diff --git a/BENCHMARK.md b/BENCHMARK.md index 14fe1f5f31b1f450124b6b13f2b88d8179f71707..61eeed2cec832d6352a0f4264a9e9c3446ca97b7 100644 --- a/BENCHMARK.md +++ b/BENCHMARK.md @@ -3,8 +3,8 @@ RLVRAMBench measures **whether a colocated language-model reinforcement learning configuration completes within a specified device-memory margin**, and whether that resource label transfers to another execution setting. -It is a measured systems dataset and a set of open transfer tasks, not a -language-model quality test, a trained predictor, or a live GPU service. +It is a measured systems dataset and a set of open evaluation tasks, not a +language-model quality test or a general memory-prediction service. The released configurations run Group Relative Policy Optimization (GRPO) with low-rank adaptation (LoRA). Generation, policy training, and weight @@ -123,6 +123,23 @@ labels must be declared; this is **not a hidden leaderboard**. A method developed on these outcomes requires new whole-case measurements for an independent generalization claim. +## Separate model-family estimation comparison + +`benchmark/estimation/` uses a different information contract from the +task-local suite above. Its ninety historical configurations supply three +leave-family-out tests: fit on sixty configurations and predict the thirty +from the remaining family. The references are donor copying, empirical +component regression with a startup guard, and regularized logistic +classification. All seeds and GPU-count variants of the test family remain +outside fitting. Historical labels were visible during method design. + +The [estimator guide](benchmark/estimation/README.md) gives the frozen +specification, prediction files, per-family scores and a CPU verification +command. The references have different failure and rejection patterns; none +uniformly improves on the others. Do not mix these fitting pools with +the original task-local donor rules or pool their repeated targets into +an inflated test-set size. + ## What this release can distinguish Within a model, the four measured workloads have identical three-state diff --git a/DATA_DICTIONARY.md b/DATA_DICTIONARY.md index ef5da5958e4d96d6f29ae62e813b0c94ff45c8ac..07d9f9f7699ad6a72093f44cdc979e8697b981bb 100644 --- a/DATA_DICTIONARY.md +++ b/DATA_DICTIONARY.md @@ -1,7 +1,8 @@ # RLVRAMBench data dictionary Static data protocol 1.0, evaluator 1.1. The separate admission protocol is -specified in `benchmark/decision/protocol.json`. +specified in `benchmark/decision/protocol.json`; the memory-estimation +comparison has its own `benchmark/estimation/protocol.json`. CSV files use UTF-8, a header row, and a blank cell for missing or unavailable information. Blank is **not zero**, success, or a memory failure. Measurements in MiB use 1 MiB = 1,048,576 bytes. Identifiers are @@ -268,3 +269,98 @@ donor attempts, actually requested target attempts, and full physical collection costs have separate ledgers. Physical GPU-seconds multiply each recorded runner invocation's elapsed seconds by its two allocated GPUs; this excludes queue time and other allocation overhead. + +## Memory-estimation comparison + +This comparison does not change the historical table counts above. +The retrospective evaluation reuses ninety historical configurations, +holding out an entire model family at a time. The separate prospective +inventory fixes twelve larger-model configurations, three seeds per +configuration, and eighteen matched two-/four-GPU pairs. A planned slot +is not necessarily a launched process or an eligible memory measurement. + +| File in `benchmark/estimation/` | Meaning | +|---|---| +| `protocol.json`, `amendment-01.json` | Frozen design and prelaunch correction | +| `model_metadata.json` | Architecture and parameter counts derived from upstream configuration and tensor headers | +| `matrix.csv`, `targets.csv` | Execution slots and the twelve configuration-level prediction targets | +| `retrospective.json` | Separate family-held-out fits, predictions, scores and source-evidence costs | +| `fitted_model.json`, `predictions.json` | Final historical-only fit and predictions sealed before prospective execution | +| `prediction_freeze.json` | Timestamp and hashes binding the frozen inputs and outputs | +| `execution_commits.json`, `execution_history.json` | Immutable execution source for each pair and disclosed runtime repairs | +| `results/configurations_flat.csv` | Browsable settings, seed coverage, observed state and all three frozen predictions | +| `results/processes.csv` | All 36 planned slots, including validation failures and unknown launch status | +| `results/configurations.csv` | Configuration-level outcomes with structured settings and seed lists | +| `results/pairs.csv` | Two-/four-GPU outcomes in each matched allocation | +| `results/evaluation.json` | Resolved-subset scores, full-panel coverage and unscored approvals | +| `results/costs.json` | Task, invocation-reservation, observed-allocation and whole-allocation time | +| `results/summary.json`, `results/provenance.json` | Counts and independently checked evidence/source hashes | +| `results/output_manifest.json` | Hashes and sizes of the twelve reconstructed JSON/CSV outputs | + +The flat table is a browsing view, not a permitted input to blind prediction: +it deliberately displays target outcomes beside predictions. `gpu_count` +is the number of devices assigned to the training process; each enclosing +allocation reserves four devices. `eligible_seed_count`, `eligible_seeds` +and `unresolved_seeds` expose the coverage behind `observed_state`. +All three seeds must be eligible before a repeated label can be assigned. +Otherwise `observed_state` is `unresolved`, even when an eligible seed +already establishes a failure. `known_memory_failure` and +`known_margin_exceedance` retain such partial eligible evidence. + +`completed_peak_mib` is available only when all three seeds complete with +validated measurements. `maximum_known_completed_peak_mib` may summarize +fewer completions; it must not be substituted for the completed repeated +outcome. Host-memory exhaustion, software failure and missing evidence do +not establish a GPU-memory label. + +Each method has a `_predicted_state` column. The method names are +`donor_copy`, `component_regression` and `logistic`. Component regression +supplies an estimated `_predicted_peak_mib`; donor copying retains +the donor's measured peak, not a newly estimated target peak. The classifier's +blank peak field is not a zero estimate. An +`_unscored_approval` marks a predicted within-margin setting whose +repeated outcome remains unresolved. Those approvals must accompany any +reported precision or failure count on the resolved subset. + +The process table separates `launched`, `eligible`, `completed`, +`online_status` and reconstructed `state`. The runner's online status is +not accepted as a label without checking its raw evidence. Unknown launch +status is distinct from `False`. `validation_errors` explains unresolved +measurements. Pair peak and runtime differences are reported only when +both members have validated completions; they describe total execution +under each device assignment, not identical generated responses. + +The four cost views are alternatives, not additive charges: +task time weights each invocation by its two or four assigned GPUs; +invocation-reservation time weights both by four; observed allocation +time includes recorded overhead through runner finalization; whole-job +time requires actual scheduler start/end accounting. Each reports known +and unknown record counts. A missing terminal record is unknown cost, +not zero cost. Preflight jobs are reported separately in execution history. + +The enhanced `allocation_accounting.json` preserves outer-job and batch-step +records separately, with the original UTC scheduler lines, their hashes, +one-second timestamp precision, and links to the recorded pair claims. +A cancelled job's batch cleanup can end later than its outer record. +Both measured spans remain explicit; they are never summed. A nonlaunched +slot has no invocation cost, but its enclosing allocation can still have a +verified scheduler cost. + +## Expanded-host follow-up + +`benchmark/host_capacity/` uses the same thirteen-file results schema as +the estimation panel, but contains eighteen new planned slots, nine pairs +and six configurations. It is a separate experiment, not a repair that +replaces original outcomes. The fixed seeds are 161–163. +The raw records are under `output/estimation_host_capacity/`. + +`study_freeze.json` seals this new design and its inherited predictions. +Provenance keeps its `study_freeze_sha256` separate from the original +`prediction_seal_sha256`. The new `protocol_sha256` identifies the expanded +host-resource condition; the original protocol and prediction seal remain +additional anchors. No predictions are fitted from this panel. + +All completion, partial-evidence and scoring definitions above still apply. +In particular, exit code zero without evidence of the requested training +step does not establish completion. Its sampled peak is not a completed-run +peak. The per-panel cost files exclude each other's allocations. diff --git a/README.md b/README.md index 1b9c2a16757dad396957b025f3e2ae8b5a265d94..d8632a9925518abfa8809476ed8a4e4785f34b09 100644 --- a/README.md +++ b/README.md @@ -1,5 +1,5 @@ --- -pretty_name: RLVRAMBench — memory outcomes and configuration decisions +pretty_name: RLVRAMBench — measured memory feasibility and configuration decisions license: mit language: - en @@ -45,6 +45,14 @@ configs: data_files: - split: recorded path: benchmark/decision/results/attempts.csv +- config_name: larger_model_configurations + data_files: + - split: measured + path: benchmark/estimation/results/configurations_flat.csv +- config_name: expanded_host_configurations + data_files: + - split: measured + path: benchmark/host_capacity/results/configurations_flat.csv - config_name: configuration_feasibility data_files: - split: configurations @@ -56,11 +64,16 @@ configs: **Which language-model training configurations can I use with the memory I have, and how much testing does that decision require?** -RLVRAMBench measures memory feasibility when response generation and -reinforcement-learning updates share the same graphics processors. It -provides measured outcomes, fixed prediction tasks, a budgeted decision -replay, reference rules, and raw evidence that explains failures. +RLVRAMBench is a measurement dataset with open evaluation tasks for a +specific language-model training system. It measures memory feasibility +when response generation and reinforcement-learning updates share the +same graphics processors. It provides measured outcomes, fixed prediction +tasks, a budgeted decision replay, reference methods, and raw failure evidence. It does **not** score a model's intelligence or learning quality. +The companion article is **Memory Feasibility in Colocated Language-Model +Reinforcement Learning: A Failure-Aware Measurement Study**. This is a +focused, extensible study, not a comprehensive cross-platform benchmark +or evidence of community adoption. The default table above is a readable twelve-candidate experiment. Each row shows the model, workload, hardware, training settings, earlier source @@ -99,13 +112,15 @@ The limit leaves 5% below the measured capacity of 40,960 MiB per device. It is a stated operating convention, **not a statistical guarantee**. One mebibyte (MiB) is 1,048,576 bytes. Failed runs do not have a counterfactual completed-run peak; blank completed-peak fields are not zero. -Unresolved attempts and known partial failure evidence remain visible. +All three slots must have eligible outcomes before the repeated label +is assigned, even if a partial record already contains a diagnosed +failure. Unresolved attempts and known partial failure evidence remain visible. These are conditional memory outcomes. They do not say that every scheduler job, software invocation, or future repetition will complete. The full attempt history is therefore separate from eligible labels. -## Two ways to use the benchmark +## Ways to use the measurements ### 1. Predict a target outcome from existing measurements @@ -195,6 +210,55 @@ Thus the comparison does not establish cold-start or GPU-time superiority of transfer. All three rules have 100% observed precision wherever they approve anything; the small panel does not estimate future failure risk. +### 3. Compare estimators with complete model families held out + +The separate `benchmark/estimation/` comparison asks whether architecture +and configuration information improves on copying an earlier outcome. +It uses ninety historical configurations. Each of three tests withholds +all thirty configurations of one model family and fits on the other sixty. +Seeds and alternative GPU counts of the test family are not fitting examples. + +References include donor copying, empirical component regression with a +one-sided startup check, and regularized logistic classification. +The regression uses completed peaks without inventing numerical targets +for failures; the classifier also learns from failure labels. +These are not exact tensor-liveness models or calibrated failure probabilities. + +Results differ by family: regression gets every Qwen label right but +approves four memory-failing Granite configurations; the classifier rejects +six usable Qwen settings. Read the separate approval errors, not just +accuracy. The historical labels were public during design, so this is +not blind method development. + +The [estimator guide](benchmark/estimation/README.md) explains inputs, +frozen predictions, acquisition costs and CPU reconstruction. + +### 4. Inspect prospective larger-model transfer + +The `larger_model_configurations` view contains twelve planned Qwen2.5-7B +settings with predictions frozen before target execution. Each row shows +how many of its three seed outcomes are eligible, any known partial failure +evidence, the resolved label when available, and each method's prediction. +This is size transfer within a familiar model family, not a new-family test. + +The original panel has 23 eligible startup memory-admission failures and +13 unresolved slots, with no completed training steps. Only six repeated +labels resolve, all to failure. Donor copying wrongly approves those six; +the fitted references reject them, but an always-failure rule does too. +Their scores on this subset do not demonstrate recovery of usable settings. +In particular, the regression's three approvals have unresolved outcomes. + +The `expanded_host_configurations` view is a separate follow-up to +host-memory exhaustion in the original panel. It retains the frozen +predictions, uses new seed slots, and increases the host-resource +allocation. Its six settings are not replacements for original targets. +All nine two-GPU invocations fail the startup memory check. The nine +four-GPU invocations exit successfully but lack the required final-step +record, leaving their three repeated labels unresolved. No validated +completion is available in this panel either. See the +[follow-up guide](benchmark/host_capacity/README.md) for the diagnostic +evidence and separate allocation costs. + ## What score should I report? Report **within-margin recall at a stated measurement budget**, together @@ -239,6 +303,11 @@ the original 588-process corpus and 24 completed batch-by-logging controls. Sixteen historical source screens are counted separately in the archive. The 48-slot admission panel is separate; it is not silently added to the unchanged static-table totals. +The estimator comparison reuses ninety of those historical configurations; +its three test folds are not additional GPU experiments. +The original larger-model panel and expanded-host follow-up have their own +configuration and process tables. Their planned slots, unresolved outcomes, +and costs are not included in the static or admission totals above. The default candidate table is a useful entry point. `configurations.csv` contains settings without outcomes. `configuration_results.csv` and diff --git a/benchmark/estimation/README.md b/benchmark/estimation/README.md new file mode 100644 index 0000000000000000000000000000000000000000..336aaee10aebcb727ad18fb6e8e633e748d488f5 --- /dev/null +++ b/benchmark/estimation/README.md @@ -0,0 +1,125 @@ +# Memory-estimation comparison + +This is a configuration-level evaluation within the RLVRAMBench measurement +study, separate from the original task-local transfer suite. It asks whether +model structure and configured work improve decisions beyond copying an +earlier measurement. It does not measure language-model quality or claim a +general memory estimator for arbitrary hardware and frameworks. + +## What is predicted? + +Each method predicts one label for a complete configuration, combining three +eligible seed outcomes: + +- `within_margin`: all three runs complete and every sampled per-device peak + is at most 38,912 MiB. +- `above_margin`: all three complete, but at least one peak exceeds that limit. +- `memory_failure`: at least one of the three eligible outcomes is a diagnosed + memory failure. + +An incomplete set of eligible seeds is unresolved, not a fourth prediction +class. Known failure evidence remains visible even when the repeated label +cannot be completed. + +## Methods and separation + +The historical fitting pool has 90 configurations: 72 boundary settings and +18 four-GPU settings. The three retrospective tests each withhold every +configuration from one model family, giving 60 fitting and 30 test +configurations. Seed processes are not separate fitting examples. +Historical labels were public during method design; this is not blind +development on an unseen dataset. + +Three fixed references use the same permitted pool: + +1. **Donor copy** selects a historical setting by workload, compound level, + GPU count, model family and parameter count, without using target outcomes. +2. **Component regression with startup guard** fits a nonnegative weighted + sum of architecture/configuration volume proxies to completed peaks. + A separate one-sided resident-actor budget check can override its class. + `peak_only_predicted_state` and `startup_guard_changed_state` distinguish + the fitted estimate from this check. This is not an implementation of + DNNMem, LLMem, or ReaL. +3. **Logistic classification** fits regularized class scores using all source + labels. Its scores are not calibrated failure probabilities. + +No failure peak is imputed as zero, capacity, or a survivor's completed peak. +Regression coefficients do not identify causal physical components. + +The separate prospective panel specifies twelve Qwen2.5-7B configurations +and 36 fresh invocations in eighteen matched two-/four-GPU pairs. All +predictions were sealed before study execution. The 7B model belongs to a +known family, and older exploratory 7B workflows existed; those records are +excluded. This is not an unseen-family or cross-hardware test. + +## How to read the result + +Report useful within-margin approvals together with approvals of diagnosed +memory failures and above-margin completions. Exact three-state accuracy is +additional information, not a sufficient admission ranking. Evaluate each +held-out family separately; do not rank methods by a single pooled score +that hides incompatible error patterns. + +`retrospective.json` contains every fold, fitted parameter, prediction, +confusion matrix and cost. The references do not uniformly improve on one +another: the regression recognizes Qwen labels well but approves four +memory-failing Granite configurations; the classifier rejects six usable +Qwen settings. + +## Files and reconstruction + +| File | Purpose | +|---|---| +| `protocol.json`, `amendment-01.json` | Original protocol and explicit prelaunch corrections | +| `model_metadata.json` | Header-derived counts, architecture, and source revisions; no model weights | +| `matrix.csv`, `targets.csv` | Frozen execution schedule and prediction settings | +| `retrospective.json` | Three leave-family-out comparisons | +| `fitted_model.json` | Final fit on all 90 historical configurations | +| `predictions.json` | Three frozen predictions per prospective configuration | +| `prediction_freeze.json` | Pre-execution timestamp and input/output hashes | +| `execution_commits.json`, `execution_history.json` | Exact execution sources and disclosed runtime repairs | +| `results/configurations_flat.csv` | One readable row per planned target, including coverage and all predictions | +| `results/evaluation.json` | Resolved-subset scores, full-panel coverage and approvals without resolved outcomes | +| `results/processes.csv`, `results/pairs.csv` | Every planned seed slot and matched pair, including unresolved evidence | +| `results/costs.json` | Separate task and allocation-time accounting, with missing records explicit | + +The result files are produced only after raw-evidence reconstruction. +The flat table deliberately includes outcomes for inspection; use the frozen +target settings and permitted historical measurements when making predictions. +An unresolved target is not a correct prediction or a safe approval. +Always report the number of unscored approvals beside scores on the resolved +subset. Host-memory failures and software failures do not establish +GPU-memory outcomes. + +The original larger-model panel is complete: 34 of 36 planned invocations +launched, 23 provide diagnosed startup memory-admission failures, and none +provides a completed training step. Thirteen slots remain unresolved, +including seven host-memory failures. Six of twelve repeated configuration +labels resolve, all to `memory_failure`. Donor copy incorrectly approves +all six. Both fitted references reject them, but so does an always-failure +rule: this failure-only subset cannot establish useful-configuration +discovery. The regression's three approvals are all unscored four-GPU +targets, not successful predictions. + +The separately frozen [expanded-host follow-up](../host_capacity/README.md) +uses new seed slots and the same predictions. It does not replace these +outcomes or merge with their denominators. + +From the reproduction code checkout, using the project-local publication +environment: + +```bash +.venv-reproduce/bin/python -m memory_tuner.verify_estimation \ + --root . --refit --output estimator-verification.json +``` + +The output file must not already exist. This checks frozen inputs, recomputes +predictions and retrospective scores, and compares fresh fits without reading +prospective outcomes or overwriting the seal. Numerical tolerance is reported; +class labels must agree exactly. The archived backend-discovery warning limits +claims about CPU timing/thread control, not the recorded predictions. + +The full historical acquisition comprises 308 attempts for 270 eligible +processes. Prospective donor copying selects ten distinct configurations, +using 34 historical attempts for thirty eligible processes. These are +deduplicated evidence costs, not demonstrated GPU-time savings. diff --git a/benchmark/estimation/allocation_accounting.json b/benchmark/estimation/allocation_accounting.json new file mode 100644 index 0000000000000000000000000000000000000000..a264139858fc2f2b7dd69fecd89efc75ae87e06f --- /dev/null +++ b/benchmark/estimation/allocation_accounting.json @@ -0,0 +1,866 @@ +[ + { + "accounting_schema_version": 1, + "allocated_gpu_count": 4, + "batch_ended_at_epoch": 1789482735.0, + "batch_started_at_epoch": 1789482644.0, + "batch_state": "FAILED", + "ended_at_epoch": 1789482735.0, + "job_id": "5054339", + "pair_id": "est-qwen25-7b-math-c2-s153", + "pair_ledger_sha256": "8a7b481ea18db0c3bdfae4ab480be6e357c735feaa755a90c0e76031a8de6b7d", + "pair_manifest_sha256": "991df8f86d667cb619f1114341213752936e091f06df8f609442286416781efc", + "raw_record_provenance": { + "batch": { + "line": "5054326_0.batch|5054339.batch|batch|FAILED|2:0|2026-09-15T14:30:44|2026-09-15T14:32:15|91|cpu=64,gres/gpu=4,mem=192G,node=1|acn13|", + "line_number": 2, + "sha256": "2aa42595f73dcef3652717d77c4c029beb0cbb7677417b79f8ce3aed0f87080d" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054326_0|5054339|rlvram-matched-gpu|FAILED|2:0|2026-09-15T14:30:44|2026-09-15T14:32:15|91|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn13|QOSGrpNodeLimit", + "line_number": 1, + "sha256": "3646af02f14eed317ed8f1a9d639a25bf3217ded7f592558bbdd952597c7db40" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_exit_code": "2:0", + "scheduler_job_id": "5054326_0", + "scheduler_node_list": "acn13", + "scheduler_scope": "outer_job_record", + "scheduler_state": "FAILED", + "started_at_epoch": 1789482644.0, + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1 + }, + { + "accounting_schema_version": 1, + "allocated_gpu_count": 4, + "batch_ended_at_epoch": 1789482792.0, + "batch_started_at_epoch": 1789482736.0, + "batch_state": "FAILED", + "ended_at_epoch": 1789482789.0, + "job_id": "5054345", + "pair_id": "est-qwen25-7b-math-c2-s151", + "pair_ledger_sha256": "3dff67c7e3f4c24f35883caef96992ab323630eab11c2bdfcce2d34ada12d403", + "pair_manifest_sha256": "3352f7e0ba5d4ae3c78325713c59a571f6c1b0a46d936e6975a437fc46565feb", + "raw_record_provenance": { + "batch": { + "line": "5054326_1.batch|5054345.batch|batch|FAILED|2:0|2026-09-15T14:32:16|2026-09-15T14:33:12|56|cpu=64,gres/gpu=4,mem=192G,node=1|acn13|", + "line_number": 5, + "sha256": "2e1f5255daa069455f9095ec95a1481549690bb185f5afa1b928747ff6b7346e" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054326_1|5054345|rlvram-matched-gpu|CANCELLED by 7547|0:0|2026-09-15T14:32:16|2026-09-15T14:33:09|53|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn13|QOSGrpNodeLimit", + "line_number": 4, + "sha256": "4c85eb09645ec15c0d8b4442d86382c262a400d1e079ff78f2eee2a643a2e3b1" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_exit_code": "0:0", + "scheduler_job_id": "5054326_1", + "scheduler_node_list": "acn13", + "scheduler_scope": "outer_job_record", + "scheduler_state": "CANCELLED by 7547", + "started_at_epoch": 1789482736.0, + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1 + }, + { + "accounting_schema_version": 1, + "allocated_gpu_count": 4, + "batch_ended_at_epoch": 1789489158.0, + "batch_started_at_epoch": 1789489149.0, + "batch_state": "FAILED", + "ended_at_epoch": 1789489158.0, + "job_id": "5054951", + "pair_id": "est-qwen25-7b-math-c3-s152", + "pair_ledger_sha256": "85f9da6513e1fe16b8dad1e6fec0d54aabe0af20601fa5883a21c81b4a73a06a", + "pair_manifest_sha256": "f4f5d85031ac044d1db31cb535d93ca010ced3f61d7e8abfae2fe2391dffe85e", + "raw_record_provenance": { + "batch": { + "line": "5054951_17.batch|5054951.batch|batch|FAILED|2:0|2026-09-15T16:19:09|2026-09-15T16:19:18|9|cpu=64,gres/gpu=4,mem=192G,node=1|acn01|", + "line_number": 53, + "sha256": "1d5a6d468589d00cc53bcb2843198f3816dc168daad7326a2b96e106b292aa07" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_17|5054951|rlvram-matched-gpu|FAILED|2:0|2026-09-15T16:19:09|2026-09-15T16:19:18|9|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn01|JobArrayTaskLimit", + "line_number": 52, + "sha256": "2e45ccf5289b1149a87417daced55a3bd75c02dab4c40eea82900a32885ff3cb" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_exit_code": "2:0", + "scheduler_job_id": "5054951_17", + "scheduler_node_list": "acn01", + "scheduler_scope": "outer_job_record", + "scheduler_state": "FAILED", + "started_at_epoch": 1789489149.0, + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1 + }, + { + "accounting_schema_version": 1, + "allocated_gpu_count": 4, + "batch_ended_at_epoch": 1789487646.0, + "batch_started_at_epoch": 1789487331.0, + "batch_state": "COMPLETED", + "ended_at_epoch": 1789487646.0, + "job_id": "5054952", + "pair_id": "est-qwen25-7b-gsm8k-c3-s153", + "pair_ledger_sha256": "0ab7e45cad162bfab6e8e6ae2c77bcbdf29c32fa102f3a39d2712159556b48cb", + "pair_manifest_sha256": "e2db8b1be2bf8cc717f5d9cecc6493baab989bc84fd92e2465486dfe15bc0b49", + "raw_record_provenance": { + "batch": { + "line": "5054951_2.batch|5054952.batch|batch|COMPLETED|0:0|2026-09-15T15:48:51|2026-09-15T15:54:06|315|cpu=64,gres/gpu=4,mem=192G,node=1|acn11|", + "line_number": 8, + "sha256": "086912b35626b5a7873378453a54b867633b963560087ed9813a949e9eb7b6fb" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_2|5054952|rlvram-matched-gpu|COMPLETED|0:0|2026-09-15T15:48:51|2026-09-15T15:54:06|315|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn11|None", + "line_number": 7, + "sha256": "f8dcda614ec3ad7343285e8f43a9a2859010369c911014c305d88c3584fc3007" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_exit_code": "0:0", + "scheduler_job_id": "5054951_2", + "scheduler_node_list": "acn11", + "scheduler_scope": "outer_job_record", + "scheduler_state": "COMPLETED", + "started_at_epoch": 1789487331.0, + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1 + }, + { + "accounting_schema_version": 1, + "allocated_gpu_count": 4, + "batch_ended_at_epoch": 1789487640.0, + "batch_started_at_epoch": 1789487332.0, + "batch_state": "COMPLETED", + "ended_at_epoch": 1789487640.0, + "job_id": "5054953", + "pair_id": "est-qwen25-7b-math-c3-s151", + "pair_ledger_sha256": "933314fb69c761afd7901fcca99b7b7ca86a2ff64e93d4cafd39a4c3c02712ac", + "pair_manifest_sha256": "56a9e2bf77364a2c7548a806774042887622e49fb384cc9ae0d540ff95105c2c", + "raw_record_provenance": { + "batch": { + "line": "5054951_3.batch|5054953.batch|batch|COMPLETED|0:0|2026-09-15T15:48:52|2026-09-15T15:54:00|308|cpu=64,gres/gpu=4,mem=192G,node=1|acn15|", + "line_number": 11, + "sha256": "dc462d83f18c0a750050b3af5addcfabbdbe6e28fafe57b7b59747a2b0833a99" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_3|5054953|rlvram-matched-gpu|COMPLETED|0:0|2026-09-15T15:48:52|2026-09-15T15:54:00|308|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn15|None", + "line_number": 10, + "sha256": "e46baf5eb652553f9ea544d7d76077d609cd8a49e596da79bbea7731e5e15cbb" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_exit_code": "0:0", + "scheduler_job_id": "5054951_3", + "scheduler_node_list": "acn15", + "scheduler_scope": "outer_job_record", + "scheduler_state": "COMPLETED", + "started_at_epoch": 1789487332.0, + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1 + }, + { + "accounting_schema_version": 1, + "allocated_gpu_count": 4, + "batch_ended_at_epoch": 1789487853.0, + "batch_started_at_epoch": 1789487332.0, + "batch_state": "OUT_OF_MEMORY", + "ended_at_epoch": 1789487853.0, + "job_id": "5054954", + "pair_id": "est-qwen25-7b-gsm8k-c2-s152", + "pair_ledger_sha256": "58395abc9d5df303ef2155cf53231c86275c1ee75370b7ac083497b225f997dc", + "pair_manifest_sha256": "06f240c40980feac67a620ecb986328a9c43f805ee3dbe35e414ea149376a59e", + "raw_record_provenance": { + "batch": { + "line": "5054951_4.batch|5054954.batch|batch|OUT_OF_MEMORY|0:125|2026-09-15T15:48:52|2026-09-15T15:57:33|521|cpu=64,gres/gpu=4,mem=192G,node=1|acn18|", + "line_number": 14, + "sha256": "632f5238a72780f5d869b5d1485bdf273f8974a60492b27e2378a04c864837ff" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_4|5054954|rlvram-matched-gpu|OUT_OF_MEMORY|0:125|2026-09-15T15:48:52|2026-09-15T15:57:33|521|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn18|None", + "line_number": 13, + "sha256": "f62e9bca018d972afec1b476f61e1166290a8623f81f87ec793ceb01a20f9b39" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_exit_code": "0:125", + "scheduler_job_id": "5054951_4", + "scheduler_node_list": "acn18", + "scheduler_scope": "outer_job_record", + "scheduler_state": "OUT_OF_MEMORY", + "started_at_epoch": 1789487332.0, + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1 + }, + { + "accounting_schema_version": 1, + "allocated_gpu_count": 4, + "batch_ended_at_epoch": 1789487922.0, + "batch_started_at_epoch": 1789487642.0, + "batch_state": "COMPLETED", + "ended_at_epoch": 1789487922.0, + "job_id": "5054977", + "pair_id": "est-qwen25-7b-code_heavy_tail-c3-s153", + "pair_ledger_sha256": "8cb922fc5199e4161a37bbb447ce56f022ff98268ca967d7adefc6d6a4fd5771", + "pair_manifest_sha256": "ec8e4a2106cd025faadd7c787be1326212b53334a8aad98ef84dd624dd9eea7b", + "raw_record_provenance": { + "batch": { + "line": "5054951_5.batch|5054977.batch|batch|COMPLETED|0:0|2026-09-15T15:54:02|2026-09-15T15:58:42|280|cpu=64,gres/gpu=4,mem=192G,node=1|acn20|", + "line_number": 17, + "sha256": "36a617a83b903953751df45ac7dd7f474a24fb59af5de346532038024a93b289" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_5|5054977|rlvram-matched-gpu|COMPLETED|0:0|2026-09-15T15:54:02|2026-09-15T15:58:42|280|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn20|JobArrayTaskLimit", + "line_number": 16, + "sha256": "ab8f8a9de2e57d569fbac28dc6f7bbabf40ac3eca3dcfe6f917c278362564e21" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_exit_code": "0:0", + "scheduler_job_id": "5054951_5", + "scheduler_node_list": "acn20", + "scheduler_scope": "outer_job_record", + "scheduler_state": "COMPLETED", + "started_at_epoch": 1789487642.0, + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1 + }, + { + "accounting_schema_version": 1, + "allocated_gpu_count": 4, + "batch_ended_at_epoch": 1789488155.0, + "batch_started_at_epoch": 1789487647.0, + "batch_state": "OUT_OF_MEMORY", + "ended_at_epoch": 1789488155.0, + "job_id": "5054981", + "pair_id": "est-qwen25-7b-gsm8k-c2-s151", + "pair_ledger_sha256": "ec8fab57145124ce0f69f1235a1021166e930084acdfb22ad6d67c48fd712354", + "pair_manifest_sha256": "223d0ca5805a06be1ba04808797cec4b35d1a0f4d2fde64bba8e75662da8719c", + "raw_record_provenance": { + "batch": { + "line": "5054951_6.batch|5054981.batch|batch|OUT_OF_MEMORY|0:125|2026-09-15T15:54:07|2026-09-15T16:02:35|508|cpu=64,gres/gpu=4,mem=192G,node=1|acn11|", + "line_number": 20, + "sha256": "3f616d0a90d5fbf5867a609035af17867a41f4df4bca8364a4a5df9dfb3367ad" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_6|5054981|rlvram-matched-gpu|OUT_OF_MEMORY|0:125|2026-09-15T15:54:07|2026-09-15T16:02:35|508|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn11|JobArrayTaskLimit", + "line_number": 19, + "sha256": "7caf085f6a50ef9b911c2dc71b9dcca53e470c381330fe99be6ab80dc1ecd509" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_exit_code": "0:125", + "scheduler_job_id": "5054951_6", + "scheduler_node_list": "acn11", + "scheduler_scope": "outer_job_record", + "scheduler_state": "OUT_OF_MEMORY", + "started_at_epoch": 1789487647.0, + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1 + }, + { + "accounting_schema_version": 1, + "allocated_gpu_count": 4, + "batch_ended_at_epoch": 1789488307.0, + "batch_started_at_epoch": 1789487853.0, + "batch_state": "OUT_OF_MEMORY", + "ended_at_epoch": 1789488307.0, + "job_id": "5054998", + "pair_id": "est-qwen25-7b-math-c2-s152", + "pair_ledger_sha256": "1064fb987e1b69fe0bf650314ce057fe2a4c8059614c565045d7155346cfbe06", + "pair_manifest_sha256": "2aeea4bb6c09da6918856caa89918ada979d4d0f64691bc1d5597c23d0ed52fd", + "raw_record_provenance": { + "batch": { + "line": "5054951_7.batch|5054998.batch|batch|OUT_OF_MEMORY|0:125|2026-09-15T15:57:33|2026-09-15T16:05:07|454|cpu=64,gres/gpu=4,mem=192G,node=1|acn18|", + "line_number": 23, + "sha256": "f9e075d6b36294e450b8d16b93fb389110ad9b8490feb881da8f51562b6cd300" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_7|5054998|rlvram-matched-gpu|OUT_OF_MEMORY|0:125|2026-09-15T15:57:33|2026-09-15T16:05:07|454|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn18|JobArrayTaskLimit", + "line_number": 22, + "sha256": "be4d8a35394b80fb173b5da704f331dc23999edcf9aee058292062a96b3a7457" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_exit_code": "0:125", + "scheduler_job_id": "5054951_7", + "scheduler_node_list": "acn18", + "scheduler_scope": "outer_job_record", + "scheduler_state": "OUT_OF_MEMORY", + "started_at_epoch": 1789487853.0, + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1 + }, + { + "accounting_schema_version": 1, + "allocated_gpu_count": 4, + "batch_ended_at_epoch": 1789488196.0, + "batch_started_at_epoch": 1789487924.0, + "batch_state": "COMPLETED", + "ended_at_epoch": 1789488196.0, + "job_id": "5055003", + "pair_id": "est-qwen25-7b-code_heavy_tail-c3-s151", + "pair_ledger_sha256": "158c05ad2355beccb78137a62d51614f4d958c7f9bb4c9bfc87ab8ecbe770e7d", + "pair_manifest_sha256": "a4fcb78a9489edfbcfa33200c5d7e9f27b030f982c3e9a28f33ca963b42d44e8", + "raw_record_provenance": { + "batch": { + "line": "5054951_8.batch|5055003.batch|batch|COMPLETED|0:0|2026-09-15T15:58:44|2026-09-15T16:03:16|272|cpu=64,gres/gpu=4,mem=192G,node=1|acn20|", + "line_number": 26, + "sha256": "62b2c4330d6284d577a0d6454951ba19d9e0ed218ce55c3cda10d42a97ee2b9b" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_8|5055003|rlvram-matched-gpu|COMPLETED|0:0|2026-09-15T15:58:44|2026-09-15T16:03:16|272|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn20|JobArrayTaskLimit", + "line_number": 25, + "sha256": "0db977269694185565b2411cd444122e886e96a4005e12e7454920c63fda0ed3" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_exit_code": "0:0", + "scheduler_job_id": "5054951_8", + "scheduler_node_list": "acn20", + "scheduler_scope": "outer_job_record", + "scheduler_state": "COMPLETED", + "started_at_epoch": 1789487924.0, + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1 + }, + { + "accounting_schema_version": 1, + "allocated_gpu_count": 4, + "batch_ended_at_epoch": 1789488585.0, + "batch_started_at_epoch": 1789488156.0, + "batch_state": "OUT_OF_MEMORY", + "ended_at_epoch": 1789488585.0, + "job_id": "5055021", + "pair_id": "est-qwen25-7b-code_heavy_tail-c2-s151", + "pair_ledger_sha256": "677ebb50042b1ef16e7bd82768cee7a2d1b0266b54e3c9794f113c7de86ace1d", + "pair_manifest_sha256": "6395afe03e9cd59953f22397461c744de2d05916ffb204a554841f8f90503496", + "raw_record_provenance": { + "batch": { + "line": "5054951_9.batch|5055021.batch|batch|OUT_OF_MEMORY|0:125|2026-09-15T16:02:36|2026-09-15T16:09:45|429|cpu=64,gres/gpu=4,mem=192G,node=1|acn11|", + "line_number": 29, + "sha256": "ed2545b6c4749a652bf9aed3cda77417e3acbf5ea51d044e1075b5ca87a4077e" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_9|5055021|rlvram-matched-gpu|OUT_OF_MEMORY|0:125|2026-09-15T16:02:36|2026-09-15T16:09:45|429|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn11|JobArrayTaskLimit", + "line_number": 28, + "sha256": "62683b8b356c3c2c7eb5adf12e3a482158a401bd57994543641da7a24e109ee3" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_exit_code": "0:125", + "scheduler_job_id": "5054951_9", + "scheduler_node_list": "acn11", + "scheduler_scope": "outer_job_record", + "scheduler_state": "OUT_OF_MEMORY", + "started_at_epoch": 1789488156.0, + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1 + }, + { + "accounting_schema_version": 1, + "allocated_gpu_count": 4, + "batch_ended_at_epoch": 1789488623.0, + "batch_started_at_epoch": 1789488197.0, + "batch_state": "OUT_OF_MEMORY", + "ended_at_epoch": 1789488623.0, + "job_id": "5055027", + "pair_id": "est-qwen25-7b-gsm8k-c2-s153", + "pair_ledger_sha256": "e3fecbf6c338a0d390915e18757a9b0ac12973f79d191f60925cab0e6a67c367", + "pair_manifest_sha256": "ea390f752b0726d01796ec9bd9cdbc7b95ba2e214c76a72933f5e840914630d6", + "raw_record_provenance": { + "batch": { + "line": "5054951_10.batch|5055027.batch|batch|OUT_OF_MEMORY|0:125|2026-09-15T16:03:17|2026-09-15T16:10:23|426|cpu=64,gres/gpu=4,mem=192G,node=1|acn20|", + "line_number": 32, + "sha256": "26ff8416d05c1f67fdbd639bfa1bf33b70049276ca4f6c64930f52a729a251b1" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_10|5055027|rlvram-matched-gpu|OUT_OF_MEMORY|0:125|2026-09-15T16:03:17|2026-09-15T16:10:23|426|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn20|JobArrayTaskLimit", + "line_number": 31, + "sha256": "1824818a792ed14ef6b19ac9e6277e656848429f2e330223e24ad101e0bab8e0" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_exit_code": "0:125", + "scheduler_job_id": "5054951_10", + "scheduler_node_list": "acn20", + "scheduler_scope": "outer_job_record", + "scheduler_state": "OUT_OF_MEMORY", + "started_at_epoch": 1789488197.0, + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1 + }, + { + "accounting_schema_version": 1, + "allocated_gpu_count": 4, + "batch_ended_at_epoch": 1789488613.0, + "batch_started_at_epoch": 1789488309.0, + "batch_state": "COMPLETED", + "ended_at_epoch": 1789488613.0, + "job_id": "5055037", + "pair_id": "est-qwen25-7b-code_heavy_tail-c3-s152", + "pair_ledger_sha256": "cb53fc6413afbaa786f81fe6a3e57cfe1b07e57d577b6beb5f0904935ebade22", + "pair_manifest_sha256": "7c27305530e8fe45b8afd153ce30cfe047562310d8ac0ae5ee4583924a7c85c6", + "raw_record_provenance": { + "batch": { + "line": "5054951_11.batch|5055037.batch|batch|COMPLETED|0:0|2026-09-15T16:05:09|2026-09-15T16:10:13|304|cpu=64,gres/gpu=4,mem=192G,node=1|acn18|", + "line_number": 35, + "sha256": "87f9c86860d88d023bf6b21fab9782dfb365502d6df6460bece1d3c034e9f8e1" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_11|5055037|rlvram-matched-gpu|COMPLETED|0:0|2026-09-15T16:05:09|2026-09-15T16:10:13|304|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn18|JobArrayTaskLimit", + "line_number": 34, + "sha256": "40a709e26f393883dee8624152f24f9d5633504cbb21392f8df02e4c80013347" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_exit_code": "0:0", + "scheduler_job_id": "5054951_11", + "scheduler_node_list": "acn18", + "scheduler_scope": "outer_job_record", + "scheduler_state": "COMPLETED", + "started_at_epoch": 1789488309.0, + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1 + }, + { + "accounting_schema_version": 1, + "allocated_gpu_count": 4, + "batch_ended_at_epoch": 1789489147.0, + "batch_started_at_epoch": 1789488586.0, + "batch_state": "OUT_OF_MEMORY", + "ended_at_epoch": 1789489147.0, + "job_id": "5055038", + "pair_id": "est-qwen25-7b-code_heavy_tail-c2-s152", + "pair_ledger_sha256": "b9f49fb1c9f629d7d00cc59e9558b0bcc1ab86462a8ecc48b357cc36086b0564", + "pair_manifest_sha256": "e537f9a68f04d79dfc3db97a988dd61656824cab1d1b99587a263c448b225dd6", + "raw_record_provenance": { + "batch": { + "line": "5054951_12.batch|5055038.batch|batch|OUT_OF_MEMORY|0:125|2026-09-15T16:09:46|2026-09-15T16:19:07|561|cpu=64,gres/gpu=4,mem=192G,node=1|acn01|", + "line_number": 38, + "sha256": "9e932163e94f3699d0ba7c3354310c866704ac2fb2f0f8da04d43862a9575a27" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_12|5055038|rlvram-matched-gpu|OUT_OF_MEMORY|0:125|2026-09-15T16:09:46|2026-09-15T16:19:07|561|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn01|JobArrayTaskLimit", + "line_number": 37, + "sha256": "b5366a5c233c938ec7d8edc0d9d53e281bf30003247fe3a175c4f1f1dbd41cba" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_exit_code": "0:125", + "scheduler_job_id": "5054951_12", + "scheduler_node_list": "acn01", + "scheduler_scope": "outer_job_record", + "scheduler_state": "OUT_OF_MEMORY", + "started_at_epoch": 1789488586.0, + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1 + }, + { + "accounting_schema_version": 1, + "allocated_gpu_count": 4, + "batch_ended_at_epoch": 1789488898.0, + "batch_started_at_epoch": 1789488613.0, + "batch_state": "COMPLETED", + "ended_at_epoch": 1789488898.0, + "job_id": "5055039", + "pair_id": "est-qwen25-7b-gsm8k-c3-s151", + "pair_ledger_sha256": "1974529d05d8aa3adf94448d49f6178553bb12da672dd9ffbe8f9d1f283f06b8", + "pair_manifest_sha256": "019720c3288d38821c459ca30347a1829315b6cbfc8a267aa2bd35f81fa32b92", + "raw_record_provenance": { + "batch": { + "line": "5054951_13.batch|5055039.batch|batch|COMPLETED|0:0|2026-09-15T16:10:13|2026-09-15T16:14:58|285|cpu=64,gres/gpu=4,mem=192G,node=1|acn08|", + "line_number": 41, + "sha256": "d974f4dfebc7b879f43cebd521c472ace1eff36724bfac10a2e127475d9b8e57" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_13|5055039|rlvram-matched-gpu|COMPLETED|0:0|2026-09-15T16:10:13|2026-09-15T16:14:58|285|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn08|JobArrayTaskLimit", + "line_number": 40, + "sha256": "507f420cee2376c7893155ce734b0cad24363f9baae3fbc2f12d7d7a1992edf4" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_exit_code": "0:0", + "scheduler_job_id": "5054951_13", + "scheduler_node_list": "acn08", + "scheduler_scope": "outer_job_record", + "scheduler_state": "COMPLETED", + "started_at_epoch": 1789488613.0, + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1 + }, + { + "accounting_schema_version": 1, + "allocated_gpu_count": 4, + "batch_ended_at_epoch": 1789488914.0, + "batch_started_at_epoch": 1789488625.0, + "batch_state": "COMPLETED", + "ended_at_epoch": 1789488914.0, + "job_id": "5055040", + "pair_id": "est-qwen25-7b-gsm8k-c3-s152", + "pair_ledger_sha256": "c92968019ad2d4c7b072941de26e1bc6621c4f980c66e27a304e703d1a3e04a5", + "pair_manifest_sha256": "a66599ba72b406d526e0c736ef1daaee8d5d6f844a26eb38f4d3f5df6b57a4dd", + "raw_record_provenance": { + "batch": { + "line": "5054951_14.batch|5055040.batch|batch|COMPLETED|0:0|2026-09-15T16:10:25|2026-09-15T16:15:14|289|cpu=64,gres/gpu=4,mem=192G,node=1|acn11|", + "line_number": 44, + "sha256": "31150809a0e2fef17253e165e2f185b4d1f73b8501d63735667e5e663f8c0a85" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_14|5055040|rlvram-matched-gpu|COMPLETED|0:0|2026-09-15T16:10:25|2026-09-15T16:15:14|289|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn11|JobArrayTaskLimit", + "line_number": 43, + "sha256": "e4d36cac674c275775e0d51dbb3a4edf4acf3dea0caa600d179b618db8cd458f" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_exit_code": "0:0", + "scheduler_job_id": "5054951_14", + "scheduler_node_list": "acn11", + "scheduler_scope": "outer_job_record", + "scheduler_state": "COMPLETED", + "started_at_epoch": 1789488625.0, + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1 + }, + { + "accounting_schema_version": 1, + "allocated_gpu_count": 4, + "batch_ended_at_epoch": 1789489176.0, + "batch_started_at_epoch": 1789488900.0, + "batch_state": "COMPLETED", + "ended_at_epoch": 1789489176.0, + "job_id": "5055042", + "pair_id": "est-qwen25-7b-math-c3-s153", + "pair_ledger_sha256": "f8bdfa5c80e6aaddc5e45ba364a1fec1f08730ec8ca6149077226468af82fb42", + "pair_manifest_sha256": "4cb4bf4a3c5922f305b91dd0985935fde9f33d62c50b707baeb014d517f9c595", + "raw_record_provenance": { + "batch": { + "line": "5054951_15.batch|5055042.batch|batch|COMPLETED|0:0|2026-09-15T16:15:00|2026-09-15T16:19:36|276|cpu=64,gres/gpu=4,mem=192G,node=1|acn08|", + "line_number": 47, + "sha256": "a6b128dbf87b3d4a01bd0d276f5d45b6f671ce64dd1941643fe632ade5b0ba5c" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_15|5055042|rlvram-matched-gpu|COMPLETED|0:0|2026-09-15T16:15:00|2026-09-15T16:19:36|276|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn08|JobArrayTaskLimit", + "line_number": 46, + "sha256": "36930ee3b0567563e4ac913fea86d244146f1e9a54ae4172fa1f5fd6e0aec5f4" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_exit_code": "0:0", + "scheduler_job_id": "5054951_15", + "scheduler_node_list": "acn08", + "scheduler_scope": "outer_job_record", + "scheduler_state": "COMPLETED", + "started_at_epoch": 1789488900.0, + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1 + }, + { + "accounting_schema_version": 1, + "allocated_gpu_count": 4, + "batch_ended_at_epoch": 1789489357.0, + "batch_started_at_epoch": 1789488914.0, + "batch_state": "OUT_OF_MEMORY", + "ended_at_epoch": 1789489357.0, + "job_id": "5055043", + "pair_id": "est-qwen25-7b-code_heavy_tail-c2-s153", + "pair_ledger_sha256": "265adb55fb58ffe85ab0f4e5e2802e00235c2654fd8429eba9595177d83aaa09", + "pair_manifest_sha256": "14764ab83644f8cd5912cebea75dae9b704ca48ba0b6bc4cb9007fd7751c599b", + "raw_record_provenance": { + "batch": { + "line": "5054951_16.batch|5055043.batch|batch|OUT_OF_MEMORY|0:125|2026-09-15T16:15:14|2026-09-15T16:22:37|443|cpu=64,gres/gpu=4,mem=192G,node=1|acn11|", + "line_number": 50, + "sha256": "3c6922fbdabfdcf51c2da2c5d0b8d478d77d4fb5ed26d1980bb9c62b966925e0" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_16|5055043|rlvram-matched-gpu|OUT_OF_MEMORY|0:125|2026-09-15T16:15:14|2026-09-15T16:22:37|443|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn11|JobArrayTaskLimit", + "line_number": 49, + "sha256": "9312728a73b1e4286376000c297d8618c5a7151fd09cf9862abcdf830cb5fcf6" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_exit_code": "0:125", + "scheduler_job_id": "5054951_16", + "scheduler_node_list": "acn11", + "scheduler_scope": "outer_job_record", + "scheduler_state": "OUT_OF_MEMORY", + "started_at_epoch": 1789488914.0, + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1 + } +] diff --git a/benchmark/estimation/allocation_accounting_sacct.psv b/benchmark/estimation/allocation_accounting_sacct.psv new file mode 100644 index 0000000000000000000000000000000000000000..2034a79471a99b35c0f42e390ccf572682326770 --- /dev/null +++ b/benchmark/estimation/allocation_accounting_sacct.psv @@ -0,0 +1,54 @@ +5054326_0|5054339|rlvram-matched-gpu|FAILED|2:0|2026-09-15T14:30:44|2026-09-15T14:32:15|91|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn13|QOSGrpNodeLimit +5054326_0.batch|5054339.batch|batch|FAILED|2:0|2026-09-15T14:30:44|2026-09-15T14:32:15|91|cpu=64,gres/gpu=4,mem=192G,node=1|acn13| +5054326_0.extern|5054339.extern|extern|COMPLETED|0:0|2026-09-15T14:30:44|2026-09-15T14:32:15|91|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn13| +5054326_1|5054345|rlvram-matched-gpu|CANCELLED by 7547|0:0|2026-09-15T14:32:16|2026-09-15T14:33:09|53|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn13|QOSGrpNodeLimit +5054326_1.batch|5054345.batch|batch|FAILED|2:0|2026-09-15T14:32:16|2026-09-15T14:33:12|56|cpu=64,gres/gpu=4,mem=192G,node=1|acn13| +5054326_1.extern|5054345.extern|extern|COMPLETED|0:0|2026-09-15T14:32:16|2026-09-15T14:33:12|56|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn13| +5054951_2|5054952|rlvram-matched-gpu|COMPLETED|0:0|2026-09-15T15:48:51|2026-09-15T15:54:06|315|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn11|None +5054951_2.batch|5054952.batch|batch|COMPLETED|0:0|2026-09-15T15:48:51|2026-09-15T15:54:06|315|cpu=64,gres/gpu=4,mem=192G,node=1|acn11| +5054951_2.extern|5054952.extern|extern|COMPLETED|0:0|2026-09-15T15:48:51|2026-09-15T15:54:06|315|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn11| +5054951_3|5054953|rlvram-matched-gpu|COMPLETED|0:0|2026-09-15T15:48:52|2026-09-15T15:54:00|308|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn15|None +5054951_3.batch|5054953.batch|batch|COMPLETED|0:0|2026-09-15T15:48:52|2026-09-15T15:54:00|308|cpu=64,gres/gpu=4,mem=192G,node=1|acn15| +5054951_3.extern|5054953.extern|extern|COMPLETED|0:0|2026-09-15T15:48:52|2026-09-15T15:54:00|308|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn15| +5054951_4|5054954|rlvram-matched-gpu|OUT_OF_MEMORY|0:125|2026-09-15T15:48:52|2026-09-15T15:57:33|521|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn18|None +5054951_4.batch|5054954.batch|batch|OUT_OF_MEMORY|0:125|2026-09-15T15:48:52|2026-09-15T15:57:33|521|cpu=64,gres/gpu=4,mem=192G,node=1|acn18| +5054951_4.extern|5054954.extern|extern|COMPLETED|0:0|2026-09-15T15:48:52|2026-09-15T15:57:33|521|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn18| +5054951_5|5054977|rlvram-matched-gpu|COMPLETED|0:0|2026-09-15T15:54:02|2026-09-15T15:58:42|280|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn20|JobArrayTaskLimit +5054951_5.batch|5054977.batch|batch|COMPLETED|0:0|2026-09-15T15:54:02|2026-09-15T15:58:42|280|cpu=64,gres/gpu=4,mem=192G,node=1|acn20| +5054951_5.extern|5054977.extern|extern|COMPLETED|0:0|2026-09-15T15:54:02|2026-09-15T15:58:42|280|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn20| +5054951_6|5054981|rlvram-matched-gpu|OUT_OF_MEMORY|0:125|2026-09-15T15:54:07|2026-09-15T16:02:35|508|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn11|JobArrayTaskLimit +5054951_6.batch|5054981.batch|batch|OUT_OF_MEMORY|0:125|2026-09-15T15:54:07|2026-09-15T16:02:35|508|cpu=64,gres/gpu=4,mem=192G,node=1|acn11| +5054951_6.extern|5054981.extern|extern|COMPLETED|0:0|2026-09-15T15:54:07|2026-09-15T16:02:35|508|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn11| +5054951_7|5054998|rlvram-matched-gpu|OUT_OF_MEMORY|0:125|2026-09-15T15:57:33|2026-09-15T16:05:07|454|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn18|JobArrayTaskLimit +5054951_7.batch|5054998.batch|batch|OUT_OF_MEMORY|0:125|2026-09-15T15:57:33|2026-09-15T16:05:07|454|cpu=64,gres/gpu=4,mem=192G,node=1|acn18| +5054951_7.extern|5054998.extern|extern|COMPLETED|0:0|2026-09-15T15:57:33|2026-09-15T16:05:07|454|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn18| +5054951_8|5055003|rlvram-matched-gpu|COMPLETED|0:0|2026-09-15T15:58:44|2026-09-15T16:03:16|272|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn20|JobArrayTaskLimit +5054951_8.batch|5055003.batch|batch|COMPLETED|0:0|2026-09-15T15:58:44|2026-09-15T16:03:16|272|cpu=64,gres/gpu=4,mem=192G,node=1|acn20| +5054951_8.extern|5055003.extern|extern|COMPLETED|0:0|2026-09-15T15:58:44|2026-09-15T16:03:16|272|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn20| +5054951_9|5055021|rlvram-matched-gpu|OUT_OF_MEMORY|0:125|2026-09-15T16:02:36|2026-09-15T16:09:45|429|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn11|JobArrayTaskLimit +5054951_9.batch|5055021.batch|batch|OUT_OF_MEMORY|0:125|2026-09-15T16:02:36|2026-09-15T16:09:45|429|cpu=64,gres/gpu=4,mem=192G,node=1|acn11| +5054951_9.extern|5055021.extern|extern|COMPLETED|0:0|2026-09-15T16:02:36|2026-09-15T16:09:45|429|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn11| +5054951_10|5055027|rlvram-matched-gpu|OUT_OF_MEMORY|0:125|2026-09-15T16:03:17|2026-09-15T16:10:23|426|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn20|JobArrayTaskLimit +5054951_10.batch|5055027.batch|batch|OUT_OF_MEMORY|0:125|2026-09-15T16:03:17|2026-09-15T16:10:23|426|cpu=64,gres/gpu=4,mem=192G,node=1|acn20| +5054951_10.extern|5055027.extern|extern|COMPLETED|0:0|2026-09-15T16:03:17|2026-09-15T16:10:23|426|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn20| +5054951_11|5055037|rlvram-matched-gpu|COMPLETED|0:0|2026-09-15T16:05:09|2026-09-15T16:10:13|304|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn18|JobArrayTaskLimit +5054951_11.batch|5055037.batch|batch|COMPLETED|0:0|2026-09-15T16:05:09|2026-09-15T16:10:13|304|cpu=64,gres/gpu=4,mem=192G,node=1|acn18| +5054951_11.extern|5055037.extern|extern|COMPLETED|0:0|2026-09-15T16:05:09|2026-09-15T16:10:13|304|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn18| +5054951_12|5055038|rlvram-matched-gpu|OUT_OF_MEMORY|0:125|2026-09-15T16:09:46|2026-09-15T16:19:07|561|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn01|JobArrayTaskLimit +5054951_12.batch|5055038.batch|batch|OUT_OF_MEMORY|0:125|2026-09-15T16:09:46|2026-09-15T16:19:07|561|cpu=64,gres/gpu=4,mem=192G,node=1|acn01| +5054951_12.extern|5055038.extern|extern|COMPLETED|0:0|2026-09-15T16:09:46|2026-09-15T16:19:07|561|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn01| +5054951_13|5055039|rlvram-matched-gpu|COMPLETED|0:0|2026-09-15T16:10:13|2026-09-15T16:14:58|285|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn08|JobArrayTaskLimit +5054951_13.batch|5055039.batch|batch|COMPLETED|0:0|2026-09-15T16:10:13|2026-09-15T16:14:58|285|cpu=64,gres/gpu=4,mem=192G,node=1|acn08| +5054951_13.extern|5055039.extern|extern|COMPLETED|0:0|2026-09-15T16:10:13|2026-09-15T16:14:58|285|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn08| +5054951_14|5055040|rlvram-matched-gpu|COMPLETED|0:0|2026-09-15T16:10:25|2026-09-15T16:15:14|289|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn11|JobArrayTaskLimit +5054951_14.batch|5055040.batch|batch|COMPLETED|0:0|2026-09-15T16:10:25|2026-09-15T16:15:14|289|cpu=64,gres/gpu=4,mem=192G,node=1|acn11| +5054951_14.extern|5055040.extern|extern|COMPLETED|0:0|2026-09-15T16:10:25|2026-09-15T16:15:14|289|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn11| +5054951_15|5055042|rlvram-matched-gpu|COMPLETED|0:0|2026-09-15T16:15:00|2026-09-15T16:19:36|276|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn08|JobArrayTaskLimit +5054951_15.batch|5055042.batch|batch|COMPLETED|0:0|2026-09-15T16:15:00|2026-09-15T16:19:36|276|cpu=64,gres/gpu=4,mem=192G,node=1|acn08| +5054951_15.extern|5055042.extern|extern|COMPLETED|0:0|2026-09-15T16:15:00|2026-09-15T16:19:36|276|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn08| +5054951_16|5055043|rlvram-matched-gpu|OUT_OF_MEMORY|0:125|2026-09-15T16:15:14|2026-09-15T16:22:37|443|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn11|JobArrayTaskLimit +5054951_16.batch|5055043.batch|batch|OUT_OF_MEMORY|0:125|2026-09-15T16:15:14|2026-09-15T16:22:37|443|cpu=64,gres/gpu=4,mem=192G,node=1|acn11| +5054951_16.extern|5055043.extern|extern|COMPLETED|0:0|2026-09-15T16:15:14|2026-09-15T16:22:37|443|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn11| +5054951_17|5054951|rlvram-matched-gpu|FAILED|2:0|2026-09-15T16:19:09|2026-09-15T16:19:18|9|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn01|JobArrayTaskLimit +5054951_17.batch|5054951.batch|batch|FAILED|2:0|2026-09-15T16:19:09|2026-09-15T16:19:18|9|cpu=64,gres/gpu=4,mem=192G,node=1|acn01| +5054951_17.extern|5054951.extern|extern|COMPLETED|0:0|2026-09-15T16:19:09|2026-09-15T16:19:18|9|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn01| diff --git a/benchmark/estimation/amendment-01.json b/benchmark/estimation/amendment-01.json new file mode 100644 index 0000000000000000000000000000000000000000..175939a64afb3c9d27a6cdc1a618c7fc8bc80264 --- /dev/null +++ b/benchmark/estimation/amendment-01.json @@ -0,0 +1,43 @@ +{ + "base_protocol_sha256": "1694e447e842600612604891e7b3d036ff0791fa84c4bdbeba54ed093feefdeb", + "amended_at_utc": "2026-09-15T14:09:16+00:00", + "before_new_target_execution": true, + "reason": "Prelaunch code review found that a nonnegative coefficient on an offload=true indicator could only increase predicted memory. Replace that indicator with the resident actor-shard volume. The shared feature change affects both regression and logistic classification. Old-source unit-test fits occurred; no final prediction seal or new standard-GRPO target execution existed at this amendment. No hyperparameter search or target outcome motivated the correction.", + "previous_feature_order": [ + "actor_shards_gib", "adapter_optimizer_gib", "generation_budget_gib", + "generation_replica_gib", "largest_unit_proxy_gib", + "checkpoint_boundary_volume_gib", "vocabulary_output_volume_gib", + "actor_parameter_offload" + ], + "updates": { + "features": [ + "actor_shards_gib", "adapter_optimizer_gib", "generation_budget_gib", + "generation_replica_gib", "largest_unit_proxy_gib", + "checkpoint_boundary_volume_gib", "vocabulary_output_volume_gib", + "resident_actor_shards_gib" + ], + "feature_scope": "Architecture/checkpoint-header and configured shape volumes only. Largest-unit proxy=max(largest decoder block, root non-block elements, largest weight tensor), not an observed wrapping map. Checkpoint-boundary and vocabulary-output volumes are predictive covariates, not exact live activation totals. No source identity, dataset name, compound level, actual generated length or outcome appears as a numeric feature. The eighth feature is zero when actor parameters are offloaded and otherwise 4(P+R)/(G*2^30) GiB. Unconditional actor shards remain a separate training-state proxy. Both estimators use the amended eight-feature order. Startup applicability reads the validated offload setting directly.", + "component_regression": "Empirical component regression WITH a one-sided startup guard. Nonnegative least squares with an intercept, float64, maxiter=10000; training-completion-only scaling and peak targets unchanged. Report the raw peak-derived class and whether the guard changed it. Coefficients are empirical, not causal or independently identified physical components.", + "scoring": "Require three eligible distinct seeds for a repeated prospective label. Unresolved configurations are not labels: report resolved-subset categorical metrics separately from all 12 planned configurations, resolved/unresolved IDs, known failure/margin flags and each method's unscored approvals. Do not impute failure peaks from capacity or surviving processes. Retrospective scores are separate by held-out family. Report the deduplicated donor acquisition subset separately from the full fitting pool." + }, + "implementation_sha256": { + "benchmark.py": "35e36caad46d2814b0ac9f8180961de5564746347cf668d5c3fc1e643fd477e8", + "memory_tuner/estimation_baselines.py": "8cc20f9697eb38cb6a19a2255e95712ac44469cbd147eaf93e7e7ed183360ad2" + }, + "additional_prelaunch_checks": [ + "Reject stored feature-name/order mismatch before prediction.", + "Reject unresolved or duplicate truth configurations in the known-label scorer.", + "Record Python, NumPy, SciPy, scikit-learn, backend and thread settings; limit final fitting to one numerical thread.", + "Recheck all effective input hashes before writing outputs and bind original protocol plus amendment hashes in the final seal.", + "The runner must validate sealed prediction outputs and effective inputs before any study invocation." + ], + "unchanged_inputs_verified_sha256": { + "benchmark/estimation/matrix.csv": "1ff175e22c76bf1af3857b4ae57cd4f78bafe0d280a295d696e96815ef85e9c8", + "benchmark/estimation/targets.csv": "f2481278127fa726f24297d81fc9df46e999315fe543167ed398e671fffa522f", + "benchmark/estimation/model_metadata.json": "2f0ce16c13e72900d2a60f9b7ad28d4dc5ed43d7915d8a88925ba9c5ffd684ad", + "benchmark/attempts.csv": "397c8caaf0d3abe06587fd649c181fe71d5b8b5f5e24e0591cc7e56d38e275c1", + "benchmark/configurations.csv": "3a77c96a67b82e01d23fb6f7a440cba23041bed1dae2749072cb9abb5b5d239d", + "benchmark/outcomes.csv": "af05cd91ed1de843735cc188e5e54dbd1645b3ed91c97f503ac3de67bc5f61c9" + }, + "unchanged_design": "Original protocol, 90 source-ID allowlist, 12 target settings, 36 scheduled invocations, all seeds/order/subsets, model snapshot and metadata, data hashes, donor rule, method hyperparameters, budgets and no-retry policy remain unchanged. No new target observation may update either GPU-count prediction." +} diff --git a/benchmark/estimation/execution_commits.json b/benchmark/estimation/execution_commits.json new file mode 100644 index 0000000000000000000000000000000000000000..bb37ad2b4719b0f6184873b24382b37d29092306 --- /dev/null +++ b/benchmark/estimation/execution_commits.json @@ -0,0 +1,20 @@ +{ + "est-qwen25-7b-math-c2-s153": "0bd659b3f69e2f779dbea84bb2da221bd0ff52db", + "est-qwen25-7b-math-c2-s151": "0bd659b3f69e2f779dbea84bb2da221bd0ff52db", + "est-qwen25-7b-gsm8k-c3-s153": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-math-c3-s151": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-gsm8k-c2-s152": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-code_heavy_tail-c3-s153": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-gsm8k-c2-s151": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-math-c2-s152": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-code_heavy_tail-c3-s151": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-code_heavy_tail-c2-s151": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-gsm8k-c2-s153": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-code_heavy_tail-c3-s152": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-code_heavy_tail-c2-s152": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-gsm8k-c3-s151": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-gsm8k-c3-s152": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-math-c3-s153": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-code_heavy_tail-c2-s153": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-math-c3-s152": "d8e0c3551d8572fb64e3b7126fbd7569a987973f" +} diff --git a/benchmark/estimation/execution_history.json b/benchmark/estimation/execution_history.json new file mode 100644 index 0000000000000000000000000000000000000000..4f95fe98af15050ffd02dd4b643a167f65219377 --- /dev/null +++ b/benchmark/estimation/execution_history.json @@ -0,0 +1,169 @@ +{ + "schema_version": 1, + "prediction_seal_sha256": "5ed714caff922a5109346a4242545d19f2824636d0f52bb0bbd2a09b40ed4666", + "original_protocol_sha256": "1694e447e842600612604891e7b3d036ff0791fa84c4bdbeba54ed093feefdeb", + "events": [ + { + "kind": "zero_training_preflight", + "job_id": "5054195", + "source_commit": "e41df1aa1ab289cb5e2b0b3beeb3a03ff66e66d3", + "status": "failed", + "elapsed_seconds": 5, + "allocated_gpus": 4, + "training_launched": false, + "cause": "Global Slurm GPU ordinals 4..7 differ from the four-device NVML job-view ordinals 0..3." + }, + { + "kind": "read_only_device_diagnostic", + "job_id": "5054207", + "status": "completed", + "elapsed_seconds": 2, + "allocated_gpus": 4, + "training_launched": false, + "finding": "Exactly four devices visible through NVML, matching CUDA_VISIBLE_DEVICES=0,1,2,3; global Slurm GPU indices retained separately. UUIDs resolve from this complete job view." + }, + { + "kind": "zero_training_preflight", + "job_id": "5054224", + "source_commit": "0bd659b3f69e2f779dbea84bb2da221bd0ff52db", + "status": "completed", + "elapsed_seconds": 53, + "allocated_gpus": 4, + "training_launched": false, + "checks": "Nonconsecutive two-device subset, then all four; actual CUDA UUIDs, Ray placement, NVML selection and post-exit all-device idle checks passed." + }, + { + "kind": "study_array", + "array_job_id": "5054326", + "source_commit": "0bd659b3f69e2f779dbea84bb2da221bd0ff52db", + "requested_pair_indices": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17], + "started_pair_indices": [0, 1], + "never_started_pair_indices": [2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17], + "cancelled_at_utc": "2026-09-15T14:33:09+00:00", + "reason": "First pair exposed deterministic vLLM integer parsing of UUID masks during import, before model initialization. Cancellation prevented repeated setup failures. Cleanup also raced an allocation-sampler subprocess.", + "started_invocation_evidence": "Three training logs diagnose the vLLM UUID-to-integer import failure. Seed151/twoGPU was interrupted during device preflight and has no matching traceback. All four invocations are infrastructure-invalid, not memory outcomes.", + "repair_scope": "Device-identification compatibility and sampler shutdown only. Frozen predictions, models, data, settings, seeds, pair order and subsets are unchanged.", + "replacement_policy": "No started pair or invocation will be rerun. Only indices never claimed may be launched once from corrected immutable source. Missing repeated labels remain unresolved." + }, + { + "kind": "expanded_zero_training_preflight", + "job_id": "5054489", + "source_commit": "26418498a5b5405a31c8ce7e4f2b9bd1cd7f9575", + "status": "failed", + "started_at_utc": "2026-09-15T15:26:01+00:00", + "ended_at_utc": "2026-09-15T15:26:40+00:00", + "elapsed_seconds": 39, + "allocated_gpus": 4, + "training_launched": false, + "cause": "A training-library import initialized CUDA. The preflight incorrectly equated zero training with zero CUDA-context initialization. The failing import was not identified by this version.", + "checks_completed": "Initial allocation idle check and preflight-start record only. No two/four-GPU runtime report or pass record was produced.", + "additional_checks": "Actual VERL and vLLM worker imports, UUID-to-NVML round-trip validation, capability queries in driver and Ray actors, preservation of physical UUID masks, and post-exit idle checks." + }, + { + "kind": "remaining_study_array", + "array_job_id": "5054693", + "source_commit": "26418498a5b5405a31c8ce7e4f2b9bd1cd7f9575", + "status": "cancelled_before_any_start", + "reason": "The expanded preflight failed; DependencyNeverSatisfied prevented every array element from starting. Cancelled only this pending array before queuing the never-claimed slots behind the corrected preflight.", + "started_pair_indices": [], + "dependency": "afterok:5054489", + "requested_pair_indices": [2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17], + "maximum_concurrency": 3, + "retry_or_replacement": false, + "source_worktree_clean_before_submission": true, + "frozen_protocol_matrix_and_prediction_seal_unchanged": true, + "runtime_scope": "Device identity compatibility and sampling shutdown only; no model, data, batching, precision, memory-control, seed, order, subset, or predictor changes." + }, + { + "kind": "corrected_import_context_zero_training_preflight", + "job_id": "5054938", + "source_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "status": "completed", + "started_at_utc": "2026-09-15T15:46:24+00:00", + "ended_at_utc": "2026-09-15T15:47:47+00:00", + "elapsed_seconds": 83, + "allocated_gpus": 4, + "training_launched": false, + "correction": "Explicitly disposable preflight children may initialize CUDA through imports. Per-import transitions are recorded; actual task UUIDs, NVML capabilities and Ray placement remain checked. All-four-device idle checks now follow both successful and failed child exits.", + "checks_completed": "Both task sizes passed CUDA/Ray/NVML identity and capability checks, with two all-four-device idle samples after each disposable child. Recorded import-created CUDA contexts arose while importing verl.workers.rollout.vllm_rollout.vllm_async_server.", + "two_gpu_runtime_report_sha256": "77f617e346869b6a617eb316468a7ac74b9156ae7e168df3ea089aba5153c81d", + "two_gpu_post_child_idle_sha256": "6c4adc25407078a374b10cebf6d810db8ee4cf724215679fa570128e5f16cd32", + "four_gpu_runtime_report_sha256": "64e83062d19c77987a2215851b2335a8d25ca6ac79020f72073d6cc1109ac97f", + "four_gpu_post_child_idle_sha256": "df025b3e93ead549afe56be266c4b0407a96d4cea243e1eab50dc29d8af91cec", + "actual_training_bridge_and_trainer_probe_exec_unchanged": true, + "frozen_protocol_matrix_and_prediction_seal_unchanged": true + }, + { + "kind": "remaining_study_array", + "array_job_id": "5054951", + "source_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "status": "completed_all_claims_terminal", + "first_started_at_utc": "2026-09-15T15:48:51+00:00", + "last_ended_at_utc": "2026-09-15T16:22:37+00:00", + "dependency": "afterok:5054938", + "requested_pair_indices": [2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17], + "maximum_concurrency": 3, + "retry_or_replacement": false, + "source_worktree_clean_before_submission": true, + "frozen_protocol_matrix_and_prediction_seal_unchanged": true + }, + { + "kind": "host_memory_censoring_diagnostic", + "observed_at_utc": "2026-09-15T16:08:43+00:00", + "array_job_id": "5054951", + "source_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "host_memory_limit_mib": 196608, + "confirmed_jobs": [ + { + "job_id": "5054954", + "pair_index": 4, + "pair_id": "est-qwen25-7b-gsm8k-c2-s152", + "node": "acn18", + "scheduler_oom_kill_events": 4, + "scheduler_max_rss_mib": 196607.84, + "ended_at_utc": "2026-09-15T15:57:33+00:00" + }, + { + "job_id": "5054981", + "pair_index": 6, + "pair_id": "est-qwen25-7b-gsm8k-c2-s151", + "node": "acn11", + "scheduler_oom_kill_events": 3, + "scheduler_max_rss_mib": 196607.59, + "ended_at_utc": "2026-09-15T16:02:35+00:00" + } + ], + "finding": "Both four-task-GPU c2 periods reached generation-engine graph capture and then lost workers at the host-memory cgroup limit. Neither completed a training step or reported CUDA out-of-memory. Partial device peaks do not establish a completed memory label.", + "eligibility": "Keep four-GPU outcomes unresolved. Earlier two-GPU admission failures remain separately subject to independent evidence validation. Retain costs and all raw evidence.", + "additional_scheduler_evidence": "Array index 7, job 5054998, also ended OUT_OF_MEMORY at 16:05:07 UTC with four oom_kill events; its detailed period diagnosis is not included in the two-job finding above.", + "original_study_resources_and_prediction_seal_unchanged": true, + "new_followup_launched": false + } + ], + "status": "complete_independently_collected", + "final_outcomes": { + "planned_slots": 36, + "launched_invocations": 34, + "eligible_memory_admission_failures": 23, + "completed_invocations": 0, + "unresolved_slots": 13, + "resolved_configurations": 6, + "resolved_configuration_states": "all memory_failure", + "unresolved_configurations": 6, + "unresolved_slot_causes": { + "initial_setup_or_interrupted_invocation": 4, + "host_memory_exhaustion": 7, + "scheduler_identity_nonlaunch": 2 + }, + "same_setting_retries_or_replacements": 0, + "accounted_outer_allocations": 18, + "outer_allocation_gpu_seconds": 23296, + "batch_step_gpu_seconds_separate_not_additive": 23308, + "preflight_and_diagnostic_gpu_seconds_separate": 728, + "results_directory": "benchmark/estimation/results", + "accounting_input": "benchmark/estimation/allocation_accounting.json", + "accounting_raw_source": "benchmark/estimation/allocation_accounting_sacct.psv" + }, + "study_resource_cap_unchanged_reserved_gpu_hours": 66, + "preflight_cost_reported_separately": true +} diff --git a/benchmark/estimation/fitted_model.json b/benchmark/estimation/fitted_model.json new file mode 100644 index 0000000000000000000000000000000000000000..50b3f14515acdbb1a0c94d0d5fad0ca2cfe66195 --- /dev/null +++ b/benchmark/estimation/fitted_model.json @@ -0,0 +1,415 @@ +{ + "classifier_classes": [ + "above_margin", + "memory_failure", + "within_margin" + ], + "classifier_coefficients": [ + [ + 0.14837732625278155, + -0.7242861977644228, + -0.030985986687828454, + -0.1694410444015409, + 0.05753193331705675, + 0.16607992470759583, + -0.3141698408276469, + 0.1874593676217934 + ], + [ + 0.5024974977973338, + 1.1551702881304888, + 2.92972148756149, + 0.32062376850070523, + 0.21846640819013424, + 0.46730560418414135, + 0.19357465202341737, + 0.28182249965838185 + ], + [ + -0.6508748240501154, + -0.4308840903660661, + -2.898735500873661, + -0.15118272409916486, + -0.27599834150719094, + -0.6333855288917364, + 0.12059518880422974, + -0.4692818672801747 + ] + ], + "classifier_intercepts": [ + 0.7271110245357251, + -2.9187339040273756, + 2.191622879491652 + ], + "classifier_iterations": [ + 38 + ], + "component_coefficients_gib": [ + 6.721051299571553, + 2.6852579170135575, + 0.0, + 25.67138002927888, + 0.0, + 2.3897827606119444, + 0.1984882140415612, + 0.0, + 4.920522564008151 + ], + "component_divisors": [ + 7.317022323608398, + 1.0, + 32.0, + 7.317022323608398, + 2.289562225341797, + 9.28125, + 18.31640625, + 7.317022323608398 + ], + "component_training_residual_norm_gib": 13.98249734490586, + "eligible_source_processes": 270, + "feature_names": [ + "actor_shards_gib", + "adapter_optimizer_gib", + "generation_budget_gib", + "generation_replica_gib", + "largest_unit_proxy_gib", + "checkpoint_boundary_volume_gib", + "vocabulary_output_volume_gib", + "resident_actor_shards_gib" + ], + "numerical_environment": { + "fit_thread_limit": 1, + "numpy": "1.24.3", + "platform": "Linux-4.18.0-553.146.1.el8_10.x86_64-x86_64-with-glibc2.28", + "python": "3.11.4", + "scikit_learn": "1.3.0", + "scipy": "1.10.1", + "thread_environment": { + "MKL_NUM_THREADS": null, + "OMP_NUM_THREADS": null, + "OPENBLAS_NUM_THREADS": null + }, + "threadpools": [ + { + "filepath": "/mnt/proj3/open-35-44/verl/rl/.venv-reproduce/lib/python3.11/site-packages/scikit_learn.libs/libgomp-a34b3233.so.1.0.0", + "internal_api": "openmp", + "num_threads": 1, + "prefix": "libgomp", + "user_api": "openmp", + "version": null + }, + { + "architecture": "Zen", + "filepath": "/mnt/proj3/open-35-44/verl/rl/.venv-reproduce/lib/python3.11/site-packages/scipy.libs/libopenblasp-r0-41284840.3.18.so", + "internal_api": "openblas", + "num_threads": 1, + "prefix": "libopenblas", + "threading_layer": "pthreads", + "user_api": "blas", + "version": "0.3.18" + } + ] + }, + "recorded_source_attempts": 308, + "regression_training_configuration_ids": [ + "cfg-072bd7669d20f8fa", + "cfg-0d9ba9215ab5683e", + "cfg-0f540d3f6fe4661b", + "cfg-0f8a31ef7ac4c551", + "cfg-1432bb8a4d9c3b8c", + "cfg-17e69bcf1b041bd9", + "cfg-18285c3cb50308d4", + "cfg-1c5278f8be90a50a", + "cfg-2afbe250af9cd3e1", + "cfg-301e4dae7048c6ec", + "cfg-30f98456fdb322c1", + "cfg-338d9224eb9d2f3a", + "cfg-371ae37408089efe", + "cfg-399f6b038d6f65f9", + "cfg-3f0f48a3c570a0c8", + "cfg-5244d13eac3116ad", + "cfg-5ff9621270e710ed", + "cfg-61c14de21a34d0e6", + "cfg-62f371a80968a0ba", + "cfg-6a4562db98b38d6f", + "cfg-6d54ac65d934f404", + "cfg-714c540f76d9525d", + "cfg-73c06764ba28a56b", + "cfg-78c440d3891ddbc2", + "cfg-7a0e81934d3398ee", + "cfg-7a9a974958c57abc", + "cfg-835067f7405e89a2", + "cfg-89ca48ce88ea694e", + "cfg-8bd7050cd7f49e78", + "cfg-9160dc45a0171a10", + "cfg-93d4f41e600c9079", + "cfg-96ccf395b84393f7", + "cfg-97107ceba8bf0cbe", + "cfg-9758870c0a0b4d69", + "cfg-9c461415e6bb9702", + "cfg-9f624ec22aa3dde0", + "cfg-a026693bae37d30b", + "cfg-a23a9ffc1f670139", + "cfg-a24e9cdd2f6143f2", + "cfg-a3987314cb2f9f12", + "cfg-a94bf7877633dfda", + "cfg-aa33c72e0a449b9c", + "cfg-acfa4acda15c3599", + "cfg-b5007322fb2e1772", + "cfg-b5352215babfeba6", + "cfg-b56fb353c2f2fa3e", + "cfg-b8dd70ab682c13b2", + "cfg-b8e90bd69c9744ea", + "cfg-bae2977da7957f72", + "cfg-bfb83ff32b7fd921", + "cfg-c0605c1607caa079", + "cfg-c2a8e6e8e140e109", + "cfg-c3f0ad157d299d39", + "cfg-cc2c35944156e5d4", + "cfg-cc5f4f0c2e1256ba", + "cfg-cd3974595f9db882", + "cfg-d02503f8c9804a2d", + "cfg-d4a213515805a9a7", + "cfg-d9397d15f234d541", + "cfg-db87c67536995ab0", + "cfg-e117f15dd5ccfb30", + "cfg-e2a205be8efba055", + "cfg-e91942abb5b4f847", + "cfg-f009e9e3891c06f4", + "cfg-f03e4e98c682240e", + "cfg-f439d2dd90137a80", + "cfg-f75b68f479767290", + "cfg-fbe1c0422f98cb38", + "cfg-fd4d4789145230df", + "cfg-fd5bb0db68b0dae3" + ], + "scaler_mean": [ + 5.465130043029785, + 0.3629150390625, + 25.333333333333332, + 6.072366714477539, + 1.2746035257975261, + 4.3475341796875, + 6.95436397128635, + 3.8458322525024413 + ], + "scaler_scale": [ + 1.5123338692603971, + 0.0900265920171816, + 7.180219742846005, + 0.9775168008234761, + 0.785841111066617, + 3.8672072396593395, + 6.986165289686755, + 2.6975162604429936 + ], + "source_cost": { + "distinct_selected_donors": { + "configuration_ids": [ + "cfg-1c5278f8be90a50a", + "cfg-30f98456fdb322c1", + "cfg-338d9224eb9d2f3a", + "cfg-399f6b038d6f65f9", + "cfg-714c540f76d9525d", + "cfg-7a9a974958c57abc", + "cfg-cc2c35944156e5d4", + "cfg-e91942abb5b4f847", + "cfg-f009e9e3891c06f4", + "cfg-f03e4e98c682240e" + ], + "configurations": 10, + "eligible_processes": 30, + "recorded_attempts": 34 + }, + "full_permitted_fitting_pool": { + "configuration_ids": [ + "cfg-072bd7669d20f8fa", + "cfg-0d9ba9215ab5683e", + "cfg-0f540d3f6fe4661b", + "cfg-0f8a31ef7ac4c551", + "cfg-10644b5d0a82d5c3", + "cfg-1432bb8a4d9c3b8c", + "cfg-17e69bcf1b041bd9", + "cfg-18285c3cb50308d4", + "cfg-1c5278f8be90a50a", + "cfg-2afbe250af9cd3e1", + "cfg-301e4dae7048c6ec", + "cfg-30f98456fdb322c1", + "cfg-32977d5aa42a41cb", + "cfg-338d9224eb9d2f3a", + "cfg-352f4dbc55fbcd3c", + "cfg-371ae37408089efe", + "cfg-399f6b038d6f65f9", + "cfg-3c86a11fdb8455cb", + "cfg-3dc042a103546eac", + "cfg-3f0f48a3c570a0c8", + "cfg-49d9bccfe873f5c6", + "cfg-5244d13eac3116ad", + "cfg-55811d23c5144bb7", + "cfg-5839ccf95c2c1722", + "cfg-5ff9621270e710ed", + "cfg-61c14de21a34d0e6", + "cfg-62f371a80968a0ba", + "cfg-6a4562db98b38d6f", + "cfg-6d54ac65d934f404", + "cfg-714c540f76d9525d", + "cfg-73c06764ba28a56b", + "cfg-78c440d3891ddbc2", + "cfg-7a0e81934d3398ee", + "cfg-7a9a974958c57abc", + "cfg-7df2428553b76877", + "cfg-7f23cab692546e3f", + "cfg-835067f7405e89a2", + "cfg-896685010703c0fb", + "cfg-89ca48ce88ea694e", + "cfg-8bd7050cd7f49e78", + "cfg-9160dc45a0171a10", + "cfg-93d4f41e600c9079", + "cfg-966f5804dc918967", + "cfg-96ccf395b84393f7", + "cfg-97107ceba8bf0cbe", + "cfg-9758870c0a0b4d69", + "cfg-9c461415e6bb9702", + "cfg-9f624ec22aa3dde0", + "cfg-a026693bae37d30b", + "cfg-a23a9ffc1f670139", + "cfg-a24e9cdd2f6143f2", + "cfg-a3987314cb2f9f12", + "cfg-a94bf7877633dfda", + "cfg-aa33c72e0a449b9c", + "cfg-acfa4acda15c3599", + "cfg-b5007322fb2e1772", + "cfg-b5352215babfeba6", + "cfg-b56fb353c2f2fa3e", + "cfg-b8dd70ab682c13b2", + "cfg-b8e90bd69c9744ea", + "cfg-bae2977da7957f72", + "cfg-bfb83ff32b7fd921", + "cfg-c0605c1607caa079", + "cfg-c2a8e6e8e140e109", + "cfg-c3f0ad157d299d39", + "cfg-ca71b56dc3189305", + "cfg-cad8341dbbff10ba", + "cfg-cc2c35944156e5d4", + "cfg-cc5f4f0c2e1256ba", + "cfg-cc900fb7731c3f49", + "cfg-cd3974595f9db882", + "cfg-d02503f8c9804a2d", + "cfg-d138048feeb8fe5d", + "cfg-d4a213515805a9a7", + "cfg-d9397d15f234d541", + "cfg-db87c67536995ab0", + "cfg-e117f15dd5ccfb30", + "cfg-e2a205be8efba055", + "cfg-e8893b12558f51dd", + "cfg-e91942abb5b4f847", + "cfg-ea5a23a40ad5ca50", + "cfg-ea92cccc91029465", + "cfg-f009e9e3891c06f4", + "cfg-f03e4e98c682240e", + "cfg-f439d2dd90137a80", + "cfg-f75b68f479767290", + "cfg-fad1b555f8b37235", + "cfg-fbe1c0422f98cb38", + "cfg-fd4d4789145230df", + "cfg-fd5bb0db68b0dae3" + ], + "configurations": 90, + "eligible_processes": 270, + "recorded_attempts": 308 + }, + "interpretation": "Historical evidence acquisition, not incremental GPU execution or a cold-start cost saving." + }, + "training_configuration_ids": [ + "cfg-072bd7669d20f8fa", + "cfg-0d9ba9215ab5683e", + "cfg-0f540d3f6fe4661b", + "cfg-0f8a31ef7ac4c551", + "cfg-10644b5d0a82d5c3", + "cfg-1432bb8a4d9c3b8c", + "cfg-17e69bcf1b041bd9", + "cfg-18285c3cb50308d4", + "cfg-1c5278f8be90a50a", + "cfg-2afbe250af9cd3e1", + "cfg-301e4dae7048c6ec", + "cfg-30f98456fdb322c1", + "cfg-32977d5aa42a41cb", + "cfg-338d9224eb9d2f3a", + "cfg-352f4dbc55fbcd3c", + "cfg-371ae37408089efe", + "cfg-399f6b038d6f65f9", + "cfg-3c86a11fdb8455cb", + "cfg-3dc042a103546eac", + "cfg-3f0f48a3c570a0c8", + "cfg-49d9bccfe873f5c6", + "cfg-5244d13eac3116ad", + "cfg-55811d23c5144bb7", + "cfg-5839ccf95c2c1722", + "cfg-5ff9621270e710ed", + "cfg-61c14de21a34d0e6", + "cfg-62f371a80968a0ba", + "cfg-6a4562db98b38d6f", + "cfg-6d54ac65d934f404", + "cfg-714c540f76d9525d", + "cfg-73c06764ba28a56b", + "cfg-78c440d3891ddbc2", + "cfg-7a0e81934d3398ee", + "cfg-7a9a974958c57abc", + "cfg-7df2428553b76877", + "cfg-7f23cab692546e3f", + "cfg-835067f7405e89a2", + "cfg-896685010703c0fb", + "cfg-89ca48ce88ea694e", + "cfg-8bd7050cd7f49e78", + "cfg-9160dc45a0171a10", + "cfg-93d4f41e600c9079", + "cfg-966f5804dc918967", + "cfg-96ccf395b84393f7", + "cfg-97107ceba8bf0cbe", + "cfg-9758870c0a0b4d69", + "cfg-9c461415e6bb9702", + "cfg-9f624ec22aa3dde0", + "cfg-a026693bae37d30b", + "cfg-a23a9ffc1f670139", + "cfg-a24e9cdd2f6143f2", + "cfg-a3987314cb2f9f12", + "cfg-a94bf7877633dfda", + "cfg-aa33c72e0a449b9c", + "cfg-acfa4acda15c3599", + "cfg-b5007322fb2e1772", + "cfg-b5352215babfeba6", + "cfg-b56fb353c2f2fa3e", + "cfg-b8dd70ab682c13b2", + "cfg-b8e90bd69c9744ea", + "cfg-bae2977da7957f72", + "cfg-bfb83ff32b7fd921", + "cfg-c0605c1607caa079", + "cfg-c2a8e6e8e140e109", + "cfg-c3f0ad157d299d39", + "cfg-ca71b56dc3189305", + "cfg-cad8341dbbff10ba", + "cfg-cc2c35944156e5d4", + "cfg-cc5f4f0c2e1256ba", + "cfg-cc900fb7731c3f49", + "cfg-cd3974595f9db882", + "cfg-d02503f8c9804a2d", + "cfg-d138048feeb8fe5d", + "cfg-d4a213515805a9a7", + "cfg-d9397d15f234d541", + "cfg-db87c67536995ab0", + "cfg-e117f15dd5ccfb30", + "cfg-e2a205be8efba055", + "cfg-e8893b12558f51dd", + "cfg-e91942abb5b4f847", + "cfg-ea5a23a40ad5ca50", + "cfg-ea92cccc91029465", + "cfg-f009e9e3891c06f4", + "cfg-f03e4e98c682240e", + "cfg-f439d2dd90137a80", + "cfg-f75b68f479767290", + "cfg-fad1b555f8b37235", + "cfg-fbe1c0422f98cb38", + "cfg-fd4d4789145230df", + "cfg-fd5bb0db68b0dae3" + ] +} diff --git a/benchmark/estimation/matrix.csv b/benchmark/estimation/matrix.csv new file mode 100644 index 0000000000000000000000000000000000000000..b9a9b009b1d365eebfc621d69389cd7d9505a429 --- /dev/null +++ b/benchmark/estimation/matrix.csv @@ -0,0 +1,37 @@ +experiment_id,run_group,study_stage,configuration_level,model_family,algorithm,dataset,model,data_dir,gpu_count,rollout_tp_size,actor_micro_batch,rollout_logprob_micro_batch,ref_logprob_micro_batch,vllm_gpu_memory_utilization,parameter_offload,optimizer_offload,free_cache_engine,max_prompt_length,max_response_length,max_model_len,max_num_seqs,rollout_n,train_batch_size,train_max_samples,val_max_samples,training_seed,total_training_steps,gpu_monitor_interval_ms,workload_name,custom_reward_function_path,custom_reward_function_name,revision_study,pair_id,period,condition,configuration_id,case_id,gpu_subset_ordinals,allocator_trace_enabled,allocator_trace_sync,save_freq,test_freq,val_before_train,resume_mode,max_actor_ckpt_to_keep,per_process_timeout_seconds +est-qwen25-7b-math-c2-s153-2gpu,prospective_estimation,prospective_estimation,c2,qwen25_7b,grpo,math,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/math,2,1,8,8,8,0.6,False,False,True,512,512,1024,128,2,64,64,64,153,1,100,,,,prospective_estimation,est-qwen25-7b-math-c2-s153,1,2gpu,est-qwen25-7b-math-c2-2gpu,est-qwen25-7b-math-2gpu,0;2,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-math-c2-s153-4gpu,prospective_estimation,prospective_estimation,c2,qwen25_7b,grpo,math,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/math,4,1,8,8,8,0.6,False,False,True,512,512,1024,128,2,64,64,64,153,1,100,,,,prospective_estimation,est-qwen25-7b-math-c2-s153,2,4gpu,est-qwen25-7b-math-c2-4gpu,est-qwen25-7b-math-4gpu,0;1;2;3,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-math-c2-s151-4gpu,prospective_estimation,prospective_estimation,c2,qwen25_7b,grpo,math,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/math,4,1,8,8,8,0.6,False,False,True,512,512,1024,128,2,64,64,64,151,1,100,,,,prospective_estimation,est-qwen25-7b-math-c2-s151,1,4gpu,est-qwen25-7b-math-c2-4gpu,est-qwen25-7b-math-4gpu,0;1;2;3,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-math-c2-s151-2gpu,prospective_estimation,prospective_estimation,c2,qwen25_7b,grpo,math,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/math,2,1,8,8,8,0.6,False,False,True,512,512,1024,128,2,64,64,64,151,1,100,,,,prospective_estimation,est-qwen25-7b-math-c2-s151,2,2gpu,est-qwen25-7b-math-c2-2gpu,est-qwen25-7b-math-2gpu,1;2,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-gsm8k-c3-s153-2gpu,prospective_estimation,prospective_estimation,c3,qwen25_7b,grpo,gsm8k,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/gsm8k,2,1,16,4,4,0.7,False,False,True,1024,2048,3072,64,2,32,32,32,153,1,100,,,,prospective_estimation,est-qwen25-7b-gsm8k-c3-s153,1,2gpu,est-qwen25-7b-gsm8k-c3-2gpu,est-qwen25-7b-gsm8k-2gpu,1;2,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-gsm8k-c3-s153-4gpu,prospective_estimation,prospective_estimation,c3,qwen25_7b,grpo,gsm8k,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/gsm8k,4,1,16,4,4,0.7,False,False,True,1024,2048,3072,64,2,32,32,32,153,1,100,,,,prospective_estimation,est-qwen25-7b-gsm8k-c3-s153,2,4gpu,est-qwen25-7b-gsm8k-c3-4gpu,est-qwen25-7b-gsm8k-4gpu,0;1;2;3,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-math-c3-s151-4gpu,prospective_estimation,prospective_estimation,c3,qwen25_7b,grpo,math,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/math,4,1,16,8,8,0.7,False,False,True,512,512,1024,128,2,64,64,64,151,1,100,,,,prospective_estimation,est-qwen25-7b-math-c3-s151,1,4gpu,est-qwen25-7b-math-c3-4gpu,est-qwen25-7b-math-4gpu,0;1;2;3,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-math-c3-s151-2gpu,prospective_estimation,prospective_estimation,c3,qwen25_7b,grpo,math,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/math,2,1,16,8,8,0.7,False,False,True,512,512,1024,128,2,64,64,64,151,1,100,,,,prospective_estimation,est-qwen25-7b-math-c3-s151,2,2gpu,est-qwen25-7b-math-c3-2gpu,est-qwen25-7b-math-2gpu,0;1,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-gsm8k-c2-s152-2gpu,prospective_estimation,prospective_estimation,c2,qwen25_7b,grpo,gsm8k,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/gsm8k,2,1,8,4,4,0.6,False,False,True,1024,2048,3072,64,2,32,32,32,152,1,100,,,,prospective_estimation,est-qwen25-7b-gsm8k-c2-s152,1,2gpu,est-qwen25-7b-gsm8k-c2-2gpu,est-qwen25-7b-gsm8k-2gpu,0;3,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-gsm8k-c2-s152-4gpu,prospective_estimation,prospective_estimation,c2,qwen25_7b,grpo,gsm8k,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/gsm8k,4,1,8,4,4,0.6,False,False,True,1024,2048,3072,64,2,32,32,32,152,1,100,,,,prospective_estimation,est-qwen25-7b-gsm8k-c2-s152,2,4gpu,est-qwen25-7b-gsm8k-c2-4gpu,est-qwen25-7b-gsm8k-4gpu,0;1;2;3,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-code_heavy_tail-c3-s153-4gpu,prospective_estimation,prospective_estimation,c3,qwen25_7b,grpo,code_heavy_tail,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/codecontests/heavy_tail,4,1,16,4,4,0.7,False,False,True,1536,1024,2560,64,2,32,32,32,153,1,100,code_heavy_tail,/mnt/proj3/open-35-44/verl/rl/memory_tuner/code_shape_reward.py,compute_score,prospective_estimation,est-qwen25-7b-code_heavy_tail-c3-s153,1,4gpu,est-qwen25-7b-code_heavy_tail-c3-4gpu,est-qwen25-7b-code_heavy_tail-4gpu,0;1;2;3,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-code_heavy_tail-c3-s153-2gpu,prospective_estimation,prospective_estimation,c3,qwen25_7b,grpo,code_heavy_tail,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/codecontests/heavy_tail,2,1,16,4,4,0.7,False,False,True,1536,1024,2560,64,2,32,32,32,153,1,100,code_heavy_tail,/mnt/proj3/open-35-44/verl/rl/memory_tuner/code_shape_reward.py,compute_score,prospective_estimation,est-qwen25-7b-code_heavy_tail-c3-s153,2,2gpu,est-qwen25-7b-code_heavy_tail-c3-2gpu,est-qwen25-7b-code_heavy_tail-2gpu,2;3,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-gsm8k-c2-s151-2gpu,prospective_estimation,prospective_estimation,c2,qwen25_7b,grpo,gsm8k,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/gsm8k,2,1,8,4,4,0.6,False,False,True,1024,2048,3072,64,2,32,32,32,151,1,100,,,,prospective_estimation,est-qwen25-7b-gsm8k-c2-s151,1,2gpu,est-qwen25-7b-gsm8k-c2-2gpu,est-qwen25-7b-gsm8k-2gpu,2;3,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-gsm8k-c2-s151-4gpu,prospective_estimation,prospective_estimation,c2,qwen25_7b,grpo,gsm8k,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/gsm8k,4,1,8,4,4,0.6,False,False,True,1024,2048,3072,64,2,32,32,32,151,1,100,,,,prospective_estimation,est-qwen25-7b-gsm8k-c2-s151,2,4gpu,est-qwen25-7b-gsm8k-c2-4gpu,est-qwen25-7b-gsm8k-4gpu,0;1;2;3,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-math-c2-s152-2gpu,prospective_estimation,prospective_estimation,c2,qwen25_7b,grpo,math,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/math,2,1,8,8,8,0.6,False,False,True,512,512,1024,128,2,64,64,64,152,1,100,,,,prospective_estimation,est-qwen25-7b-math-c2-s152,1,2gpu,est-qwen25-7b-math-c2-2gpu,est-qwen25-7b-math-2gpu,0;3,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-math-c2-s152-4gpu,prospective_estimation,prospective_estimation,c2,qwen25_7b,grpo,math,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/math,4,1,8,8,8,0.6,False,False,True,512,512,1024,128,2,64,64,64,152,1,100,,,,prospective_estimation,est-qwen25-7b-math-c2-s152,2,4gpu,est-qwen25-7b-math-c2-4gpu,est-qwen25-7b-math-4gpu,0;1;2;3,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-code_heavy_tail-c3-s151-4gpu,prospective_estimation,prospective_estimation,c3,qwen25_7b,grpo,code_heavy_tail,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/codecontests/heavy_tail,4,1,16,4,4,0.7,False,False,True,1536,1024,2560,64,2,32,32,32,151,1,100,code_heavy_tail,/mnt/proj3/open-35-44/verl/rl/memory_tuner/code_shape_reward.py,compute_score,prospective_estimation,est-qwen25-7b-code_heavy_tail-c3-s151,1,4gpu,est-qwen25-7b-code_heavy_tail-c3-4gpu,est-qwen25-7b-code_heavy_tail-4gpu,0;1;2;3,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-code_heavy_tail-c3-s151-2gpu,prospective_estimation,prospective_estimation,c3,qwen25_7b,grpo,code_heavy_tail,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/codecontests/heavy_tail,2,1,16,4,4,0.7,False,False,True,1536,1024,2560,64,2,32,32,32,151,1,100,code_heavy_tail,/mnt/proj3/open-35-44/verl/rl/memory_tuner/code_shape_reward.py,compute_score,prospective_estimation,est-qwen25-7b-code_heavy_tail-c3-s151,2,2gpu,est-qwen25-7b-code_heavy_tail-c3-2gpu,est-qwen25-7b-code_heavy_tail-2gpu,0;2,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-code_heavy_tail-c2-s151-2gpu,prospective_estimation,prospective_estimation,c2,qwen25_7b,grpo,code_heavy_tail,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/codecontests/heavy_tail,2,1,8,4,4,0.6,False,False,True,1536,1024,2560,64,2,32,32,32,151,1,100,code_heavy_tail,/mnt/proj3/open-35-44/verl/rl/memory_tuner/code_shape_reward.py,compute_score,prospective_estimation,est-qwen25-7b-code_heavy_tail-c2-s151,1,2gpu,est-qwen25-7b-code_heavy_tail-c2-2gpu,est-qwen25-7b-code_heavy_tail-2gpu,0;1,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-code_heavy_tail-c2-s151-4gpu,prospective_estimation,prospective_estimation,c2,qwen25_7b,grpo,code_heavy_tail,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/codecontests/heavy_tail,4,1,8,4,4,0.6,False,False,True,1536,1024,2560,64,2,32,32,32,151,1,100,code_heavy_tail,/mnt/proj3/open-35-44/verl/rl/memory_tuner/code_shape_reward.py,compute_score,prospective_estimation,est-qwen25-7b-code_heavy_tail-c2-s151,2,4gpu,est-qwen25-7b-code_heavy_tail-c2-4gpu,est-qwen25-7b-code_heavy_tail-4gpu,0;1;2;3,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-gsm8k-c2-s153-4gpu,prospective_estimation,prospective_estimation,c2,qwen25_7b,grpo,gsm8k,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/gsm8k,4,1,8,4,4,0.6,False,False,True,1024,2048,3072,64,2,32,32,32,153,1,100,,,,prospective_estimation,est-qwen25-7b-gsm8k-c2-s153,1,4gpu,est-qwen25-7b-gsm8k-c2-4gpu,est-qwen25-7b-gsm8k-4gpu,0;1;2;3,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-gsm8k-c2-s153-2gpu,prospective_estimation,prospective_estimation,c2,qwen25_7b,grpo,gsm8k,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/gsm8k,2,1,8,4,4,0.6,False,False,True,1024,2048,3072,64,2,32,32,32,153,1,100,,,,prospective_estimation,est-qwen25-7b-gsm8k-c2-s153,2,2gpu,est-qwen25-7b-gsm8k-c2-2gpu,est-qwen25-7b-gsm8k-2gpu,0;2,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-code_heavy_tail-c3-s152-2gpu,prospective_estimation,prospective_estimation,c3,qwen25_7b,grpo,code_heavy_tail,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/codecontests/heavy_tail,2,1,16,4,4,0.7,False,False,True,1536,1024,2560,64,2,32,32,32,152,1,100,code_heavy_tail,/mnt/proj3/open-35-44/verl/rl/memory_tuner/code_shape_reward.py,compute_score,prospective_estimation,est-qwen25-7b-code_heavy_tail-c3-s152,1,2gpu,est-qwen25-7b-code_heavy_tail-c3-2gpu,est-qwen25-7b-code_heavy_tail-2gpu,0;1,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-code_heavy_tail-c3-s152-4gpu,prospective_estimation,prospective_estimation,c3,qwen25_7b,grpo,code_heavy_tail,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/codecontests/heavy_tail,4,1,16,4,4,0.7,False,False,True,1536,1024,2560,64,2,32,32,32,152,1,100,code_heavy_tail,/mnt/proj3/open-35-44/verl/rl/memory_tuner/code_shape_reward.py,compute_score,prospective_estimation,est-qwen25-7b-code_heavy_tail-c3-s152,2,4gpu,est-qwen25-7b-code_heavy_tail-c3-4gpu,est-qwen25-7b-code_heavy_tail-4gpu,0;1;2;3,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-code_heavy_tail-c2-s152-2gpu,prospective_estimation,prospective_estimation,c2,qwen25_7b,grpo,code_heavy_tail,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/codecontests/heavy_tail,2,1,8,4,4,0.6,False,False,True,1536,1024,2560,64,2,32,32,32,152,1,100,code_heavy_tail,/mnt/proj3/open-35-44/verl/rl/memory_tuner/code_shape_reward.py,compute_score,prospective_estimation,est-qwen25-7b-code_heavy_tail-c2-s152,1,2gpu,est-qwen25-7b-code_heavy_tail-c2-2gpu,est-qwen25-7b-code_heavy_tail-2gpu,1;3,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-code_heavy_tail-c2-s152-4gpu,prospective_estimation,prospective_estimation,c2,qwen25_7b,grpo,code_heavy_tail,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/codecontests/heavy_tail,4,1,8,4,4,0.6,False,False,True,1536,1024,2560,64,2,32,32,32,152,1,100,code_heavy_tail,/mnt/proj3/open-35-44/verl/rl/memory_tuner/code_shape_reward.py,compute_score,prospective_estimation,est-qwen25-7b-code_heavy_tail-c2-s152,2,4gpu,est-qwen25-7b-code_heavy_tail-c2-4gpu,est-qwen25-7b-code_heavy_tail-4gpu,0;1;2;3,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-gsm8k-c3-s151-4gpu,prospective_estimation,prospective_estimation,c3,qwen25_7b,grpo,gsm8k,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/gsm8k,4,1,16,4,4,0.7,False,False,True,1024,2048,3072,64,2,32,32,32,151,1,100,,,,prospective_estimation,est-qwen25-7b-gsm8k-c3-s151,1,4gpu,est-qwen25-7b-gsm8k-c3-4gpu,est-qwen25-7b-gsm8k-4gpu,0;1;2;3,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-gsm8k-c3-s151-2gpu,prospective_estimation,prospective_estimation,c3,qwen25_7b,grpo,gsm8k,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/gsm8k,2,1,16,4,4,0.7,False,False,True,1024,2048,3072,64,2,32,32,32,151,1,100,,,,prospective_estimation,est-qwen25-7b-gsm8k-c3-s151,2,2gpu,est-qwen25-7b-gsm8k-c3-2gpu,est-qwen25-7b-gsm8k-2gpu,1;2,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-gsm8k-c3-s152-4gpu,prospective_estimation,prospective_estimation,c3,qwen25_7b,grpo,gsm8k,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/gsm8k,4,1,16,4,4,0.7,False,False,True,1024,2048,3072,64,2,32,32,32,152,1,100,,,,prospective_estimation,est-qwen25-7b-gsm8k-c3-s152,1,4gpu,est-qwen25-7b-gsm8k-c3-4gpu,est-qwen25-7b-gsm8k-4gpu,0;1;2;3,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-gsm8k-c3-s152-2gpu,prospective_estimation,prospective_estimation,c3,qwen25_7b,grpo,gsm8k,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/gsm8k,2,1,16,4,4,0.7,False,False,True,1024,2048,3072,64,2,32,32,32,152,1,100,,,,prospective_estimation,est-qwen25-7b-gsm8k-c3-s152,2,2gpu,est-qwen25-7b-gsm8k-c3-2gpu,est-qwen25-7b-gsm8k-2gpu,1;3,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-math-c3-s153-2gpu,prospective_estimation,prospective_estimation,c3,qwen25_7b,grpo,math,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/math,2,1,16,8,8,0.7,False,False,True,512,512,1024,128,2,64,64,64,153,1,100,,,,prospective_estimation,est-qwen25-7b-math-c3-s153,1,2gpu,est-qwen25-7b-math-c3-2gpu,est-qwen25-7b-math-2gpu,0;3,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-math-c3-s153-4gpu,prospective_estimation,prospective_estimation,c3,qwen25_7b,grpo,math,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/math,4,1,16,8,8,0.7,False,False,True,512,512,1024,128,2,64,64,64,153,1,100,,,,prospective_estimation,est-qwen25-7b-math-c3-s153,2,4gpu,est-qwen25-7b-math-c3-4gpu,est-qwen25-7b-math-4gpu,0;1;2;3,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-code_heavy_tail-c2-s153-4gpu,prospective_estimation,prospective_estimation,c2,qwen25_7b,grpo,code_heavy_tail,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/codecontests/heavy_tail,4,1,8,4,4,0.6,False,False,True,1536,1024,2560,64,2,32,32,32,153,1,100,code_heavy_tail,/mnt/proj3/open-35-44/verl/rl/memory_tuner/code_shape_reward.py,compute_score,prospective_estimation,est-qwen25-7b-code_heavy_tail-c2-s153,1,4gpu,est-qwen25-7b-code_heavy_tail-c2-4gpu,est-qwen25-7b-code_heavy_tail-4gpu,0;1;2;3,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-code_heavy_tail-c2-s153-2gpu,prospective_estimation,prospective_estimation,c2,qwen25_7b,grpo,code_heavy_tail,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/codecontests/heavy_tail,2,1,8,4,4,0.6,False,False,True,1536,1024,2560,64,2,32,32,32,153,1,100,code_heavy_tail,/mnt/proj3/open-35-44/verl/rl/memory_tuner/code_shape_reward.py,compute_score,prospective_estimation,est-qwen25-7b-code_heavy_tail-c2-s153,2,2gpu,est-qwen25-7b-code_heavy_tail-c2-2gpu,est-qwen25-7b-code_heavy_tail-2gpu,2;3,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-math-c3-s152-4gpu,prospective_estimation,prospective_estimation,c3,qwen25_7b,grpo,math,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/math,4,1,16,8,8,0.7,False,False,True,512,512,1024,128,2,64,64,64,152,1,100,,,,prospective_estimation,est-qwen25-7b-math-c3-s152,1,4gpu,est-qwen25-7b-math-c3-4gpu,est-qwen25-7b-math-4gpu,0;1;2;3,True,1,-1,-1,False,disable,1,1500 +est-qwen25-7b-math-c3-s152-2gpu,prospective_estimation,prospective_estimation,c3,qwen25_7b,grpo,math,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/math,2,1,16,8,8,0.7,False,False,True,512,512,1024,128,2,64,64,64,152,1,100,,,,prospective_estimation,est-qwen25-7b-math-c3-s152,2,2gpu,est-qwen25-7b-math-c3-2gpu,est-qwen25-7b-math-2gpu,1;3,True,1,-1,-1,False,disable,1,1500 diff --git a/benchmark/estimation/model_metadata.json b/benchmark/estimation/model_metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..99aa0cbeb336fd311e91773fd146bddb55dd1e87 --- /dev/null +++ b/benchmark/estimation/model_metadata.json @@ -0,0 +1,108 @@ +{ + "Qwen/Qwen2.5-3B-Instruct": { + "all_linear_layer_adapter_elements": 119734272, + "config_sha256": "eed00b17e22553979d090fa492e587e92885e328914c8e0b0b78f0a0d3576b3b", + "floating_checkpoint_bytes": 6171877376, + "floating_checkpoint_elements": 3085938688, + "head_dim": 128, + "hidden_size": 2048, + "intermediate_size": 11008, + "largest_transformer_layer_elements": 77076992, + "largest_weight_tensor_elements": 311164928, + "lora_rank": 64, + "model_id": "Qwen/Qwen2.5-3B-Instruct", + "non_transformer_layer_elements": 311166976, + "num_attention_heads": 16, + "num_hidden_layers": 36, + "num_key_value_heads": 2, + "revision": "aa8e72537993ba99e69dfaafa59ed015b17504d1", + "scope": "Numeric architecture/checkpoint-header inputs only; no weight payload. Adapter count assumes rank-64 all-linear transformer layers, excluding embeddings/output head. Floating checkpoint elements may include buffers; no separately allocated reference model is assumed.", + "shard_header_sha256": { + "model-00001-of-00002.safetensors": "8d1c57f12799cf1a917776014a04eff3bfaf1cb1dafa2ca3a12f92ade34e65bd", + "model-00002-of-00002.safetensors": "5bb7481ec0a898ebaa815dbdd881d83752dfc05e445e19a7571800dbd025cf8e" + }, + "tensor_count": 434, + "tie_word_embeddings": true, + "vocab_size": 151936 + }, + "Qwen/Qwen2.5-7B-Instruct": { + "all_linear_layer_adapter_elements": 161480704, + "config_sha256": "7463bb0ea78315365e6c6b74de4e73bbcc8359dfb0c5a737584e077d42c0b03c", + "floating_checkpoint_bytes": 15231233024, + "floating_checkpoint_elements": 7615616512, + "head_dim": 128, + "hidden_size": 3584, + "intermediate_size": 18944, + "largest_transformer_layer_elements": 233057792, + "largest_weight_tensor_elements": 544997376, + "lora_rank": 64, + "model_id": "Qwen/Qwen2.5-7B-Instruct", + "non_transformer_layer_elements": 1089998336, + "num_attention_heads": 28, + "num_hidden_layers": 28, + "num_key_value_heads": 4, + "revision": "a09a35458c702b33eeacc393d103063234e8bc28", + "scope": "Numeric architecture/checkpoint-header inputs only; no weight payload. Adapter count assumes rank-64 all-linear transformer layers, excluding embeddings/output head. Floating checkpoint elements may include buffers; no separately allocated reference model is assumed.", + "shard_header_sha256": { + "model-00001-of-00004.safetensors": "24ce65896dba011b065dd8fcbf877559a6701fe80516f7d4d039d082147cb2e8", + "model-00002-of-00004.safetensors": "69f0c2c87ae06b271181d6964cf235aa9ce7931f64cde7a1a5511ba16070c739", + "model-00003-of-00004.safetensors": "a859e7527f6cf5f144667f65c356c01374f95ed56f40cbe6dca4cb7752164066", + "model-00004-of-00004.safetensors": "181209f962ac9f71cb2bf72adbd5727cf4fbbd646cf80558576ccb5c13e75b71" + }, + "tensor_count": 339, + "tie_word_embeddings": false, + "vocab_size": 152064 + }, + "ibm-granite/granite-3.3-2b-instruct": { + "all_linear_layer_adapter_elements": 112721920, + "config_sha256": "9202d328d8368958aab7dad89a8e6aa35c250b3a2eedd2d8850ee0bcceca65f6", + "floating_checkpoint_bytes": 5067079680, + "floating_checkpoint_elements": 2533539840, + "head_dim": 64, + "hidden_size": 2048, + "intermediate_size": 8192, + "largest_transformer_layer_elements": 60821504, + "largest_weight_tensor_elements": 100677632, + "lora_rank": 64, + "model_id": "ibm-granite/granite-3.3-2b-instruct", + "non_transformer_layer_elements": 100679680, + "num_attention_heads": 32, + "num_hidden_layers": 40, + "num_key_value_heads": 8, + "revision": "707f574c62054322f6b5b04b6d075f0a8f05e0f0", + "scope": "Numeric architecture/checkpoint-header inputs only; no weight payload. Adapter count assumes rank-64 all-linear transformer layers, excluding embeddings/output head. Floating checkpoint elements may include buffers; no separately allocated reference model is assumed.", + "shard_header_sha256": { + "model-00001-of-00002.safetensors": "744c520c1a1d794563fed804e54d99dd424713d4346c3edc06dfdd17c6afe1c0", + "model-00002-of-00002.safetensors": "039f729e7d783e8514dcac3af74ef1e9ad7c8c235e7d92671000e778ab937d79" + }, + "tensor_count": 362, + "tie_word_embeddings": true, + "vocab_size": 49159 + }, + "microsoft/Phi-4-mini-instruct": { + "all_linear_layer_adapter_elements": 92274688, + "config_sha256": "ac65d86061d3d0d704ee2511fd0eb8713ef19eb6eedba17c3080a4165d5b933b", + "floating_checkpoint_bytes": 7672043520, + "floating_checkpoint_elements": 3836021760, + "head_dim": 128, + "hidden_size": 3072, + "intermediate_size": 8192, + "largest_transformer_layer_elements": 100669440, + "largest_weight_tensor_elements": 614596608, + "lora_rank": 64, + "model_id": "microsoft/Phi-4-mini-instruct", + "non_transformer_layer_elements": 614599680, + "num_attention_heads": 24, + "num_hidden_layers": 32, + "num_key_value_heads": 8, + "revision": "cfbefacb99257ffa30c83adab238a50856ac3083", + "scope": "Numeric architecture/checkpoint-header inputs only; no weight payload. Adapter count assumes rank-64 all-linear transformer layers, excluding embeddings/output head. Floating checkpoint elements may include buffers; no separately allocated reference model is assumed.", + "shard_header_sha256": { + "model-00001-of-00002.safetensors": "29dd019fbd86197c6d12794e36555ab5bbd74a3b0acfe7c0736176bbea03a2ca", + "model-00002-of-00002.safetensors": "54aed6db58d46324a40d42491b03f1e6d0c06da5022d61457d992c4f570d4c41" + }, + "tensor_count": 194, + "tie_word_embeddings": true, + "vocab_size": 200064 + } +} diff --git a/benchmark/estimation/numerical_diagnostic.json b/benchmark/estimation/numerical_diagnostic.json new file mode 100644 index 0000000000000000000000000000000000000000..fa44b8d3669007e18d9475a06629fc2c64aa8d8f --- /dev/null +++ b/benchmark/estimation/numerical_diagnostic.json @@ -0,0 +1,43 @@ +{ + "schema_version": 1, + "diagnostic_date": "2026-09-15", + "prediction_seal_sha256": "5ed714caff922a5109346a4242545d19f2824636d0f52bb0bbd2a09b40ed4666", + "original_seal_and_predictions_unchanged": true, + "interpreter": "project-local .venv-reproduce/bin/python", + "versions": { + "python": "3.11.4", + "numpy": "1.24.3", + "scipy": "1.10.1", + "scikit-learn": "1.3.0", + "threadpoolctl": "2.2.0" + }, + "numpy_backend": { + "library": "libopenblas64_p-r0-15028c96.3.21.so", + "sha256": "554bde1d8a0c71d8dc21ae74de05c44da4fff5dbc6791a819f6acf5adfe90bd9", + "configuration": "OpenBLAS 0.3.21 USE64BITINT DYNAMIC_ARCH NO_AFFINITY Zen MAX_THREADS=64", + "historical_fitting_thread_limit": "not established", + "cause_of_discovery_warning": "threadpoolctl looked for unsuffixed openblas_get_config; this NumPy build exports openblas_get_config64_. Its fallback returned None and the callback raised AttributeError on split().", + "diagnostic_getter_threads": 64, + "diagnostic_thread_count_interpretation": "Getter result in a later read-only process, not a historical fitting measurement or active-worker count." + }, + "other_reported_pools": "The original seal records SciPy OpenBLAS 0.3.18 and scikit-learn OpenMP at one thread; that does not establish the NumPy pool.", + "independent_saved_solution_checks": { + "prospective_prediction_rows_exact": 36, + "retrospective_prediction_rows_exact": 270, + "retrospective_scores_and_costs_exact": true, + "nnls_optimality_checked": true, + "all_parameters_finite": true, + "nnls_coefficients_nonnegative": true, + "smallest_scaled_gradient_above": -1e-12, + "largest_active_coordinate_gradient_magnitude_below": 1e-12, + "conclusion": "No scientific correctness blocker found; CPU performance/thread-control claim remains limited." + }, + "separate_parent_fresh_fit_check": { + "status": "passed", + "relative_tolerance": 1e-9, + "absolute_tolerance": 1e-6, + "categorical_labels_exact": true, + "prospective_outcomes_read": false, + "original_outputs_overwritten": false + } +} diff --git a/benchmark/estimation/prediction_freeze.json b/benchmark/estimation/prediction_freeze.json new file mode 100644 index 0000000000000000000000000000000000000000..cfb3d0592b5be6c7ef9c22110a95f85d596180fb --- /dev/null +++ b/benchmark/estimation/prediction_freeze.json @@ -0,0 +1,62 @@ +{ + "amendments_sha256": { + "amendment-01.json": "b02fe0419f865c4cc3d77fb35bb4205a54f3aa9605fdadc7b86a9a11b172d25e" + }, + "effective_input_sha256": { + "benchmark.py": "35e36caad46d2814b0ac9f8180961de5564746347cf668d5c3fc1e643fd477e8", + "benchmark/attempts.csv": "397c8caaf0d3abe06587fd649c181fe71d5b8b5f5e24e0591cc7e56d38e275c1", + "benchmark/configurations.csv": "3a77c96a67b82e01d23fb6f7a440cba23041bed1dae2749072cb9abb5b5d239d", + "benchmark/estimation/matrix.csv": "1ff175e22c76bf1af3857b4ae57cd4f78bafe0d280a295d696e96815ef85e9c8", + "benchmark/estimation/model_metadata.json": "2f0ce16c13e72900d2a60f9b7ad28d4dc5ed43d7915d8a88925ba9c5ffd684ad", + "benchmark/estimation/targets.csv": "f2481278127fa726f24297d81fc9df46e999315fe543167ed398e671fffa522f", + "benchmark/outcomes.csv": "af05cd91ed1de843735cc188e5e54dbd1645b3ed91c97f503ac3de67bc5f61c9", + "memory_tuner/estimation_baselines.py": "8cc20f9697eb38cb6a19a2255e95712ac44469cbd147eaf93e7e7ed183360ad2", + "memory_tuner/model_memory_metadata.py": "8845a1197b49d9463ebe5dc8c953463dd2106361c54e7b12244057c48aaa2712", + "memory_tuner/plan_estimation_study.py": "16855f584441528870fcca38c3a19eafd1234c3eacb9b43cc9eebdd4bd6ad862" + }, + "files_sha256": { + "fitted_model.json": "8b84aa1ae57fe56fd478817dbbce505487cbef6b26652ae2c8f45e9fa5ef5df4", + "predictions.json": "96d7f2ae11145f0985c0faf743333d941afbcd3b80966a857b1af0af15306ace", + "retrospective.json": "8ada76da63ab91e0b1ca6276a98f2b096979fe6acd7578c1223469d55d399f3a" + }, + "fit_and_predict_cpu_wall_seconds": 0.08008193783462048, + "frozen_at_utc": "2026-09-15T14:10:21.535819+00:00", + "implementation_sha256": "8cc20f9697eb38cb6a19a2255e95712ac44469cbd147eaf93e7e7ed183360ad2", + "numerical_environment": { + "fit_thread_limit": 1, + "numpy": "1.24.3", + "platform": "Linux-4.18.0-553.146.1.el8_10.x86_64-x86_64-with-glibc2.28", + "python": "3.11.4", + "scikit_learn": "1.3.0", + "scipy": "1.10.1", + "thread_environment": { + "MKL_NUM_THREADS": null, + "OMP_NUM_THREADS": null, + "OPENBLAS_NUM_THREADS": null + }, + "threadpools": [ + { + "filepath": "/mnt/proj3/open-35-44/verl/rl/.venv-reproduce/lib/python3.11/site-packages/scikit_learn.libs/libgomp-a34b3233.so.1.0.0", + "internal_api": "openmp", + "num_threads": 1, + "prefix": "libgomp", + "user_api": "openmp", + "version": null + }, + { + "architecture": "Zen", + "filepath": "/mnt/proj3/open-35-44/verl/rl/.venv-reproduce/lib/python3.11/site-packages/scipy.libs/libopenblasp-r0-41284840.3.18.so", + "internal_api": "openblas", + "num_threads": 1, + "prefix": "libopenblas", + "threading_layer": "pthreads", + "user_api": "blas", + "version": "0.3.18" + } + ] + }, + "predictions": 36, + "protocol_sha256": "1694e447e842600612604891e7b3d036ff0791fa84c4bdbeba54ed093feefdeb", + "status": "sealed_before_new_standard_grpo_execution", + "target_configurations": 12 +} diff --git a/benchmark/estimation/predictions.json b/benchmark/estimation/predictions.json new file mode 100644 index 0000000000000000000000000000000000000000..bee8d62ed13f33ea9e0556ca534dd177f80a8f40 --- /dev/null +++ b/benchmark/estimation/predictions.json @@ -0,0 +1,578 @@ +[ + { + "configuration_id": "est-qwen25-7b-gsm8k-c2-2gpu", + "donor_configuration_id": "cfg-338d9224eb9d2f3a", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 33108.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39409.6357421875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "est-qwen25-7b-gsm8k-c2-2gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "memory_failure", + "predicted_peak_mib": 46461.218916058904, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39409.6357421875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "est-qwen25-7b-gsm8k-c2-2gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39409.6357421875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "est-qwen25-7b-gsm8k-c2-4gpu", + "donor_configuration_id": "cfg-399f6b038d6f65f9", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 29840.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31992.81787109375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "est-qwen25-7b-gsm8k-c2-4gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 38751.70456713468, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31992.81787109375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "est-qwen25-7b-gsm8k-c2-4gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "above_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31992.81787109375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "est-qwen25-7b-gsm8k-c3-2gpu", + "donor_configuration_id": "cfg-7a9a974958c57abc", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 37140.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 43505.6357421875, + "predicted_startup_budget_failure": true + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "est-qwen25-7b-gsm8k-c3-2gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "memory_failure", + "predicted_peak_mib": 49851.347837658235, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 43505.6357421875, + "predicted_startup_budget_failure": true + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "est-qwen25-7b-gsm8k-c3-2gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 43505.6357421875, + "predicted_startup_budget_failure": true + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "est-qwen25-7b-gsm8k-c3-4gpu", + "donor_configuration_id": "cfg-cc2c35944156e5d4", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 33872.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 36088.81787109375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "est-qwen25-7b-gsm8k-c3-4gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "memory_failure", + "predicted_peak_mib": 42141.83348873401, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 36088.81787109375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "est-qwen25-7b-gsm8k-c3-4gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 36088.81787109375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "est-qwen25-7b-math-c2-2gpu", + "donor_configuration_id": "cfg-30f98456fdb322c1", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 33446.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39409.6357421875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "est-qwen25-7b-math-c2-2gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "memory_failure", + "predicted_peak_mib": 46391.75739749115, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39409.6357421875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "est-qwen25-7b-math-c2-2gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39409.6357421875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "est-qwen25-7b-math-c2-4gpu", + "donor_configuration_id": "cfg-f03e4e98c682240e", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 30180.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31992.81787109375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "est-qwen25-7b-math-c2-4gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 38682.24304856692, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31992.81787109375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "est-qwen25-7b-math-c2-4gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "above_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31992.81787109375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "est-qwen25-7b-math-c3-2gpu", + "donor_configuration_id": "cfg-f009e9e3891c06f4", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 37478.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 43505.6357421875, + "predicted_startup_budget_failure": true + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "est-qwen25-7b-math-c3-2gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "memory_failure", + "predicted_peak_mib": 49712.42480052272, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 43505.6357421875, + "predicted_startup_budget_failure": true + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "est-qwen25-7b-math-c3-2gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 43505.6357421875, + "predicted_startup_budget_failure": true + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "est-qwen25-7b-math-c3-4gpu", + "donor_configuration_id": "cfg-714c540f76d9525d", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 34212.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 36088.81787109375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "est-qwen25-7b-math-c3-4gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "memory_failure", + "predicted_peak_mib": 42002.910451598495, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 36088.81787109375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "est-qwen25-7b-math-c3-4gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 36088.81787109375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "est-qwen25-7b-code_heavy_tail-c2-2gpu", + "donor_configuration_id": "cfg-e91942abb5b4f847", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 33436.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39409.6357421875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "est-qwen25-7b-code_heavy_tail-c2-2gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "memory_failure", + "predicted_peak_mib": 46443.853536416966, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39409.6357421875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "est-qwen25-7b-code_heavy_tail-c2-2gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39409.6357421875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "est-qwen25-7b-code_heavy_tail-c2-4gpu", + "donor_configuration_id": "cfg-e91942abb5b4f847", + "donor_gpu_count_matches": false, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 33436.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31992.81787109375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "est-qwen25-7b-code_heavy_tail-c2-4gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 38734.33918749274, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31992.81787109375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "est-qwen25-7b-code_heavy_tail-c2-4gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "above_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31992.81787109375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "est-qwen25-7b-code_heavy_tail-c3-2gpu", + "donor_configuration_id": "cfg-1c5278f8be90a50a", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 37682.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 43505.6357421875, + "predicted_startup_budget_failure": true + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "est-qwen25-7b-code_heavy_tail-c3-2gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "memory_failure", + "predicted_peak_mib": 49816.61707837435, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 43505.6357421875, + "predicted_startup_budget_failure": true + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "est-qwen25-7b-code_heavy_tail-c3-2gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 43505.6357421875, + "predicted_startup_budget_failure": true + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "est-qwen25-7b-code_heavy_tail-c3-4gpu", + "donor_configuration_id": "cfg-1c5278f8be90a50a", + "donor_gpu_count_matches": false, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 37682.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 36088.81787109375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "est-qwen25-7b-code_heavy_tail-c3-4gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "memory_failure", + "predicted_peak_mib": 42107.102729450125, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 36088.81787109375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "est-qwen25-7b-code_heavy_tail-c3-4gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 36088.81787109375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + } +] diff --git a/benchmark/estimation/protocol.json b/benchmark/estimation/protocol.json new file mode 100644 index 0000000000000000000000000000000000000000..89f1fc45684e2129acdaee179eab8a880aad42cc --- /dev/null +++ b/benchmark/estimation/protocol.json @@ -0,0 +1,206 @@ +{ + "allocation_gpus": 4, + "cache_policy": "Fresh process and Ray/runtime identifiers per period. Shared immutable model snapshot and identical preparation; writable compiler/cache policy identical between periods and recorded by runner. Order randomization does not eliminate all thermal/cache carryover. No hardware-confinement claim from a mask alone.", + "classifier": { + "C": 1.0, + "class_weight": "balanced", + "convergence": "ConvergenceWarning or nonfinite parameters is a hard failure", + "decision": "argmax; no probability calibration or safety interpretation", + "implementation": "sklearn.linear_model.LogisticRegression", + "max_iter": 10000, + "multi_class": "multinomial", + "penalty": "l2", + "random_state": 20260915, + "scaling": "Training-fold StandardScaler", + "solver": "lbfgs" + }, + "collection_order_seed": 2026091503, + "comparison": "Two versus four task GPUs on the same four-GPU allocation; matched global prompts, compound level, seed, source and horizon. Generated responses may differ; this is the total execution consequence, not identical tensor work or pure memory sharding.", + "component_regression": "Ordinary nonnegative least squares with intercept, maxiter10000, float64. Only configurations with three completed repeats supply numerical targets, their maximum external completed peak. Feature divisors=max(training-completion feature max,1). This is empirical whole-run component regression, not exact tensor liveness or a sum/max of measured stage peaks. Intercept and coefficients jointly fit runtime/activation effects; coefficients are not causal components.", + "cost": "Report full permitted fitting pool's actual attempts separately from donor-copy selected subset. Completed-only regression still reports full pool acquisition. Existing source reuse has zero incremental GPU collection; a cold start does not. CPU fitting/prediction time separately.", + "cost_scope": "Both periods reserve four GPUs. Report task-provisioned GPU-time (2*t2+4*t4), reservation during invocations 4*(t2+t4), and whole allocation4*T separately. Include failed/unresolved work; queue wait is separate. None is utilization-weighted time.", + "data_sha256": { + "data/codecontests/heavy_tail/test.parquet": "e9d077a023de7f80aa013c5541c33b5978a43e203ce916045bc72c4fe5c62a82", + "data/codecontests/heavy_tail/train.parquet": "0bc6ec8fd7ef1618a604417faa4bb13cb9a0117dcffb75685b2c80c1da95f697", + "data/gsm8k/test.parquet": "0dcd50ed32caa9d8f434d4e3ef8031bc3e49b0f823655a53be5c002e727f17dd", + "data/gsm8k/train.parquet": "8fb13a0cd8621e5fcacdee3ffe28b3a369975478ea7ab514781f3f190cba2488", + "data/math/test.parquet": "e0cb48b1e49874c71f4a02dff944eba08844067d98dc5bdc89c004644e6b597d", + "data/math/train.parquet": "4aa303577e968be6f810600730c970b14be659a9ae9072cc2afefd04d67e7cb8" + }, + "device_capacity_mib": 40960, + "donor_rule": "Same workload and compound level required. Prefer same GPU count, then same architecture-family prior, then smallest absolute log parameter-count ratio; tie-break by configuration_id. A missing exact GPU match uses a disclosed different-GPU donor. No outcomes affect donor selection.", + "evaluation_seeds": [ + 151, + 152, + 153 + ], + "failure_policy": "No reruns/replacements/requeues. Each period has a persistent launch ledger. Continue the second period after any first-period outcome only after all four devices pass identity/cleanup/idle checks. Otherwise retain not-started. Missing GPU identity/evidence is unresolved, not OOM. Early validated OOM need not reach full worker initialization. Do not drop incomplete pairs.", + "feature_scope": "Architecture/checkpoint-header and configured shape volumes only. Largest-unit proxy=max(largest decoder block, root non-block elements, largest weight tensor); not an observed wrapping map. Checkpoint-boundary and vocabulary-output volumes are predictive covariates, not exact live activation totals. No source identity, dataset name, compound level, actual generated length or outcome appears as a numeric feature.", + "features": [ + "actor_shards_gib", + "adapter_optimizer_gib", + "generation_budget_gib", + "generation_replica_gib", + "largest_unit_proxy_gib", + "checkpoint_boundary_volume_gib", + "vocabulary_output_volume_gib", + "actor_parameter_offload" + ], + "frozen_at_utc": "2026-09-15T13:33:20.483317+00:00", + "gpus": "NVIDIA A100-SXM4-40GB", + "input_sha256": { + "benchmark/attempts.csv": "397c8caaf0d3abe06587fd649c181fe71d5b8b5f5e24e0591cc7e56d38e275c1", + "benchmark/configurations.csv": "3a77c96a67b82e01d23fb6f7a440cba23041bed1dae2749072cb9abb5b5d239d", + "benchmark/estimation/matrix.csv": "1ff175e22c76bf1af3857b4ae57cd4f78bafe0d280a295d696e96815ef85e9c8", + "benchmark/estimation/model_metadata.json": "2f0ce16c13e72900d2a60f9b7ad28d4dc5ed43d7915d8a88925ba9c5ffd684ad", + "benchmark/estimation/targets.csv": "f2481278127fa726f24297d81fc9df46e999315fe543167ed398e671fffa522f", + "benchmark/outcomes.csv": "af05cd91ed1de843735cc188e5e54dbd1645b3ed91c97f503ac3de67bc5f61c9", + "memory_tuner/estimation_baselines.py": "0658f92a18147fd67909ca49fa3dc32020217cfd61d34651161c95baa80dd55a", + "memory_tuner/model_memory_metadata.py": "8845a1197b49d9463ebe5dc8c953463dd2106361c54e7b12244057c48aaa2712", + "memory_tuner/plan_estimation_study.py": "16855f584441528870fcca38c3a19eafd1234c3eacb9b43cc9eebdd4bd6ad862" + }, + "label_mapping": "Component regression first rejects a positive resident-actor startup-budget check as memory_failure. Otherwise regression peak > capacity predicts memory_failure; otherwise > margin predicts above_margin; otherwise within_margin. This is an evaluated heuristic, not a claim that every allocation failure requires an observed capacity crossing.", + "memory_margin_limit_mib": 38912, + "model": "Qwen/Qwen2.5-7B-Instruct", + "model_revision": "a09a35458c702b33eeacc393d103063234e8bc28", + "outcome_rule": "A configuration has a repeated memory label only if all three seed slots have validated memory outcomes. Any diagnosed memory failure then implies memory_failure; otherwise any completed peak above38912MiB implies above_margin; otherwise within_margin. Unresolved/not-started slots remain explicit with known failure/margin flags, not replaced or discarded.", + "pair_order_seed": 2026091501, + "planned_pairs": 18, + "planned_processes": 36, + "prior_visibility": "Old labels and prior manuscript findings were public during design. Retrospective holdout is not blind method development. Historical7B exploratory work used other workflows and is excluded from fitting/selection. New standardGRPO predictions are frozen before their outcomes, not an unseen-family/hardware evaluation.", + "prospective_design": "All 90 old configurations fit predictions for 12 new 7B configurations before execution.", + "protocol_version": "estimation-1.0", + "randomization": "9 two-GPU-first and 9 four-GPU-first pairs, balanced as closely as possible within each three-seed workload/level cell. Each of six two-of-four allocation-ordinal subsets occurs three times, assigned independently of period order. Physical UUIDs resolve before either period.", + "resource_cap": { + "gpus_per_allocation": 4, + "idle_timeout_seconds": 180, + "initial_allocations": 18, + "max_concurrent_allocations": 3, + "maximum_reserved_gpu_hours": 66, + "payload_timeout_seconds": 1500, + "requeue": false, + "retries": 0, + "wall_minutes_per_allocation": 55 + }, + "retrospective_design": "Three leave-model-family-out folds, 60 fitting / 30 test configurations each.", + "source_configuration_ids": [ + "cfg-072bd7669d20f8fa", + "cfg-0d9ba9215ab5683e", + "cfg-0f540d3f6fe4661b", + "cfg-0f8a31ef7ac4c551", + "cfg-10644b5d0a82d5c3", + "cfg-1432bb8a4d9c3b8c", + "cfg-17e69bcf1b041bd9", + "cfg-18285c3cb50308d4", + "cfg-1c5278f8be90a50a", + "cfg-2afbe250af9cd3e1", + "cfg-301e4dae7048c6ec", + "cfg-30f98456fdb322c1", + "cfg-32977d5aa42a41cb", + "cfg-338d9224eb9d2f3a", + "cfg-352f4dbc55fbcd3c", + "cfg-371ae37408089efe", + "cfg-399f6b038d6f65f9", + "cfg-3c86a11fdb8455cb", + "cfg-3dc042a103546eac", + "cfg-3f0f48a3c570a0c8", + "cfg-49d9bccfe873f5c6", + "cfg-5244d13eac3116ad", + "cfg-55811d23c5144bb7", + "cfg-5839ccf95c2c1722", + "cfg-5ff9621270e710ed", + "cfg-61c14de21a34d0e6", + "cfg-62f371a80968a0ba", + "cfg-6a4562db98b38d6f", + "cfg-6d54ac65d934f404", + "cfg-714c540f76d9525d", + "cfg-73c06764ba28a56b", + "cfg-78c440d3891ddbc2", + "cfg-7a0e81934d3398ee", + "cfg-7a9a974958c57abc", + "cfg-7df2428553b76877", + "cfg-7f23cab692546e3f", + "cfg-835067f7405e89a2", + "cfg-896685010703c0fb", + "cfg-89ca48ce88ea694e", + "cfg-8bd7050cd7f49e78", + "cfg-9160dc45a0171a10", + "cfg-93d4f41e600c9079", + "cfg-966f5804dc918967", + "cfg-96ccf395b84393f7", + "cfg-97107ceba8bf0cbe", + "cfg-9758870c0a0b4d69", + "cfg-9c461415e6bb9702", + "cfg-9f624ec22aa3dde0", + "cfg-a026693bae37d30b", + "cfg-a23a9ffc1f670139", + "cfg-a24e9cdd2f6143f2", + "cfg-a3987314cb2f9f12", + "cfg-a94bf7877633dfda", + "cfg-aa33c72e0a449b9c", + "cfg-acfa4acda15c3599", + "cfg-b5007322fb2e1772", + "cfg-b5352215babfeba6", + "cfg-b56fb353c2f2fa3e", + "cfg-b8dd70ab682c13b2", + "cfg-b8e90bd69c9744ea", + "cfg-bae2977da7957f72", + "cfg-bfb83ff32b7fd921", + "cfg-c0605c1607caa079", + "cfg-c2a8e6e8e140e109", + "cfg-c3f0ad157d299d39", + "cfg-ca71b56dc3189305", + "cfg-cad8341dbbff10ba", + "cfg-cc2c35944156e5d4", + "cfg-cc5f4f0c2e1256ba", + "cfg-cc900fb7731c3f49", + "cfg-cd3974595f9db882", + "cfg-d02503f8c9804a2d", + "cfg-d138048feeb8fe5d", + "cfg-d4a213515805a9a7", + "cfg-d9397d15f234d541", + "cfg-db87c67536995ab0", + "cfg-e117f15dd5ccfb30", + "cfg-e2a205be8efba055", + "cfg-e8893b12558f51dd", + "cfg-e91942abb5b4f847", + "cfg-ea5a23a40ad5ca50", + "cfg-ea92cccc91029465", + "cfg-f009e9e3891c06f4", + "cfg-f03e4e98c682240e", + "cfg-f439d2dd90137a80", + "cfg-f75b68f479767290", + "cfg-fad1b555f8b37235", + "cfg-fbe1c0422f98cb38", + "cfg-fd4d4789145230df", + "cfg-fd5bb0db68b0dae3" + ], + "source_configurations": 90, + "source_selection": "Only 72 boundary and 18 four_gpu source configurations; all prior7B exploration, other controls and the 48-invocation prospective admission panel are excluded. No new two-GPU outcome updates predictions for four GPUs or vice versa.", + "source_studies": [ + "boundary", + "four_gpu" + ], + "startup_check": "Separate one-sided diagnostic only for resident actor: 4(P+R)/G+u*C>C. Shared-reference counted once; generation budget includes its replica/cache. Assumes actor shards resident before vLLM free-memory check. Passing does not predict completion or margin compliance; no target observation used.", + "status": "specified_before_fitting_and_new_standard_grpo_outcomes", + "subset_seed": 2026091502, + "target_configuration_ids": [ + "est-qwen25-7b-gsm8k-c2-2gpu", + "est-qwen25-7b-gsm8k-c2-4gpu", + "est-qwen25-7b-gsm8k-c3-2gpu", + "est-qwen25-7b-gsm8k-c3-4gpu", + "est-qwen25-7b-math-c2-2gpu", + "est-qwen25-7b-math-c2-4gpu", + "est-qwen25-7b-math-c3-2gpu", + "est-qwen25-7b-math-c3-4gpu", + "est-qwen25-7b-code_heavy_tail-c2-2gpu", + "est-qwen25-7b-code_heavy_tail-c2-4gpu", + "est-qwen25-7b-code_heavy_tail-c3-2gpu", + "est-qwen25-7b-code_heavy_tail-c3-4gpu" + ], + "target_configurations": 12, + "task_gpu_counts": [ + 2, + 4 + ] +} diff --git a/benchmark/estimation/results/configurations.csv b/benchmark/estimation/results/configurations.csv new file mode 100644 index 0000000000000000000000000000000000000000..bf85e562b5a94e2d92fd780daa2728b2a2d1137b --- /dev/null +++ b/benchmark/estimation/results/configurations.csv @@ -0,0 +1,13 @@ +completed_peak_mib,completed_processes,configuration_id,eligible_processes,eligible_seeds,known_margin_exceedance,known_memory_failure,maximum_known_completed_peak_mib,planned_seeds,settings,state,unresolved_seeds +,0,est-qwen25-7b-gsm8k-c2-2gpu,3,"[151, 152, 153]",False,True,,"[151, 152, 153]","{""accelerator"": ""NVIDIA A100-SXM4-40GB"", ""actor_micro_batch"": ""8"", ""adaptation"": ""LoRA"", ""algorithm"": ""GRPO"", ""allocator_trace_enabled"": ""True"", ""allocator_trace_sync"": ""1"", ""case_id"": ""est-qwen25-7b-gsm8k-2gpu"", ""condition"": """", ""configuration_id"": ""est-qwen25-7b-gsm8k-c2-2gpu"", ""configuration_level"": ""c2"", ""device_capacity_mib"": ""40960"", ""free_cache_engine"": ""True"", ""gpu_count"": ""2"", ""gpu_monitor_interval_ms"": ""100"", ""legacy_dataset_id"": ""gsm8k"", ""margin_limit_mib"": ""38912"", ""max_actor_ckpt_to_keep"": ""1"", ""max_model_len"": ""3072"", ""max_num_seqs"": ""64"", ""max_prompt_length"": ""1024"", ""max_response_length"": ""2048"", ""model_family"": ""qwen25_7b"", ""model_id"": ""Qwen/Qwen2.5-7B-Instruct"", ""model_name"": ""Qwen2.5-7B-Instruct"", ""optimizer_offload"": ""False"", ""parameter_offload"": ""False"", ""peak_measurement"": ""maximum externally sampled per-device memory"", ""ref_logprob_micro_batch"": ""4"", ""resume_mode"": ""disable"", ""rollout_logprob_micro_batch"": ""4"", ""rollout_n"": ""2"", ""rollout_tp_size"": ""1"", ""save_freq"": ""-1"", ""study"": ""prospective_estimation"", ""test_freq"": ""-1"", ""total_training_steps"": ""1"", ""train_batch_size"": ""32"", ""train_max_samples"": ""32"", ""val_before_train"": ""False"", ""val_max_samples"": ""32"", ""vllm_gpu_memory_utilization"": ""0.6"", ""workload_id"": ""gsm8k"", ""workload_name"": ""GSM8K mathematics""}",memory_failure,[] +,0,est-qwen25-7b-gsm8k-c2-4gpu,0,[],False,False,,"[151, 152, 153]","{""accelerator"": ""NVIDIA A100-SXM4-40GB"", ""actor_micro_batch"": ""8"", ""adaptation"": ""LoRA"", ""algorithm"": ""GRPO"", ""allocator_trace_enabled"": ""True"", ""allocator_trace_sync"": ""1"", ""case_id"": ""est-qwen25-7b-gsm8k-4gpu"", ""condition"": """", ""configuration_id"": ""est-qwen25-7b-gsm8k-c2-4gpu"", ""configuration_level"": ""c2"", ""device_capacity_mib"": ""40960"", ""free_cache_engine"": ""True"", ""gpu_count"": ""4"", ""gpu_monitor_interval_ms"": ""100"", ""legacy_dataset_id"": ""gsm8k"", ""margin_limit_mib"": ""38912"", ""max_actor_ckpt_to_keep"": ""1"", ""max_model_len"": ""3072"", ""max_num_seqs"": ""64"", ""max_prompt_length"": ""1024"", ""max_response_length"": ""2048"", ""model_family"": ""qwen25_7b"", ""model_id"": ""Qwen/Qwen2.5-7B-Instruct"", ""model_name"": ""Qwen2.5-7B-Instruct"", ""optimizer_offload"": ""False"", ""parameter_offload"": ""False"", ""peak_measurement"": ""maximum externally sampled per-device memory"", ""ref_logprob_micro_batch"": ""4"", ""resume_mode"": ""disable"", ""rollout_logprob_micro_batch"": ""4"", ""rollout_n"": ""2"", ""rollout_tp_size"": ""1"", ""save_freq"": ""-1"", ""study"": ""prospective_estimation"", ""test_freq"": ""-1"", ""total_training_steps"": ""1"", ""train_batch_size"": ""32"", ""train_max_samples"": ""32"", ""val_before_train"": ""False"", ""val_max_samples"": ""32"", ""vllm_gpu_memory_utilization"": ""0.6"", ""workload_id"": ""gsm8k"", ""workload_name"": ""GSM8K mathematics""}",unresolved,"[151, 152, 153]" +,0,est-qwen25-7b-gsm8k-c3-2gpu,3,"[151, 152, 153]",False,True,,"[151, 152, 153]","{""accelerator"": ""NVIDIA A100-SXM4-40GB"", ""actor_micro_batch"": ""16"", ""adaptation"": ""LoRA"", ""algorithm"": ""GRPO"", ""allocator_trace_enabled"": ""True"", ""allocator_trace_sync"": ""1"", ""case_id"": ""est-qwen25-7b-gsm8k-2gpu"", ""condition"": """", ""configuration_id"": ""est-qwen25-7b-gsm8k-c3-2gpu"", ""configuration_level"": ""c3"", ""device_capacity_mib"": ""40960"", ""free_cache_engine"": ""True"", ""gpu_count"": ""2"", ""gpu_monitor_interval_ms"": ""100"", ""legacy_dataset_id"": ""gsm8k"", ""margin_limit_mib"": ""38912"", ""max_actor_ckpt_to_keep"": ""1"", ""max_model_len"": ""3072"", ""max_num_seqs"": ""64"", ""max_prompt_length"": ""1024"", ""max_response_length"": ""2048"", ""model_family"": ""qwen25_7b"", ""model_id"": ""Qwen/Qwen2.5-7B-Instruct"", ""model_name"": ""Qwen2.5-7B-Instruct"", ""optimizer_offload"": ""False"", ""parameter_offload"": ""False"", ""peak_measurement"": ""maximum externally sampled per-device memory"", ""ref_logprob_micro_batch"": ""4"", ""resume_mode"": ""disable"", ""rollout_logprob_micro_batch"": ""4"", ""rollout_n"": ""2"", ""rollout_tp_size"": ""1"", ""save_freq"": ""-1"", ""study"": ""prospective_estimation"", ""test_freq"": ""-1"", ""total_training_steps"": ""1"", ""train_batch_size"": ""32"", ""train_max_samples"": ""32"", ""val_before_train"": ""False"", ""val_max_samples"": ""32"", ""vllm_gpu_memory_utilization"": ""0.7"", ""workload_id"": ""gsm8k"", ""workload_name"": ""GSM8K mathematics""}",memory_failure,[] +,0,est-qwen25-7b-gsm8k-c3-4gpu,3,"[151, 152, 153]",False,True,,"[151, 152, 153]","{""accelerator"": ""NVIDIA A100-SXM4-40GB"", ""actor_micro_batch"": ""16"", ""adaptation"": ""LoRA"", ""algorithm"": ""GRPO"", ""allocator_trace_enabled"": ""True"", ""allocator_trace_sync"": ""1"", ""case_id"": ""est-qwen25-7b-gsm8k-4gpu"", ""condition"": """", ""configuration_id"": ""est-qwen25-7b-gsm8k-c3-4gpu"", ""configuration_level"": ""c3"", ""device_capacity_mib"": ""40960"", ""free_cache_engine"": ""True"", ""gpu_count"": ""4"", ""gpu_monitor_interval_ms"": ""100"", ""legacy_dataset_id"": ""gsm8k"", ""margin_limit_mib"": ""38912"", ""max_actor_ckpt_to_keep"": ""1"", ""max_model_len"": ""3072"", ""max_num_seqs"": ""64"", ""max_prompt_length"": ""1024"", ""max_response_length"": ""2048"", ""model_family"": ""qwen25_7b"", ""model_id"": ""Qwen/Qwen2.5-7B-Instruct"", ""model_name"": ""Qwen2.5-7B-Instruct"", ""optimizer_offload"": ""False"", ""parameter_offload"": ""False"", ""peak_measurement"": ""maximum externally sampled per-device memory"", ""ref_logprob_micro_batch"": ""4"", ""resume_mode"": ""disable"", ""rollout_logprob_micro_batch"": ""4"", ""rollout_n"": ""2"", ""rollout_tp_size"": ""1"", ""save_freq"": ""-1"", ""study"": ""prospective_estimation"", ""test_freq"": ""-1"", ""total_training_steps"": ""1"", ""train_batch_size"": ""32"", ""train_max_samples"": ""32"", ""val_before_train"": ""False"", ""val_max_samples"": ""32"", ""vllm_gpu_memory_utilization"": ""0.7"", ""workload_id"": ""gsm8k"", ""workload_name"": ""GSM8K mathematics""}",memory_failure,[] +,0,est-qwen25-7b-math-c2-2gpu,1,[152],False,True,,"[151, 152, 153]","{""accelerator"": ""NVIDIA A100-SXM4-40GB"", ""actor_micro_batch"": ""8"", ""adaptation"": ""LoRA"", ""algorithm"": ""GRPO"", ""allocator_trace_enabled"": ""True"", ""allocator_trace_sync"": ""1"", ""case_id"": ""est-qwen25-7b-math-2gpu"", ""condition"": """", ""configuration_id"": ""est-qwen25-7b-math-c2-2gpu"", ""configuration_level"": ""c2"", ""device_capacity_mib"": ""40960"", ""free_cache_engine"": ""True"", ""gpu_count"": ""2"", ""gpu_monitor_interval_ms"": ""100"", ""legacy_dataset_id"": ""math"", ""margin_limit_mib"": ""38912"", ""max_actor_ckpt_to_keep"": ""1"", ""max_model_len"": ""1024"", ""max_num_seqs"": ""128"", ""max_prompt_length"": ""512"", ""max_response_length"": ""512"", ""model_family"": ""qwen25_7b"", ""model_id"": ""Qwen/Qwen2.5-7B-Instruct"", ""model_name"": ""Qwen2.5-7B-Instruct"", ""optimizer_offload"": ""False"", ""parameter_offload"": ""False"", ""peak_measurement"": ""maximum externally sampled per-device memory"", ""ref_logprob_micro_batch"": ""8"", ""resume_mode"": ""disable"", ""rollout_logprob_micro_batch"": ""8"", ""rollout_n"": ""2"", ""rollout_tp_size"": ""1"", ""save_freq"": ""-1"", ""study"": ""prospective_estimation"", ""test_freq"": ""-1"", ""total_training_steps"": ""1"", ""train_batch_size"": ""64"", ""train_max_samples"": ""64"", ""val_before_train"": ""False"", ""val_max_samples"": ""64"", ""vllm_gpu_memory_utilization"": ""0.6"", ""workload_id"": ""math"", ""workload_name"": ""MATH mathematics""}",unresolved,"[151, 153]" +,0,est-qwen25-7b-math-c2-4gpu,0,[],False,False,,"[151, 152, 153]","{""accelerator"": ""NVIDIA A100-SXM4-40GB"", ""actor_micro_batch"": ""8"", ""adaptation"": ""LoRA"", ""algorithm"": ""GRPO"", ""allocator_trace_enabled"": ""True"", ""allocator_trace_sync"": ""1"", ""case_id"": ""est-qwen25-7b-math-4gpu"", ""condition"": """", ""configuration_id"": ""est-qwen25-7b-math-c2-4gpu"", ""configuration_level"": ""c2"", ""device_capacity_mib"": ""40960"", ""free_cache_engine"": ""True"", ""gpu_count"": ""4"", ""gpu_monitor_interval_ms"": ""100"", ""legacy_dataset_id"": ""math"", ""margin_limit_mib"": ""38912"", ""max_actor_ckpt_to_keep"": ""1"", ""max_model_len"": ""1024"", ""max_num_seqs"": ""128"", ""max_prompt_length"": ""512"", ""max_response_length"": ""512"", ""model_family"": ""qwen25_7b"", ""model_id"": ""Qwen/Qwen2.5-7B-Instruct"", ""model_name"": ""Qwen2.5-7B-Instruct"", ""optimizer_offload"": ""False"", ""parameter_offload"": ""False"", ""peak_measurement"": ""maximum externally sampled per-device memory"", ""ref_logprob_micro_batch"": ""8"", ""resume_mode"": ""disable"", ""rollout_logprob_micro_batch"": ""8"", ""rollout_n"": ""2"", ""rollout_tp_size"": ""1"", ""save_freq"": ""-1"", ""study"": ""prospective_estimation"", ""test_freq"": ""-1"", ""total_training_steps"": ""1"", ""train_batch_size"": ""64"", ""train_max_samples"": ""64"", ""val_before_train"": ""False"", ""val_max_samples"": ""64"", ""vllm_gpu_memory_utilization"": ""0.6"", ""workload_id"": ""math"", ""workload_name"": ""MATH mathematics""}",unresolved,"[151, 152, 153]" +,0,est-qwen25-7b-math-c3-2gpu,2,"[151, 153]",False,True,,"[151, 152, 153]","{""accelerator"": ""NVIDIA A100-SXM4-40GB"", ""actor_micro_batch"": ""16"", ""adaptation"": ""LoRA"", ""algorithm"": ""GRPO"", ""allocator_trace_enabled"": ""True"", ""allocator_trace_sync"": ""1"", ""case_id"": ""est-qwen25-7b-math-2gpu"", ""condition"": """", ""configuration_id"": ""est-qwen25-7b-math-c3-2gpu"", ""configuration_level"": ""c3"", ""device_capacity_mib"": ""40960"", ""free_cache_engine"": ""True"", ""gpu_count"": ""2"", ""gpu_monitor_interval_ms"": ""100"", ""legacy_dataset_id"": ""math"", ""margin_limit_mib"": ""38912"", ""max_actor_ckpt_to_keep"": ""1"", ""max_model_len"": ""1024"", ""max_num_seqs"": ""128"", ""max_prompt_length"": ""512"", ""max_response_length"": ""512"", ""model_family"": ""qwen25_7b"", ""model_id"": ""Qwen/Qwen2.5-7B-Instruct"", ""model_name"": ""Qwen2.5-7B-Instruct"", ""optimizer_offload"": ""False"", ""parameter_offload"": ""False"", ""peak_measurement"": ""maximum externally sampled per-device memory"", ""ref_logprob_micro_batch"": ""8"", ""resume_mode"": ""disable"", ""rollout_logprob_micro_batch"": ""8"", ""rollout_n"": ""2"", ""rollout_tp_size"": ""1"", ""save_freq"": ""-1"", ""study"": ""prospective_estimation"", ""test_freq"": ""-1"", ""total_training_steps"": ""1"", ""train_batch_size"": ""64"", ""train_max_samples"": ""64"", ""val_before_train"": ""False"", ""val_max_samples"": ""64"", ""vllm_gpu_memory_utilization"": ""0.7"", ""workload_id"": ""math"", ""workload_name"": ""MATH mathematics""}",unresolved,[152] +,0,est-qwen25-7b-math-c3-4gpu,2,"[151, 153]",False,True,,"[151, 152, 153]","{""accelerator"": ""NVIDIA A100-SXM4-40GB"", ""actor_micro_batch"": ""16"", ""adaptation"": ""LoRA"", ""algorithm"": ""GRPO"", ""allocator_trace_enabled"": ""True"", ""allocator_trace_sync"": ""1"", ""case_id"": ""est-qwen25-7b-math-4gpu"", ""condition"": """", ""configuration_id"": ""est-qwen25-7b-math-c3-4gpu"", ""configuration_level"": ""c3"", ""device_capacity_mib"": ""40960"", ""free_cache_engine"": ""True"", ""gpu_count"": ""4"", ""gpu_monitor_interval_ms"": ""100"", ""legacy_dataset_id"": ""math"", ""margin_limit_mib"": ""38912"", ""max_actor_ckpt_to_keep"": ""1"", ""max_model_len"": ""1024"", ""max_num_seqs"": ""128"", ""max_prompt_length"": ""512"", ""max_response_length"": ""512"", ""model_family"": ""qwen25_7b"", ""model_id"": ""Qwen/Qwen2.5-7B-Instruct"", ""model_name"": ""Qwen2.5-7B-Instruct"", ""optimizer_offload"": ""False"", ""parameter_offload"": ""False"", ""peak_measurement"": ""maximum externally sampled per-device memory"", ""ref_logprob_micro_batch"": ""8"", ""resume_mode"": ""disable"", ""rollout_logprob_micro_batch"": ""8"", ""rollout_n"": ""2"", ""rollout_tp_size"": ""1"", ""save_freq"": ""-1"", ""study"": ""prospective_estimation"", ""test_freq"": ""-1"", ""total_training_steps"": ""1"", ""train_batch_size"": ""64"", ""train_max_samples"": ""64"", ""val_before_train"": ""False"", ""val_max_samples"": ""64"", ""vllm_gpu_memory_utilization"": ""0.7"", ""workload_id"": ""math"", ""workload_name"": ""MATH mathematics""}",unresolved,[152] +,0,est-qwen25-7b-code_heavy_tail-c2-2gpu,3,"[151, 152, 153]",False,True,,"[151, 152, 153]","{""accelerator"": ""NVIDIA A100-SXM4-40GB"", ""actor_micro_batch"": ""8"", ""adaptation"": ""LoRA"", ""algorithm"": ""GRPO"", ""allocator_trace_enabled"": ""True"", ""allocator_trace_sync"": ""1"", ""case_id"": ""est-qwen25-7b-code_heavy_tail-2gpu"", ""condition"": """", ""configuration_id"": ""est-qwen25-7b-code_heavy_tail-c2-2gpu"", ""configuration_level"": ""c2"", ""device_capacity_mib"": ""40960"", ""free_cache_engine"": ""True"", ""gpu_count"": ""2"", ""gpu_monitor_interval_ms"": ""100"", ""legacy_dataset_id"": ""code_heavy_tail"", ""margin_limit_mib"": ""38912"", ""max_actor_ckpt_to_keep"": ""1"", ""max_model_len"": ""2560"", ""max_num_seqs"": ""64"", ""max_prompt_length"": ""1536"", ""max_response_length"": ""1024"", ""model_family"": ""qwen25_7b"", ""model_id"": ""Qwen/Qwen2.5-7B-Instruct"", ""model_name"": ""Qwen2.5-7B-Instruct"", ""optimizer_offload"": ""False"", ""parameter_offload"": ""False"", ""peak_measurement"": ""maximum externally sampled per-device memory"", ""ref_logprob_micro_batch"": ""4"", ""resume_mode"": ""disable"", ""rollout_logprob_micro_batch"": ""4"", ""rollout_n"": ""2"", ""rollout_tp_size"": ""1"", ""save_freq"": ""-1"", ""study"": ""prospective_estimation"", ""test_freq"": ""-1"", ""total_training_steps"": ""1"", ""train_batch_size"": ""32"", ""train_max_samples"": ""32"", ""val_before_train"": ""False"", ""val_max_samples"": ""32"", ""vllm_gpu_memory_utilization"": ""0.6"", ""workload_id"": ""longer_prompt_code"", ""workload_name"": ""Longer-prompt code""}",memory_failure,[] +,0,est-qwen25-7b-code_heavy_tail-c2-4gpu,0,[],False,False,,"[151, 152, 153]","{""accelerator"": ""NVIDIA A100-SXM4-40GB"", ""actor_micro_batch"": ""8"", ""adaptation"": ""LoRA"", ""algorithm"": ""GRPO"", ""allocator_trace_enabled"": ""True"", ""allocator_trace_sync"": ""1"", ""case_id"": ""est-qwen25-7b-code_heavy_tail-4gpu"", ""condition"": """", ""configuration_id"": ""est-qwen25-7b-code_heavy_tail-c2-4gpu"", ""configuration_level"": ""c2"", ""device_capacity_mib"": ""40960"", ""free_cache_engine"": ""True"", ""gpu_count"": ""4"", ""gpu_monitor_interval_ms"": ""100"", ""legacy_dataset_id"": ""code_heavy_tail"", ""margin_limit_mib"": ""38912"", ""max_actor_ckpt_to_keep"": ""1"", ""max_model_len"": ""2560"", ""max_num_seqs"": ""64"", ""max_prompt_length"": ""1536"", ""max_response_length"": ""1024"", ""model_family"": ""qwen25_7b"", ""model_id"": ""Qwen/Qwen2.5-7B-Instruct"", ""model_name"": ""Qwen2.5-7B-Instruct"", ""optimizer_offload"": ""False"", ""parameter_offload"": ""False"", ""peak_measurement"": ""maximum externally sampled per-device memory"", ""ref_logprob_micro_batch"": ""4"", ""resume_mode"": ""disable"", ""rollout_logprob_micro_batch"": ""4"", ""rollout_n"": ""2"", ""rollout_tp_size"": ""1"", ""save_freq"": ""-1"", ""study"": ""prospective_estimation"", ""test_freq"": ""-1"", ""total_training_steps"": ""1"", ""train_batch_size"": ""32"", ""train_max_samples"": ""32"", ""val_before_train"": ""False"", ""val_max_samples"": ""32"", ""vllm_gpu_memory_utilization"": ""0.6"", ""workload_id"": ""longer_prompt_code"", ""workload_name"": ""Longer-prompt code""}",unresolved,"[151, 152, 153]" +,0,est-qwen25-7b-code_heavy_tail-c3-2gpu,3,"[151, 152, 153]",False,True,,"[151, 152, 153]","{""accelerator"": ""NVIDIA A100-SXM4-40GB"", ""actor_micro_batch"": ""16"", ""adaptation"": ""LoRA"", ""algorithm"": ""GRPO"", ""allocator_trace_enabled"": ""True"", ""allocator_trace_sync"": ""1"", ""case_id"": ""est-qwen25-7b-code_heavy_tail-2gpu"", ""condition"": """", ""configuration_id"": ""est-qwen25-7b-code_heavy_tail-c3-2gpu"", ""configuration_level"": ""c3"", ""device_capacity_mib"": ""40960"", ""free_cache_engine"": ""True"", ""gpu_count"": ""2"", ""gpu_monitor_interval_ms"": ""100"", ""legacy_dataset_id"": ""code_heavy_tail"", ""margin_limit_mib"": ""38912"", ""max_actor_ckpt_to_keep"": ""1"", ""max_model_len"": ""2560"", ""max_num_seqs"": ""64"", ""max_prompt_length"": ""1536"", ""max_response_length"": ""1024"", ""model_family"": ""qwen25_7b"", ""model_id"": ""Qwen/Qwen2.5-7B-Instruct"", ""model_name"": ""Qwen2.5-7B-Instruct"", ""optimizer_offload"": ""False"", ""parameter_offload"": ""False"", ""peak_measurement"": ""maximum externally sampled per-device memory"", ""ref_logprob_micro_batch"": ""4"", ""resume_mode"": ""disable"", ""rollout_logprob_micro_batch"": ""4"", ""rollout_n"": ""2"", ""rollout_tp_size"": ""1"", ""save_freq"": ""-1"", ""study"": ""prospective_estimation"", ""test_freq"": ""-1"", ""total_training_steps"": ""1"", ""train_batch_size"": ""32"", ""train_max_samples"": ""32"", ""val_before_train"": ""False"", ""val_max_samples"": ""32"", ""vllm_gpu_memory_utilization"": ""0.7"", ""workload_id"": ""longer_prompt_code"", ""workload_name"": ""Longer-prompt code""}",memory_failure,[] +,0,est-qwen25-7b-code_heavy_tail-c3-4gpu,3,"[151, 152, 153]",False,True,,"[151, 152, 153]","{""accelerator"": ""NVIDIA A100-SXM4-40GB"", ""actor_micro_batch"": ""16"", ""adaptation"": ""LoRA"", ""algorithm"": ""GRPO"", ""allocator_trace_enabled"": ""True"", ""allocator_trace_sync"": ""1"", ""case_id"": ""est-qwen25-7b-code_heavy_tail-4gpu"", ""condition"": """", ""configuration_id"": ""est-qwen25-7b-code_heavy_tail-c3-4gpu"", ""configuration_level"": ""c3"", ""device_capacity_mib"": ""40960"", ""free_cache_engine"": ""True"", ""gpu_count"": ""4"", ""gpu_monitor_interval_ms"": ""100"", ""legacy_dataset_id"": ""code_heavy_tail"", ""margin_limit_mib"": ""38912"", ""max_actor_ckpt_to_keep"": ""1"", ""max_model_len"": ""2560"", ""max_num_seqs"": ""64"", ""max_prompt_length"": ""1536"", ""max_response_length"": ""1024"", ""model_family"": ""qwen25_7b"", ""model_id"": ""Qwen/Qwen2.5-7B-Instruct"", ""model_name"": ""Qwen2.5-7B-Instruct"", ""optimizer_offload"": ""False"", ""parameter_offload"": ""False"", ""peak_measurement"": ""maximum externally sampled per-device memory"", ""ref_logprob_micro_batch"": ""4"", ""resume_mode"": ""disable"", ""rollout_logprob_micro_batch"": ""4"", ""rollout_n"": ""2"", ""rollout_tp_size"": ""1"", ""save_freq"": ""-1"", ""study"": ""prospective_estimation"", ""test_freq"": ""-1"", ""total_training_steps"": ""1"", ""train_batch_size"": ""32"", ""train_max_samples"": ""32"", ""val_before_train"": ""False"", ""val_max_samples"": ""32"", ""vllm_gpu_memory_utilization"": ""0.7"", ""workload_id"": ""longer_prompt_code"", ""workload_name"": ""Longer-prompt code""}",memory_failure,[] diff --git a/benchmark/estimation/results/configurations.json b/benchmark/estimation/results/configurations.json new file mode 100644 index 0000000000000000000000000000000000000000..68c62e25b2f3e3253afc72124b152fed3f4821c8 --- /dev/null +++ b/benchmark/estimation/results/configurations.json @@ -0,0 +1,797 @@ +[ + { + "completed_peak_mib": null, + "completed_processes": 0, + "configuration_id": "est-qwen25-7b-gsm8k-c2-2gpu", + "eligible_processes": 3, + "eligible_seeds": [ + 151, + 152, + 153 + ], + "known_margin_exceedance": false, + "known_memory_failure": true, + "maximum_known_completed_peak_mib": null, + "planned_seeds": [ + 151, + 152, + 153 + ], + "settings": { + "accelerator": "NVIDIA A100-SXM4-40GB", + "actor_micro_batch": "8", + "adaptation": "LoRA", + "algorithm": "GRPO", + "allocator_trace_enabled": "True", + "allocator_trace_sync": "1", + "case_id": "est-qwen25-7b-gsm8k-2gpu", + "condition": "", + "configuration_id": "est-qwen25-7b-gsm8k-c2-2gpu", + "configuration_level": "c2", + "device_capacity_mib": "40960", + "free_cache_engine": "True", + "gpu_count": "2", + "gpu_monitor_interval_ms": "100", + "legacy_dataset_id": "gsm8k", + "margin_limit_mib": "38912", + "max_actor_ckpt_to_keep": "1", + "max_model_len": "3072", + "max_num_seqs": "64", + "max_prompt_length": "1024", + "max_response_length": "2048", + "model_family": "qwen25_7b", + "model_id": "Qwen/Qwen2.5-7B-Instruct", + "model_name": "Qwen2.5-7B-Instruct", + "optimizer_offload": "False", + "parameter_offload": "False", + "peak_measurement": "maximum externally sampled per-device memory", + "ref_logprob_micro_batch": "4", + "resume_mode": "disable", + "rollout_logprob_micro_batch": "4", + "rollout_n": "2", + "rollout_tp_size": "1", + "save_freq": "-1", + "study": "prospective_estimation", + "test_freq": "-1", + "total_training_steps": "1", + "train_batch_size": "32", + "train_max_samples": "32", + "val_before_train": "False", + "val_max_samples": "32", + "vllm_gpu_memory_utilization": "0.6", + "workload_id": "gsm8k", + "workload_name": "GSM8K mathematics" + }, + "state": "memory_failure", + "unresolved_seeds": [] + }, + { + "completed_peak_mib": null, + "completed_processes": 0, + "configuration_id": "est-qwen25-7b-gsm8k-c2-4gpu", + "eligible_processes": 0, + "eligible_seeds": [], + "known_margin_exceedance": false, + "known_memory_failure": false, + "maximum_known_completed_peak_mib": null, + "planned_seeds": [ + 151, + 152, + 153 + ], + "settings": { + "accelerator": "NVIDIA A100-SXM4-40GB", + "actor_micro_batch": "8", + "adaptation": "LoRA", + "algorithm": "GRPO", + "allocator_trace_enabled": "True", + "allocator_trace_sync": "1", + "case_id": "est-qwen25-7b-gsm8k-4gpu", + "condition": "", + "configuration_id": "est-qwen25-7b-gsm8k-c2-4gpu", + "configuration_level": "c2", + "device_capacity_mib": "40960", + "free_cache_engine": "True", + "gpu_count": "4", + "gpu_monitor_interval_ms": "100", + "legacy_dataset_id": "gsm8k", + "margin_limit_mib": "38912", + "max_actor_ckpt_to_keep": "1", + "max_model_len": "3072", + "max_num_seqs": "64", + "max_prompt_length": "1024", + "max_response_length": "2048", + "model_family": "qwen25_7b", + "model_id": "Qwen/Qwen2.5-7B-Instruct", + "model_name": "Qwen2.5-7B-Instruct", + "optimizer_offload": "False", + "parameter_offload": "False", + "peak_measurement": "maximum externally sampled per-device memory", + "ref_logprob_micro_batch": "4", + "resume_mode": "disable", + "rollout_logprob_micro_batch": "4", + "rollout_n": "2", + "rollout_tp_size": "1", + "save_freq": "-1", + "study": "prospective_estimation", + "test_freq": "-1", + "total_training_steps": "1", + "train_batch_size": "32", + "train_max_samples": "32", + "val_before_train": "False", + "val_max_samples": "32", + "vllm_gpu_memory_utilization": "0.6", + "workload_id": "gsm8k", + "workload_name": "GSM8K mathematics" + }, + "state": "unresolved", + "unresolved_seeds": [ + 151, + 152, + 153 + ] + }, + { + "completed_peak_mib": null, + "completed_processes": 0, + "configuration_id": "est-qwen25-7b-gsm8k-c3-2gpu", + "eligible_processes": 3, + "eligible_seeds": [ + 151, + 152, + 153 + ], + "known_margin_exceedance": false, + "known_memory_failure": true, + "maximum_known_completed_peak_mib": null, + "planned_seeds": [ + 151, + 152, + 153 + ], + "settings": { + "accelerator": "NVIDIA A100-SXM4-40GB", + "actor_micro_batch": "16", + "adaptation": "LoRA", + "algorithm": "GRPO", + "allocator_trace_enabled": "True", + "allocator_trace_sync": "1", + "case_id": "est-qwen25-7b-gsm8k-2gpu", + "condition": "", + "configuration_id": "est-qwen25-7b-gsm8k-c3-2gpu", + "configuration_level": "c3", + "device_capacity_mib": "40960", + "free_cache_engine": "True", + "gpu_count": "2", + "gpu_monitor_interval_ms": "100", + "legacy_dataset_id": "gsm8k", + "margin_limit_mib": "38912", + "max_actor_ckpt_to_keep": "1", + "max_model_len": "3072", + "max_num_seqs": "64", + "max_prompt_length": "1024", + "max_response_length": "2048", + "model_family": "qwen25_7b", + "model_id": "Qwen/Qwen2.5-7B-Instruct", + "model_name": "Qwen2.5-7B-Instruct", + "optimizer_offload": "False", + "parameter_offload": "False", + "peak_measurement": "maximum externally sampled per-device memory", + "ref_logprob_micro_batch": "4", + "resume_mode": "disable", + "rollout_logprob_micro_batch": "4", + "rollout_n": "2", + "rollout_tp_size": "1", + "save_freq": "-1", + "study": "prospective_estimation", + "test_freq": "-1", + "total_training_steps": "1", + "train_batch_size": "32", + "train_max_samples": "32", + "val_before_train": "False", + "val_max_samples": "32", + "vllm_gpu_memory_utilization": "0.7", + "workload_id": "gsm8k", + "workload_name": "GSM8K mathematics" + }, + "state": "memory_failure", + "unresolved_seeds": [] + }, + { + "completed_peak_mib": null, + "completed_processes": 0, + "configuration_id": "est-qwen25-7b-gsm8k-c3-4gpu", + "eligible_processes": 3, + "eligible_seeds": [ + 151, + 152, + 153 + ], + "known_margin_exceedance": false, + "known_memory_failure": true, + "maximum_known_completed_peak_mib": null, + "planned_seeds": [ + 151, + 152, + 153 + ], + "settings": { + "accelerator": "NVIDIA A100-SXM4-40GB", + "actor_micro_batch": "16", + "adaptation": "LoRA", + "algorithm": "GRPO", + "allocator_trace_enabled": "True", + "allocator_trace_sync": "1", + "case_id": "est-qwen25-7b-gsm8k-4gpu", + "condition": "", + "configuration_id": "est-qwen25-7b-gsm8k-c3-4gpu", + "configuration_level": "c3", + "device_capacity_mib": "40960", + "free_cache_engine": "True", + "gpu_count": "4", + "gpu_monitor_interval_ms": "100", + "legacy_dataset_id": "gsm8k", + "margin_limit_mib": "38912", + "max_actor_ckpt_to_keep": "1", + "max_model_len": "3072", + "max_num_seqs": "64", + "max_prompt_length": "1024", + "max_response_length": "2048", + "model_family": "qwen25_7b", + "model_id": "Qwen/Qwen2.5-7B-Instruct", + "model_name": "Qwen2.5-7B-Instruct", + "optimizer_offload": "False", + "parameter_offload": "False", + "peak_measurement": "maximum externally sampled per-device memory", + "ref_logprob_micro_batch": "4", + "resume_mode": "disable", + "rollout_logprob_micro_batch": "4", + "rollout_n": "2", + "rollout_tp_size": "1", + "save_freq": "-1", + "study": "prospective_estimation", + "test_freq": "-1", + "total_training_steps": "1", + "train_batch_size": "32", + "train_max_samples": "32", + "val_before_train": "False", + "val_max_samples": "32", + "vllm_gpu_memory_utilization": "0.7", + "workload_id": "gsm8k", + "workload_name": "GSM8K mathematics" + }, + "state": "memory_failure", + "unresolved_seeds": [] + }, + { + "completed_peak_mib": null, + "completed_processes": 0, + "configuration_id": "est-qwen25-7b-math-c2-2gpu", + "eligible_processes": 1, + "eligible_seeds": [ + 152 + ], + "known_margin_exceedance": false, + "known_memory_failure": true, + "maximum_known_completed_peak_mib": null, + "planned_seeds": [ + 151, + 152, + 153 + ], + "settings": { + "accelerator": "NVIDIA A100-SXM4-40GB", + "actor_micro_batch": "8", + "adaptation": "LoRA", + "algorithm": "GRPO", + "allocator_trace_enabled": "True", + "allocator_trace_sync": "1", + "case_id": "est-qwen25-7b-math-2gpu", + "condition": "", + "configuration_id": "est-qwen25-7b-math-c2-2gpu", + "configuration_level": "c2", + "device_capacity_mib": "40960", + "free_cache_engine": "True", + "gpu_count": "2", + "gpu_monitor_interval_ms": "100", + "legacy_dataset_id": "math", + "margin_limit_mib": "38912", + "max_actor_ckpt_to_keep": "1", + "max_model_len": "1024", + "max_num_seqs": "128", + "max_prompt_length": "512", + "max_response_length": "512", + "model_family": "qwen25_7b", + "model_id": "Qwen/Qwen2.5-7B-Instruct", + "model_name": "Qwen2.5-7B-Instruct", + "optimizer_offload": "False", + "parameter_offload": "False", + "peak_measurement": "maximum externally sampled per-device memory", + "ref_logprob_micro_batch": "8", + "resume_mode": "disable", + "rollout_logprob_micro_batch": "8", + "rollout_n": "2", + "rollout_tp_size": "1", + "save_freq": "-1", + "study": "prospective_estimation", + "test_freq": "-1", + "total_training_steps": "1", + "train_batch_size": "64", + "train_max_samples": "64", + "val_before_train": "False", + "val_max_samples": "64", + "vllm_gpu_memory_utilization": "0.6", + "workload_id": "math", + "workload_name": "MATH mathematics" + }, + "state": "unresolved", + "unresolved_seeds": [ + 151, + 153 + ] + }, + { + "completed_peak_mib": null, + "completed_processes": 0, + "configuration_id": "est-qwen25-7b-math-c2-4gpu", + "eligible_processes": 0, + "eligible_seeds": [], + "known_margin_exceedance": false, + "known_memory_failure": false, + "maximum_known_completed_peak_mib": null, + "planned_seeds": [ + 151, + 152, + 153 + ], + "settings": { + "accelerator": "NVIDIA A100-SXM4-40GB", + "actor_micro_batch": "8", + "adaptation": "LoRA", + "algorithm": "GRPO", + "allocator_trace_enabled": "True", + "allocator_trace_sync": "1", + "case_id": "est-qwen25-7b-math-4gpu", + "condition": "", + "configuration_id": "est-qwen25-7b-math-c2-4gpu", + "configuration_level": "c2", + "device_capacity_mib": "40960", + "free_cache_engine": "True", + "gpu_count": "4", + "gpu_monitor_interval_ms": "100", + "legacy_dataset_id": "math", + "margin_limit_mib": "38912", + "max_actor_ckpt_to_keep": "1", + "max_model_len": "1024", + "max_num_seqs": "128", + "max_prompt_length": "512", + "max_response_length": "512", + "model_family": "qwen25_7b", + "model_id": "Qwen/Qwen2.5-7B-Instruct", + "model_name": "Qwen2.5-7B-Instruct", + "optimizer_offload": "False", + "parameter_offload": "False", + "peak_measurement": "maximum externally sampled per-device memory", + "ref_logprob_micro_batch": "8", + "resume_mode": "disable", + "rollout_logprob_micro_batch": "8", + "rollout_n": "2", + "rollout_tp_size": "1", + "save_freq": "-1", + "study": "prospective_estimation", + "test_freq": "-1", + "total_training_steps": "1", + "train_batch_size": "64", + "train_max_samples": "64", + "val_before_train": "False", + "val_max_samples": "64", + "vllm_gpu_memory_utilization": "0.6", + "workload_id": "math", + "workload_name": "MATH mathematics" + }, + "state": "unresolved", + "unresolved_seeds": [ + 151, + 152, + 153 + ] + }, + { + "completed_peak_mib": null, + "completed_processes": 0, + "configuration_id": "est-qwen25-7b-math-c3-2gpu", + "eligible_processes": 2, + "eligible_seeds": [ + 151, + 153 + ], + "known_margin_exceedance": false, + "known_memory_failure": true, + "maximum_known_completed_peak_mib": null, + "planned_seeds": [ + 151, + 152, + 153 + ], + "settings": { + "accelerator": "NVIDIA A100-SXM4-40GB", + "actor_micro_batch": "16", + "adaptation": "LoRA", + "algorithm": "GRPO", + "allocator_trace_enabled": "True", + "allocator_trace_sync": "1", + "case_id": "est-qwen25-7b-math-2gpu", + "condition": "", + "configuration_id": "est-qwen25-7b-math-c3-2gpu", + "configuration_level": "c3", + "device_capacity_mib": "40960", + "free_cache_engine": "True", + "gpu_count": "2", + "gpu_monitor_interval_ms": "100", + "legacy_dataset_id": "math", + "margin_limit_mib": "38912", + "max_actor_ckpt_to_keep": "1", + "max_model_len": "1024", + "max_num_seqs": "128", + "max_prompt_length": "512", + "max_response_length": "512", + "model_family": "qwen25_7b", + "model_id": "Qwen/Qwen2.5-7B-Instruct", + "model_name": "Qwen2.5-7B-Instruct", + "optimizer_offload": "False", + "parameter_offload": "False", + "peak_measurement": "maximum externally sampled per-device memory", + "ref_logprob_micro_batch": "8", + "resume_mode": "disable", + "rollout_logprob_micro_batch": "8", + "rollout_n": "2", + "rollout_tp_size": "1", + "save_freq": "-1", + "study": "prospective_estimation", + "test_freq": "-1", + "total_training_steps": "1", + "train_batch_size": "64", + "train_max_samples": "64", + "val_before_train": "False", + "val_max_samples": "64", + "vllm_gpu_memory_utilization": "0.7", + "workload_id": "math", + "workload_name": "MATH mathematics" + }, + "state": "unresolved", + "unresolved_seeds": [ + 152 + ] + }, + { + "completed_peak_mib": null, + "completed_processes": 0, + "configuration_id": "est-qwen25-7b-math-c3-4gpu", + "eligible_processes": 2, + "eligible_seeds": [ + 151, + 153 + ], + "known_margin_exceedance": false, + "known_memory_failure": true, + "maximum_known_completed_peak_mib": null, + "planned_seeds": [ + 151, + 152, + 153 + ], + "settings": { + "accelerator": "NVIDIA A100-SXM4-40GB", + "actor_micro_batch": "16", + "adaptation": "LoRA", + "algorithm": "GRPO", + "allocator_trace_enabled": "True", + "allocator_trace_sync": "1", + "case_id": "est-qwen25-7b-math-4gpu", + "condition": "", + "configuration_id": "est-qwen25-7b-math-c3-4gpu", + "configuration_level": "c3", + "device_capacity_mib": "40960", + "free_cache_engine": "True", + "gpu_count": "4", + "gpu_monitor_interval_ms": "100", + "legacy_dataset_id": "math", + "margin_limit_mib": "38912", + "max_actor_ckpt_to_keep": "1", + "max_model_len": "1024", + "max_num_seqs": "128", + "max_prompt_length": "512", + "max_response_length": "512", + "model_family": "qwen25_7b", + "model_id": "Qwen/Qwen2.5-7B-Instruct", + "model_name": "Qwen2.5-7B-Instruct", + "optimizer_offload": "False", + "parameter_offload": "False", + "peak_measurement": "maximum externally sampled per-device memory", + "ref_logprob_micro_batch": "8", + "resume_mode": "disable", + "rollout_logprob_micro_batch": "8", + "rollout_n": "2", + "rollout_tp_size": "1", + "save_freq": "-1", + "study": "prospective_estimation", + "test_freq": "-1", + "total_training_steps": "1", + "train_batch_size": "64", + "train_max_samples": "64", + "val_before_train": "False", + "val_max_samples": "64", + "vllm_gpu_memory_utilization": "0.7", + "workload_id": "math", + "workload_name": "MATH mathematics" + }, + "state": "unresolved", + "unresolved_seeds": [ + 152 + ] + }, + { + "completed_peak_mib": null, + "completed_processes": 0, + "configuration_id": "est-qwen25-7b-code_heavy_tail-c2-2gpu", + "eligible_processes": 3, + "eligible_seeds": [ + 151, + 152, + 153 + ], + "known_margin_exceedance": false, + "known_memory_failure": true, + "maximum_known_completed_peak_mib": null, + "planned_seeds": [ + 151, + 152, + 153 + ], + "settings": { + "accelerator": "NVIDIA A100-SXM4-40GB", + "actor_micro_batch": "8", + "adaptation": "LoRA", + "algorithm": "GRPO", + "allocator_trace_enabled": "True", + "allocator_trace_sync": "1", + "case_id": "est-qwen25-7b-code_heavy_tail-2gpu", + "condition": "", + "configuration_id": "est-qwen25-7b-code_heavy_tail-c2-2gpu", + "configuration_level": "c2", + "device_capacity_mib": "40960", + "free_cache_engine": "True", + "gpu_count": "2", + "gpu_monitor_interval_ms": "100", + "legacy_dataset_id": "code_heavy_tail", + "margin_limit_mib": "38912", + "max_actor_ckpt_to_keep": "1", + "max_model_len": "2560", + "max_num_seqs": "64", + "max_prompt_length": "1536", + "max_response_length": "1024", + "model_family": "qwen25_7b", + "model_id": "Qwen/Qwen2.5-7B-Instruct", + "model_name": "Qwen2.5-7B-Instruct", + "optimizer_offload": "False", + "parameter_offload": "False", + "peak_measurement": "maximum externally sampled per-device memory", + "ref_logprob_micro_batch": "4", + "resume_mode": "disable", + "rollout_logprob_micro_batch": "4", + "rollout_n": "2", + "rollout_tp_size": "1", + "save_freq": "-1", + "study": "prospective_estimation", + "test_freq": "-1", + "total_training_steps": "1", + "train_batch_size": "32", + "train_max_samples": "32", + "val_before_train": "False", + "val_max_samples": "32", + "vllm_gpu_memory_utilization": "0.6", + "workload_id": "longer_prompt_code", + "workload_name": "Longer-prompt code" + }, + "state": "memory_failure", + "unresolved_seeds": [] + }, + { + "completed_peak_mib": null, + "completed_processes": 0, + "configuration_id": "est-qwen25-7b-code_heavy_tail-c2-4gpu", + "eligible_processes": 0, + "eligible_seeds": [], + "known_margin_exceedance": false, + "known_memory_failure": false, + "maximum_known_completed_peak_mib": null, + "planned_seeds": [ + 151, + 152, + 153 + ], + "settings": { + "accelerator": "NVIDIA A100-SXM4-40GB", + "actor_micro_batch": "8", + "adaptation": "LoRA", + "algorithm": "GRPO", + "allocator_trace_enabled": "True", + "allocator_trace_sync": "1", + "case_id": "est-qwen25-7b-code_heavy_tail-4gpu", + "condition": "", + "configuration_id": "est-qwen25-7b-code_heavy_tail-c2-4gpu", + "configuration_level": "c2", + "device_capacity_mib": "40960", + "free_cache_engine": "True", + "gpu_count": "4", + "gpu_monitor_interval_ms": "100", + "legacy_dataset_id": "code_heavy_tail", + "margin_limit_mib": "38912", + "max_actor_ckpt_to_keep": "1", + "max_model_len": "2560", + "max_num_seqs": "64", + "max_prompt_length": "1536", + "max_response_length": "1024", + "model_family": "qwen25_7b", + "model_id": "Qwen/Qwen2.5-7B-Instruct", + "model_name": "Qwen2.5-7B-Instruct", + "optimizer_offload": "False", + "parameter_offload": "False", + "peak_measurement": "maximum externally sampled per-device memory", + "ref_logprob_micro_batch": "4", + "resume_mode": "disable", + "rollout_logprob_micro_batch": "4", + "rollout_n": "2", + "rollout_tp_size": "1", + "save_freq": "-1", + "study": "prospective_estimation", + "test_freq": "-1", + "total_training_steps": "1", + "train_batch_size": "32", + "train_max_samples": "32", + "val_before_train": "False", + "val_max_samples": "32", + "vllm_gpu_memory_utilization": "0.6", + "workload_id": "longer_prompt_code", + "workload_name": "Longer-prompt code" + }, + "state": "unresolved", + "unresolved_seeds": [ + 151, + 152, + 153 + ] + }, + { + "completed_peak_mib": null, + "completed_processes": 0, + "configuration_id": "est-qwen25-7b-code_heavy_tail-c3-2gpu", + "eligible_processes": 3, + "eligible_seeds": [ + 151, + 152, + 153 + ], + "known_margin_exceedance": false, + "known_memory_failure": true, + "maximum_known_completed_peak_mib": null, + "planned_seeds": [ + 151, + 152, + 153 + ], + "settings": { + "accelerator": "NVIDIA A100-SXM4-40GB", + "actor_micro_batch": "16", + "adaptation": "LoRA", + "algorithm": "GRPO", + "allocator_trace_enabled": "True", + "allocator_trace_sync": "1", + "case_id": "est-qwen25-7b-code_heavy_tail-2gpu", + "condition": "", + "configuration_id": "est-qwen25-7b-code_heavy_tail-c3-2gpu", + "configuration_level": "c3", + "device_capacity_mib": "40960", + "free_cache_engine": "True", + "gpu_count": "2", + "gpu_monitor_interval_ms": "100", + "legacy_dataset_id": "code_heavy_tail", + "margin_limit_mib": "38912", + "max_actor_ckpt_to_keep": "1", + "max_model_len": "2560", + "max_num_seqs": "64", + "max_prompt_length": "1536", + "max_response_length": "1024", + "model_family": "qwen25_7b", + "model_id": "Qwen/Qwen2.5-7B-Instruct", + "model_name": "Qwen2.5-7B-Instruct", + "optimizer_offload": "False", + "parameter_offload": "False", + "peak_measurement": "maximum externally sampled per-device memory", + "ref_logprob_micro_batch": "4", + "resume_mode": "disable", + "rollout_logprob_micro_batch": "4", + "rollout_n": "2", + "rollout_tp_size": "1", + "save_freq": "-1", + "study": "prospective_estimation", + "test_freq": "-1", + "total_training_steps": "1", + "train_batch_size": "32", + "train_max_samples": "32", + "val_before_train": "False", + "val_max_samples": "32", + "vllm_gpu_memory_utilization": "0.7", + "workload_id": "longer_prompt_code", + "workload_name": "Longer-prompt code" + }, + "state": "memory_failure", + "unresolved_seeds": [] + }, + { + "completed_peak_mib": null, + "completed_processes": 0, + "configuration_id": "est-qwen25-7b-code_heavy_tail-c3-4gpu", + "eligible_processes": 3, + "eligible_seeds": [ + 151, + 152, + 153 + ], + "known_margin_exceedance": false, + "known_memory_failure": true, + "maximum_known_completed_peak_mib": null, + "planned_seeds": [ + 151, + 152, + 153 + ], + "settings": { + "accelerator": "NVIDIA A100-SXM4-40GB", + "actor_micro_batch": "16", + "adaptation": "LoRA", + "algorithm": "GRPO", + "allocator_trace_enabled": "True", + "allocator_trace_sync": "1", + "case_id": "est-qwen25-7b-code_heavy_tail-4gpu", + "condition": "", + "configuration_id": "est-qwen25-7b-code_heavy_tail-c3-4gpu", + "configuration_level": "c3", + "device_capacity_mib": "40960", + "free_cache_engine": "True", + "gpu_count": "4", + "gpu_monitor_interval_ms": "100", + "legacy_dataset_id": "code_heavy_tail", + "margin_limit_mib": "38912", + "max_actor_ckpt_to_keep": "1", + "max_model_len": "2560", + "max_num_seqs": "64", + "max_prompt_length": "1536", + "max_response_length": "1024", + "model_family": "qwen25_7b", + "model_id": "Qwen/Qwen2.5-7B-Instruct", + "model_name": "Qwen2.5-7B-Instruct", + "optimizer_offload": "False", + "parameter_offload": "False", + "peak_measurement": "maximum externally sampled per-device memory", + "ref_logprob_micro_batch": "4", + "resume_mode": "disable", + "rollout_logprob_micro_batch": "4", + "rollout_n": "2", + "rollout_tp_size": "1", + "save_freq": "-1", + "study": "prospective_estimation", + "test_freq": "-1", + "total_training_steps": "1", + "train_batch_size": "32", + "train_max_samples": "32", + "val_before_train": "False", + "val_max_samples": "32", + "vllm_gpu_memory_utilization": "0.7", + "workload_id": "longer_prompt_code", + "workload_name": "Longer-prompt code" + }, + "state": "memory_failure", + "unresolved_seeds": [] + } +] diff --git a/benchmark/estimation/results/configurations_flat.csv b/benchmark/estimation/results/configurations_flat.csv new file mode 100644 index 0000000000000000000000000000000000000000..24af528a1e7b6b9953a965279f1bc7b666667b44 --- /dev/null +++ b/benchmark/estimation/results/configurations_flat.csv @@ -0,0 +1,13 @@ +all_three_seeds_eligible,completed_peak_mib,completed_processes,component_regression_predicted_peak_mib,component_regression_predicted_state,component_regression_unscored_approval,configuration_id,configuration_level,donor_copy_predicted_peak_mib,donor_copy_predicted_state,donor_copy_unscored_approval,eligible_seed_count,eligible_seeds,gpu_count,known_margin_exceedance,known_memory_failure,logistic_predicted_peak_mib,logistic_predicted_state,logistic_unscored_approval,maximum_known_completed_peak_mib,model,observed_state,required_seeds,unresolved,unresolved_seeds,workload +True,,0,46461.218916058904,memory_failure,False,est-qwen25-7b-gsm8k-c2-2gpu,c2,33108.0,within_margin,False,3,151;152;153,2,False,True,,memory_failure,False,,Qwen/Qwen2.5-7B-Instruct,memory_failure,3,False,,gsm8k +False,,0,38751.70456713468,within_margin,True,est-qwen25-7b-gsm8k-c2-4gpu,c2,29840.0,within_margin,True,0,,4,False,False,,above_margin,False,,Qwen/Qwen2.5-7B-Instruct,unresolved,3,True,151;152;153,gsm8k +True,,0,49851.347837658235,memory_failure,False,est-qwen25-7b-gsm8k-c3-2gpu,c3,37140.0,within_margin,False,3,151;152;153,2,False,True,,memory_failure,False,,Qwen/Qwen2.5-7B-Instruct,memory_failure,3,False,,gsm8k +True,,0,42141.83348873401,memory_failure,False,est-qwen25-7b-gsm8k-c3-4gpu,c3,33872.0,within_margin,False,3,151;152;153,4,False,True,,memory_failure,False,,Qwen/Qwen2.5-7B-Instruct,memory_failure,3,False,,gsm8k +False,,0,46391.75739749115,memory_failure,False,est-qwen25-7b-math-c2-2gpu,c2,33446.0,within_margin,True,1,152,2,False,True,,memory_failure,False,,Qwen/Qwen2.5-7B-Instruct,unresolved,3,True,151;153,math +False,,0,38682.24304856692,within_margin,True,est-qwen25-7b-math-c2-4gpu,c2,30180.0,within_margin,True,0,,4,False,False,,above_margin,False,,Qwen/Qwen2.5-7B-Instruct,unresolved,3,True,151;152;153,math +False,,0,49712.42480052272,memory_failure,False,est-qwen25-7b-math-c3-2gpu,c3,37478.0,within_margin,True,2,151;153,2,False,True,,memory_failure,False,,Qwen/Qwen2.5-7B-Instruct,unresolved,3,True,152,math +False,,0,42002.910451598495,memory_failure,False,est-qwen25-7b-math-c3-4gpu,c3,34212.0,within_margin,True,2,151;153,4,False,True,,memory_failure,False,,Qwen/Qwen2.5-7B-Instruct,unresolved,3,True,152,math +True,,0,46443.853536416966,memory_failure,False,est-qwen25-7b-code_heavy_tail-c2-2gpu,c2,33436.0,within_margin,False,3,151;152;153,2,False,True,,memory_failure,False,,Qwen/Qwen2.5-7B-Instruct,memory_failure,3,False,,code_heavy_tail +False,,0,38734.33918749274,within_margin,True,est-qwen25-7b-code_heavy_tail-c2-4gpu,c2,33436.0,within_margin,True,0,,4,False,False,,above_margin,False,,Qwen/Qwen2.5-7B-Instruct,unresolved,3,True,151;152;153,code_heavy_tail +True,,0,49816.61707837435,memory_failure,False,est-qwen25-7b-code_heavy_tail-c3-2gpu,c3,37682.0,within_margin,False,3,151;152;153,2,False,True,,memory_failure,False,,Qwen/Qwen2.5-7B-Instruct,memory_failure,3,False,,code_heavy_tail +True,,0,42107.102729450125,memory_failure,False,est-qwen25-7b-code_heavy_tail-c3-4gpu,c3,37682.0,within_margin,False,3,151;152;153,4,False,True,,memory_failure,False,,Qwen/Qwen2.5-7B-Instruct,memory_failure,3,False,,code_heavy_tail diff --git a/benchmark/estimation/results/configurations_flat.json b/benchmark/estimation/results/configurations_flat.json new file mode 100644 index 0000000000000000000000000000000000000000..80a56277015645739a00d00b4d34f31a21015c0d --- /dev/null +++ b/benchmark/estimation/results/configurations_flat.json @@ -0,0 +1,338 @@ +[ + { + "all_three_seeds_eligible": true, + "completed_peak_mib": null, + "completed_processes": 0, + "component_regression_predicted_peak_mib": 46461.218916058904, + "component_regression_predicted_state": "memory_failure", + "component_regression_unscored_approval": false, + "configuration_id": "est-qwen25-7b-gsm8k-c2-2gpu", + "configuration_level": "c2", + "donor_copy_predicted_peak_mib": 33108.0, + "donor_copy_predicted_state": "within_margin", + "donor_copy_unscored_approval": false, + "eligible_seed_count": 3, + "eligible_seeds": "151;152;153", + "gpu_count": 2, + "known_margin_exceedance": false, + "known_memory_failure": true, + "logistic_predicted_peak_mib": null, + "logistic_predicted_state": "memory_failure", + "logistic_unscored_approval": false, + "maximum_known_completed_peak_mib": null, + "model": "Qwen/Qwen2.5-7B-Instruct", + "observed_state": "memory_failure", + "required_seeds": 3, + "unresolved": false, + "unresolved_seeds": "", + "workload": "gsm8k" + }, + { + "all_three_seeds_eligible": false, + "completed_peak_mib": null, + "completed_processes": 0, + "component_regression_predicted_peak_mib": 38751.70456713468, + "component_regression_predicted_state": "within_margin", + "component_regression_unscored_approval": true, + "configuration_id": "est-qwen25-7b-gsm8k-c2-4gpu", + "configuration_level": "c2", + "donor_copy_predicted_peak_mib": 29840.0, + "donor_copy_predicted_state": "within_margin", + "donor_copy_unscored_approval": true, + "eligible_seed_count": 0, + "eligible_seeds": "", + "gpu_count": 4, + "known_margin_exceedance": false, + "known_memory_failure": false, + "logistic_predicted_peak_mib": null, + "logistic_predicted_state": "above_margin", + "logistic_unscored_approval": false, + "maximum_known_completed_peak_mib": null, + "model": "Qwen/Qwen2.5-7B-Instruct", + "observed_state": "unresolved", + "required_seeds": 3, + "unresolved": true, + "unresolved_seeds": "151;152;153", + "workload": "gsm8k" + }, + { + "all_three_seeds_eligible": true, + "completed_peak_mib": null, + "completed_processes": 0, + "component_regression_predicted_peak_mib": 49851.347837658235, + "component_regression_predicted_state": "memory_failure", + "component_regression_unscored_approval": false, + "configuration_id": "est-qwen25-7b-gsm8k-c3-2gpu", + "configuration_level": "c3", + "donor_copy_predicted_peak_mib": 37140.0, + "donor_copy_predicted_state": "within_margin", + "donor_copy_unscored_approval": false, + "eligible_seed_count": 3, + "eligible_seeds": "151;152;153", + "gpu_count": 2, + "known_margin_exceedance": false, + "known_memory_failure": true, + "logistic_predicted_peak_mib": null, + "logistic_predicted_state": "memory_failure", + "logistic_unscored_approval": false, + "maximum_known_completed_peak_mib": null, + "model": "Qwen/Qwen2.5-7B-Instruct", + "observed_state": "memory_failure", + "required_seeds": 3, + "unresolved": false, + "unresolved_seeds": "", + "workload": "gsm8k" + }, + { + "all_three_seeds_eligible": true, + "completed_peak_mib": null, + "completed_processes": 0, + "component_regression_predicted_peak_mib": 42141.83348873401, + "component_regression_predicted_state": "memory_failure", + "component_regression_unscored_approval": false, + "configuration_id": "est-qwen25-7b-gsm8k-c3-4gpu", + "configuration_level": "c3", + "donor_copy_predicted_peak_mib": 33872.0, + "donor_copy_predicted_state": "within_margin", + "donor_copy_unscored_approval": false, + "eligible_seed_count": 3, + "eligible_seeds": "151;152;153", + "gpu_count": 4, + "known_margin_exceedance": false, + "known_memory_failure": true, + "logistic_predicted_peak_mib": null, + "logistic_predicted_state": "memory_failure", + "logistic_unscored_approval": false, + "maximum_known_completed_peak_mib": null, + "model": "Qwen/Qwen2.5-7B-Instruct", + "observed_state": "memory_failure", + "required_seeds": 3, + "unresolved": false, + "unresolved_seeds": "", + "workload": "gsm8k" + }, + { + "all_three_seeds_eligible": false, + "completed_peak_mib": null, + "completed_processes": 0, + "component_regression_predicted_peak_mib": 46391.75739749115, + "component_regression_predicted_state": "memory_failure", + "component_regression_unscored_approval": false, + "configuration_id": "est-qwen25-7b-math-c2-2gpu", + "configuration_level": "c2", + "donor_copy_predicted_peak_mib": 33446.0, + "donor_copy_predicted_state": "within_margin", + "donor_copy_unscored_approval": true, + "eligible_seed_count": 1, + "eligible_seeds": "152", + "gpu_count": 2, + "known_margin_exceedance": false, + "known_memory_failure": true, + "logistic_predicted_peak_mib": null, + "logistic_predicted_state": "memory_failure", + "logistic_unscored_approval": false, + "maximum_known_completed_peak_mib": null, + "model": "Qwen/Qwen2.5-7B-Instruct", + "observed_state": "unresolved", + "required_seeds": 3, + "unresolved": true, + "unresolved_seeds": "151;153", + "workload": "math" + }, + { + "all_three_seeds_eligible": false, + "completed_peak_mib": null, + "completed_processes": 0, + "component_regression_predicted_peak_mib": 38682.24304856692, + "component_regression_predicted_state": "within_margin", + "component_regression_unscored_approval": true, + "configuration_id": "est-qwen25-7b-math-c2-4gpu", + "configuration_level": "c2", + "donor_copy_predicted_peak_mib": 30180.0, + "donor_copy_predicted_state": "within_margin", + "donor_copy_unscored_approval": true, + "eligible_seed_count": 0, + "eligible_seeds": "", + "gpu_count": 4, + "known_margin_exceedance": false, + "known_memory_failure": false, + "logistic_predicted_peak_mib": null, + "logistic_predicted_state": "above_margin", + "logistic_unscored_approval": false, + "maximum_known_completed_peak_mib": null, + "model": "Qwen/Qwen2.5-7B-Instruct", + "observed_state": "unresolved", + "required_seeds": 3, + "unresolved": true, + "unresolved_seeds": "151;152;153", + "workload": "math" + }, + { + "all_three_seeds_eligible": false, + "completed_peak_mib": null, + "completed_processes": 0, + "component_regression_predicted_peak_mib": 49712.42480052272, + "component_regression_predicted_state": "memory_failure", + "component_regression_unscored_approval": false, + "configuration_id": "est-qwen25-7b-math-c3-2gpu", + "configuration_level": "c3", + "donor_copy_predicted_peak_mib": 37478.0, + "donor_copy_predicted_state": "within_margin", + "donor_copy_unscored_approval": true, + "eligible_seed_count": 2, + "eligible_seeds": "151;153", + "gpu_count": 2, + "known_margin_exceedance": false, + "known_memory_failure": true, + "logistic_predicted_peak_mib": null, + "logistic_predicted_state": "memory_failure", + "logistic_unscored_approval": false, + "maximum_known_completed_peak_mib": null, + "model": "Qwen/Qwen2.5-7B-Instruct", + "observed_state": "unresolved", + "required_seeds": 3, + "unresolved": true, + "unresolved_seeds": "152", + "workload": "math" + }, + { + "all_three_seeds_eligible": false, + "completed_peak_mib": null, + "completed_processes": 0, + "component_regression_predicted_peak_mib": 42002.910451598495, + "component_regression_predicted_state": "memory_failure", + "component_regression_unscored_approval": false, + "configuration_id": "est-qwen25-7b-math-c3-4gpu", + "configuration_level": "c3", + "donor_copy_predicted_peak_mib": 34212.0, + "donor_copy_predicted_state": "within_margin", + "donor_copy_unscored_approval": true, + "eligible_seed_count": 2, + "eligible_seeds": "151;153", + "gpu_count": 4, + "known_margin_exceedance": false, + "known_memory_failure": true, + "logistic_predicted_peak_mib": null, + "logistic_predicted_state": "memory_failure", + "logistic_unscored_approval": false, + "maximum_known_completed_peak_mib": null, + "model": "Qwen/Qwen2.5-7B-Instruct", + "observed_state": "unresolved", + "required_seeds": 3, + "unresolved": true, + "unresolved_seeds": "152", + "workload": "math" + }, + { + "all_three_seeds_eligible": true, + "completed_peak_mib": null, + "completed_processes": 0, + "component_regression_predicted_peak_mib": 46443.853536416966, + "component_regression_predicted_state": "memory_failure", + "component_regression_unscored_approval": false, + "configuration_id": "est-qwen25-7b-code_heavy_tail-c2-2gpu", + "configuration_level": "c2", + "donor_copy_predicted_peak_mib": 33436.0, + "donor_copy_predicted_state": "within_margin", + "donor_copy_unscored_approval": false, + "eligible_seed_count": 3, + "eligible_seeds": "151;152;153", + "gpu_count": 2, + "known_margin_exceedance": false, + "known_memory_failure": true, + "logistic_predicted_peak_mib": null, + "logistic_predicted_state": "memory_failure", + "logistic_unscored_approval": false, + "maximum_known_completed_peak_mib": null, + "model": "Qwen/Qwen2.5-7B-Instruct", + "observed_state": "memory_failure", + "required_seeds": 3, + "unresolved": false, + "unresolved_seeds": "", + "workload": "code_heavy_tail" + }, + { + "all_three_seeds_eligible": false, + "completed_peak_mib": null, + "completed_processes": 0, + "component_regression_predicted_peak_mib": 38734.33918749274, + "component_regression_predicted_state": "within_margin", + "component_regression_unscored_approval": true, + "configuration_id": "est-qwen25-7b-code_heavy_tail-c2-4gpu", + "configuration_level": "c2", + "donor_copy_predicted_peak_mib": 33436.0, + "donor_copy_predicted_state": "within_margin", + "donor_copy_unscored_approval": true, + "eligible_seed_count": 0, + "eligible_seeds": "", + "gpu_count": 4, + "known_margin_exceedance": false, + "known_memory_failure": false, + "logistic_predicted_peak_mib": null, + "logistic_predicted_state": "above_margin", + "logistic_unscored_approval": false, + "maximum_known_completed_peak_mib": null, + "model": "Qwen/Qwen2.5-7B-Instruct", + "observed_state": "unresolved", + "required_seeds": 3, + "unresolved": true, + "unresolved_seeds": "151;152;153", + "workload": "code_heavy_tail" + }, + { + "all_three_seeds_eligible": true, + "completed_peak_mib": null, + "completed_processes": 0, + "component_regression_predicted_peak_mib": 49816.61707837435, + "component_regression_predicted_state": "memory_failure", + "component_regression_unscored_approval": false, + "configuration_id": "est-qwen25-7b-code_heavy_tail-c3-2gpu", + "configuration_level": "c3", + "donor_copy_predicted_peak_mib": 37682.0, + "donor_copy_predicted_state": "within_margin", + "donor_copy_unscored_approval": false, + "eligible_seed_count": 3, + "eligible_seeds": "151;152;153", + "gpu_count": 2, + "known_margin_exceedance": false, + "known_memory_failure": true, + "logistic_predicted_peak_mib": null, + "logistic_predicted_state": "memory_failure", + "logistic_unscored_approval": false, + "maximum_known_completed_peak_mib": null, + "model": "Qwen/Qwen2.5-7B-Instruct", + "observed_state": "memory_failure", + "required_seeds": 3, + "unresolved": false, + "unresolved_seeds": "", + "workload": "code_heavy_tail" + }, + { + "all_three_seeds_eligible": true, + "completed_peak_mib": null, + "completed_processes": 0, + "component_regression_predicted_peak_mib": 42107.102729450125, + "component_regression_predicted_state": "memory_failure", + "component_regression_unscored_approval": false, + "configuration_id": "est-qwen25-7b-code_heavy_tail-c3-4gpu", + "configuration_level": "c3", + "donor_copy_predicted_peak_mib": 37682.0, + "donor_copy_predicted_state": "within_margin", + "donor_copy_unscored_approval": false, + "eligible_seed_count": 3, + "eligible_seeds": "151;152;153", + "gpu_count": 4, + "known_margin_exceedance": false, + "known_memory_failure": true, + "logistic_predicted_peak_mib": null, + "logistic_predicted_state": "memory_failure", + "logistic_unscored_approval": false, + "maximum_known_completed_peak_mib": null, + "model": "Qwen/Qwen2.5-7B-Instruct", + "observed_state": "memory_failure", + "required_seeds": 3, + "unresolved": false, + "unresolved_seeds": "", + "workload": "code_heavy_tail" + } +] diff --git a/benchmark/estimation/results/costs.json b/benchmark/estimation/results/costs.json new file mode 100644 index 0000000000000000000000000000000000000000..16ac138c91c83696d73e85fa51b9c2fd7e64dcff --- /dev/null +++ b/benchmark/estimation/results/costs.json @@ -0,0 +1,972 @@ +{ + "allocations": [ + { + "accounting_valid": true, + "allocation_gpu_count": 4, + "batch_ended_at_epoch": 1789482735.0, + "batch_minus_outer_end_seconds": 0.0, + "batch_started_at_epoch": 1789482644.0, + "batch_state": "FAILED", + "batch_step_gpu_seconds": 364.0, + "errors": [], + "job_id": "5054339", + "ledger_ended_at_epoch": 1789482735.009265, + "ledger_minus_outer_end_seconds": 0.009264945983886719, + "observed_allocation_gpu_seconds": 364.02373695373535, + "pair_id": "est-qwen25-7b-math-c2-s153", + "pair_ledger_sha256": "8a7b481ea18db0c3bdfae4ab480be6e357c735feaa755a90c0e76031a8de6b7d", + "pair_manifest_sha256": "991df8f86d667cb619f1114341213752936e091f06df8f609442286416781efc", + "raw_record_provenance": { + "batch": { + "line": "5054326_0.batch|5054339.batch|batch|FAILED|2:0|2026-09-15T14:30:44|2026-09-15T14:32:15|91|cpu=64,gres/gpu=4,mem=192G,node=1|acn13|", + "line_number": 2, + "sha256": "2aa42595f73dcef3652717d77c4c029beb0cbb7677417b79f8ce3aed0f87080d" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054326_0|5054339|rlvram-matched-gpu|FAILED|2:0|2026-09-15T14:30:44|2026-09-15T14:32:15|91|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn13|QOSGrpNodeLimit", + "line_number": 1, + "sha256": "3646af02f14eed317ed8f1a9d639a25bf3217ded7f592558bbdd952597c7db40" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_ended_at_epoch": 1789482735.0, + "scheduler_scope": "outer_job_record", + "scheduler_started_at_epoch": 1789482644.0, + "scheduler_state": "FAILED", + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1, + "whole_allocation_gpu_seconds": 364.0 + }, + { + "accounting_valid": true, + "allocation_gpu_count": 4, + "batch_ended_at_epoch": 1789482792.0, + "batch_minus_outer_end_seconds": 3.0, + "batch_started_at_epoch": 1789482736.0, + "batch_state": "FAILED", + "batch_step_gpu_seconds": 224.0, + "errors": [], + "job_id": "5054345", + "ledger_ended_at_epoch": 1789482792.4472399, + "ledger_minus_outer_end_seconds": 3.447239875793457, + "observed_allocation_gpu_seconds": 225.77890491485596, + "pair_id": "est-qwen25-7b-math-c2-s151", + "pair_ledger_sha256": "3dff67c7e3f4c24f35883caef96992ab323630eab11c2bdfcce2d34ada12d403", + "pair_manifest_sha256": "3352f7e0ba5d4ae3c78325713c59a571f6c1b0a46d936e6975a437fc46565feb", + "raw_record_provenance": { + "batch": { + "line": "5054326_1.batch|5054345.batch|batch|FAILED|2:0|2026-09-15T14:32:16|2026-09-15T14:33:12|56|cpu=64,gres/gpu=4,mem=192G,node=1|acn13|", + "line_number": 5, + "sha256": "2e1f5255daa069455f9095ec95a1481549690bb185f5afa1b928747ff6b7346e" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054326_1|5054345|rlvram-matched-gpu|CANCELLED by 7547|0:0|2026-09-15T14:32:16|2026-09-15T14:33:09|53|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn13|QOSGrpNodeLimit", + "line_number": 4, + "sha256": "4c85eb09645ec15c0d8b4442d86382c262a400d1e079ff78f2eee2a643a2e3b1" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_ended_at_epoch": 1789482789.0, + "scheduler_scope": "outer_job_record", + "scheduler_started_at_epoch": 1789482736.0, + "scheduler_state": "CANCELLED by 7547", + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1, + "whole_allocation_gpu_seconds": 212.0 + }, + { + "accounting_valid": true, + "allocation_gpu_count": 4, + "batch_ended_at_epoch": 1789487646.0, + "batch_minus_outer_end_seconds": 0.0, + "batch_started_at_epoch": 1789487331.0, + "batch_state": "COMPLETED", + "batch_step_gpu_seconds": 1260.0, + "errors": [], + "job_id": "5054952", + "ledger_ended_at_epoch": 1789487646.3902702, + "ledger_minus_outer_end_seconds": 0.3902702331542969, + "observed_allocation_gpu_seconds": 1261.5456819534302, + "pair_id": "est-qwen25-7b-gsm8k-c3-s153", + "pair_ledger_sha256": "0ab7e45cad162bfab6e8e6ae2c77bcbdf29c32fa102f3a39d2712159556b48cb", + "pair_manifest_sha256": "e2db8b1be2bf8cc717f5d9cecc6493baab989bc84fd92e2465486dfe15bc0b49", + "raw_record_provenance": { + "batch": { + "line": "5054951_2.batch|5054952.batch|batch|COMPLETED|0:0|2026-09-15T15:48:51|2026-09-15T15:54:06|315|cpu=64,gres/gpu=4,mem=192G,node=1|acn11|", + "line_number": 8, + "sha256": "086912b35626b5a7873378453a54b867633b963560087ed9813a949e9eb7b6fb" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_2|5054952|rlvram-matched-gpu|COMPLETED|0:0|2026-09-15T15:48:51|2026-09-15T15:54:06|315|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn11|None", + "line_number": 7, + "sha256": "f8dcda614ec3ad7343285e8f43a9a2859010369c911014c305d88c3584fc3007" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_ended_at_epoch": 1789487646.0, + "scheduler_scope": "outer_job_record", + "scheduler_started_at_epoch": 1789487331.0, + "scheduler_state": "COMPLETED", + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1, + "whole_allocation_gpu_seconds": 1260.0 + }, + { + "accounting_valid": true, + "allocation_gpu_count": 4, + "batch_ended_at_epoch": 1789487640.0, + "batch_minus_outer_end_seconds": 0.0, + "batch_started_at_epoch": 1789487332.0, + "batch_state": "COMPLETED", + "batch_step_gpu_seconds": 1232.0, + "errors": [], + "job_id": "5054953", + "ledger_ended_at_epoch": 1789487640.8308427, + "ledger_minus_outer_end_seconds": 0.8308427333831787, + "observed_allocation_gpu_seconds": 1235.308614730835, + "pair_id": "est-qwen25-7b-math-c3-s151", + "pair_ledger_sha256": "933314fb69c761afd7901fcca99b7b7ca86a2ff64e93d4cafd39a4c3c02712ac", + "pair_manifest_sha256": "56a9e2bf77364a2c7548a806774042887622e49fb384cc9ae0d540ff95105c2c", + "raw_record_provenance": { + "batch": { + "line": "5054951_3.batch|5054953.batch|batch|COMPLETED|0:0|2026-09-15T15:48:52|2026-09-15T15:54:00|308|cpu=64,gres/gpu=4,mem=192G,node=1|acn15|", + "line_number": 11, + "sha256": "dc462d83f18c0a750050b3af5addcfabbdbe6e28fafe57b7b59747a2b0833a99" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_3|5054953|rlvram-matched-gpu|COMPLETED|0:0|2026-09-15T15:48:52|2026-09-15T15:54:00|308|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn15|None", + "line_number": 10, + "sha256": "e46baf5eb652553f9ea544d7d76077d609cd8a49e596da79bbea7731e5e15cbb" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_ended_at_epoch": 1789487640.0, + "scheduler_scope": "outer_job_record", + "scheduler_started_at_epoch": 1789487332.0, + "scheduler_state": "COMPLETED", + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1, + "whole_allocation_gpu_seconds": 1232.0 + }, + { + "accounting_valid": true, + "allocation_gpu_count": 4, + "batch_ended_at_epoch": 1789487853.0, + "batch_minus_outer_end_seconds": 0.0, + "batch_started_at_epoch": 1789487332.0, + "batch_state": "OUT_OF_MEMORY", + "batch_step_gpu_seconds": 2084.0, + "errors": [], + "job_id": "5054954", + "ledger_ended_at_epoch": 1789487853.0304253, + "ledger_minus_outer_end_seconds": 0.030425310134887695, + "observed_allocation_gpu_seconds": 2084.1130266189575, + "pair_id": "est-qwen25-7b-gsm8k-c2-s152", + "pair_ledger_sha256": "58395abc9d5df303ef2155cf53231c86275c1ee75370b7ac083497b225f997dc", + "pair_manifest_sha256": "06f240c40980feac67a620ecb986328a9c43f805ee3dbe35e414ea149376a59e", + "raw_record_provenance": { + "batch": { + "line": "5054951_4.batch|5054954.batch|batch|OUT_OF_MEMORY|0:125|2026-09-15T15:48:52|2026-09-15T15:57:33|521|cpu=64,gres/gpu=4,mem=192G,node=1|acn18|", + "line_number": 14, + "sha256": "632f5238a72780f5d869b5d1485bdf273f8974a60492b27e2378a04c864837ff" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_4|5054954|rlvram-matched-gpu|OUT_OF_MEMORY|0:125|2026-09-15T15:48:52|2026-09-15T15:57:33|521|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn18|None", + "line_number": 13, + "sha256": "f62e9bca018d972afec1b476f61e1166290a8623f81f87ec793ceb01a20f9b39" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_ended_at_epoch": 1789487853.0, + "scheduler_scope": "outer_job_record", + "scheduler_started_at_epoch": 1789487332.0, + "scheduler_state": "OUT_OF_MEMORY", + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1, + "whole_allocation_gpu_seconds": 2084.0 + }, + { + "accounting_valid": true, + "allocation_gpu_count": 4, + "batch_ended_at_epoch": 1789487922.0, + "batch_minus_outer_end_seconds": 0.0, + "batch_started_at_epoch": 1789487642.0, + "batch_state": "COMPLETED", + "batch_step_gpu_seconds": 1120.0, + "errors": [], + "job_id": "5054977", + "ledger_ended_at_epoch": 1789487922.6665664, + "ledger_minus_outer_end_seconds": 0.6665663719177246, + "observed_allocation_gpu_seconds": 1122.6565771102905, + "pair_id": "est-qwen25-7b-code_heavy_tail-c3-s153", + "pair_ledger_sha256": "8cb922fc5199e4161a37bbb447ce56f022ff98268ca967d7adefc6d6a4fd5771", + "pair_manifest_sha256": "ec8e4a2106cd025faadd7c787be1326212b53334a8aad98ef84dd624dd9eea7b", + "raw_record_provenance": { + "batch": { + "line": "5054951_5.batch|5054977.batch|batch|COMPLETED|0:0|2026-09-15T15:54:02|2026-09-15T15:58:42|280|cpu=64,gres/gpu=4,mem=192G,node=1|acn20|", + "line_number": 17, + "sha256": "36a617a83b903953751df45ac7dd7f474a24fb59af5de346532038024a93b289" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_5|5054977|rlvram-matched-gpu|COMPLETED|0:0|2026-09-15T15:54:02|2026-09-15T15:58:42|280|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn20|JobArrayTaskLimit", + "line_number": 16, + "sha256": "ab8f8a9de2e57d569fbac28dc6f7bbabf40ac3eca3dcfe6f917c278362564e21" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_ended_at_epoch": 1789487922.0, + "scheduler_scope": "outer_job_record", + "scheduler_started_at_epoch": 1789487642.0, + "scheduler_state": "COMPLETED", + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1, + "whole_allocation_gpu_seconds": 1120.0 + }, + { + "accounting_valid": true, + "allocation_gpu_count": 4, + "batch_ended_at_epoch": 1789488155.0, + "batch_minus_outer_end_seconds": 0.0, + "batch_started_at_epoch": 1789487647.0, + "batch_state": "OUT_OF_MEMORY", + "batch_step_gpu_seconds": 2032.0, + "errors": [], + "job_id": "5054981", + "ledger_ended_at_epoch": 1789488155.6469514, + "ledger_minus_outer_end_seconds": 0.64695143699646, + "observed_allocation_gpu_seconds": 2034.5588026046753, + "pair_id": "est-qwen25-7b-gsm8k-c2-s151", + "pair_ledger_sha256": "ec8fab57145124ce0f69f1235a1021166e930084acdfb22ad6d67c48fd712354", + "pair_manifest_sha256": "223d0ca5805a06be1ba04808797cec4b35d1a0f4d2fde64bba8e75662da8719c", + "raw_record_provenance": { + "batch": { + "line": "5054951_6.batch|5054981.batch|batch|OUT_OF_MEMORY|0:125|2026-09-15T15:54:07|2026-09-15T16:02:35|508|cpu=64,gres/gpu=4,mem=192G,node=1|acn11|", + "line_number": 20, + "sha256": "3f616d0a90d5fbf5867a609035af17867a41f4df4bca8364a4a5df9dfb3367ad" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_6|5054981|rlvram-matched-gpu|OUT_OF_MEMORY|0:125|2026-09-15T15:54:07|2026-09-15T16:02:35|508|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn11|JobArrayTaskLimit", + "line_number": 19, + "sha256": "7caf085f6a50ef9b911c2dc71b9dcca53e470c381330fe99be6ab80dc1ecd509" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_ended_at_epoch": 1789488155.0, + "scheduler_scope": "outer_job_record", + "scheduler_started_at_epoch": 1789487647.0, + "scheduler_state": "OUT_OF_MEMORY", + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1, + "whole_allocation_gpu_seconds": 2032.0 + }, + { + "accounting_valid": true, + "allocation_gpu_count": 4, + "batch_ended_at_epoch": 1789488307.0, + "batch_minus_outer_end_seconds": 0.0, + "batch_started_at_epoch": 1789487853.0, + "batch_state": "OUT_OF_MEMORY", + "batch_step_gpu_seconds": 1816.0, + "errors": [], + "job_id": "5054998", + "ledger_ended_at_epoch": 1789488307.1036394, + "ledger_minus_outer_end_seconds": 0.10363936424255371, + "observed_allocation_gpu_seconds": 1816.4061088562012, + "pair_id": "est-qwen25-7b-math-c2-s152", + "pair_ledger_sha256": "1064fb987e1b69fe0bf650314ce057fe2a4c8059614c565045d7155346cfbe06", + "pair_manifest_sha256": "2aeea4bb6c09da6918856caa89918ada979d4d0f64691bc1d5597c23d0ed52fd", + "raw_record_provenance": { + "batch": { + "line": "5054951_7.batch|5054998.batch|batch|OUT_OF_MEMORY|0:125|2026-09-15T15:57:33|2026-09-15T16:05:07|454|cpu=64,gres/gpu=4,mem=192G,node=1|acn18|", + "line_number": 23, + "sha256": "f9e075d6b36294e450b8d16b93fb389110ad9b8490feb881da8f51562b6cd300" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_7|5054998|rlvram-matched-gpu|OUT_OF_MEMORY|0:125|2026-09-15T15:57:33|2026-09-15T16:05:07|454|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn18|JobArrayTaskLimit", + "line_number": 22, + "sha256": "be4d8a35394b80fb173b5da704f331dc23999edcf9aee058292062a96b3a7457" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_ended_at_epoch": 1789488307.0, + "scheduler_scope": "outer_job_record", + "scheduler_started_at_epoch": 1789487853.0, + "scheduler_state": "OUT_OF_MEMORY", + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1, + "whole_allocation_gpu_seconds": 1816.0 + }, + { + "accounting_valid": true, + "allocation_gpu_count": 4, + "batch_ended_at_epoch": 1789488196.0, + "batch_minus_outer_end_seconds": 0.0, + "batch_started_at_epoch": 1789487924.0, + "batch_state": "COMPLETED", + "batch_step_gpu_seconds": 1088.0, + "errors": [], + "job_id": "5055003", + "ledger_ended_at_epoch": 1789488196.6952138, + "ledger_minus_outer_end_seconds": 0.695213794708252, + "observed_allocation_gpu_seconds": 1090.770034790039, + "pair_id": "est-qwen25-7b-code_heavy_tail-c3-s151", + "pair_ledger_sha256": "158c05ad2355beccb78137a62d51614f4d958c7f9bb4c9bfc87ab8ecbe770e7d", + "pair_manifest_sha256": "a4fcb78a9489edfbcfa33200c5d7e9f27b030f982c3e9a28f33ca963b42d44e8", + "raw_record_provenance": { + "batch": { + "line": "5054951_8.batch|5055003.batch|batch|COMPLETED|0:0|2026-09-15T15:58:44|2026-09-15T16:03:16|272|cpu=64,gres/gpu=4,mem=192G,node=1|acn20|", + "line_number": 26, + "sha256": "62b2c4330d6284d577a0d6454951ba19d9e0ed218ce55c3cda10d42a97ee2b9b" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_8|5055003|rlvram-matched-gpu|COMPLETED|0:0|2026-09-15T15:58:44|2026-09-15T16:03:16|272|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn20|JobArrayTaskLimit", + "line_number": 25, + "sha256": "0db977269694185565b2411cd444122e886e96a4005e12e7454920c63fda0ed3" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_ended_at_epoch": 1789488196.0, + "scheduler_scope": "outer_job_record", + "scheduler_started_at_epoch": 1789487924.0, + "scheduler_state": "COMPLETED", + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1, + "whole_allocation_gpu_seconds": 1088.0 + }, + { + "accounting_valid": true, + "allocation_gpu_count": 4, + "batch_ended_at_epoch": 1789488585.0, + "batch_minus_outer_end_seconds": 0.0, + "batch_started_at_epoch": 1789488156.0, + "batch_state": "OUT_OF_MEMORY", + "batch_step_gpu_seconds": 1716.0, + "errors": [], + "job_id": "5055021", + "ledger_ended_at_epoch": 1789488585.6639504, + "ledger_minus_outer_end_seconds": 0.6639504432678223, + "observed_allocation_gpu_seconds": 1718.618740081787, + "pair_id": "est-qwen25-7b-code_heavy_tail-c2-s151", + "pair_ledger_sha256": "677ebb50042b1ef16e7bd82768cee7a2d1b0266b54e3c9794f113c7de86ace1d", + "pair_manifest_sha256": "6395afe03e9cd59953f22397461c744de2d05916ffb204a554841f8f90503496", + "raw_record_provenance": { + "batch": { + "line": "5054951_9.batch|5055021.batch|batch|OUT_OF_MEMORY|0:125|2026-09-15T16:02:36|2026-09-15T16:09:45|429|cpu=64,gres/gpu=4,mem=192G,node=1|acn11|", + "line_number": 29, + "sha256": "ed2545b6c4749a652bf9aed3cda77417e3acbf5ea51d044e1075b5ca87a4077e" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_9|5055021|rlvram-matched-gpu|OUT_OF_MEMORY|0:125|2026-09-15T16:02:36|2026-09-15T16:09:45|429|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn11|JobArrayTaskLimit", + "line_number": 28, + "sha256": "62683b8b356c3c2c7eb5adf12e3a482158a401bd57994543641da7a24e109ee3" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_ended_at_epoch": 1789488585.0, + "scheduler_scope": "outer_job_record", + "scheduler_started_at_epoch": 1789488156.0, + "scheduler_state": "OUT_OF_MEMORY", + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1, + "whole_allocation_gpu_seconds": 1716.0 + }, + { + "accounting_valid": true, + "allocation_gpu_count": 4, + "batch_ended_at_epoch": 1789488623.0, + "batch_minus_outer_end_seconds": 0.0, + "batch_started_at_epoch": 1789488197.0, + "batch_state": "OUT_OF_MEMORY", + "batch_step_gpu_seconds": 1704.0, + "errors": [], + "job_id": "5055027", + "ledger_ended_at_epoch": 1789488623.825692, + "ledger_minus_outer_end_seconds": 0.8256919384002686, + "observed_allocation_gpu_seconds": 1707.2920722961426, + "pair_id": "est-qwen25-7b-gsm8k-c2-s153", + "pair_ledger_sha256": "e3fecbf6c338a0d390915e18757a9b0ac12973f79d191f60925cab0e6a67c367", + "pair_manifest_sha256": "ea390f752b0726d01796ec9bd9cdbc7b95ba2e214c76a72933f5e840914630d6", + "raw_record_provenance": { + "batch": { + "line": "5054951_10.batch|5055027.batch|batch|OUT_OF_MEMORY|0:125|2026-09-15T16:03:17|2026-09-15T16:10:23|426|cpu=64,gres/gpu=4,mem=192G,node=1|acn20|", + "line_number": 32, + "sha256": "26ff8416d05c1f67fdbd639bfa1bf33b70049276ca4f6c64930f52a729a251b1" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_10|5055027|rlvram-matched-gpu|OUT_OF_MEMORY|0:125|2026-09-15T16:03:17|2026-09-15T16:10:23|426|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn20|JobArrayTaskLimit", + "line_number": 31, + "sha256": "1824818a792ed14ef6b19ac9e6277e656848429f2e330223e24ad101e0bab8e0" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_ended_at_epoch": 1789488623.0, + "scheduler_scope": "outer_job_record", + "scheduler_started_at_epoch": 1789488197.0, + "scheduler_state": "OUT_OF_MEMORY", + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1, + "whole_allocation_gpu_seconds": 1704.0 + }, + { + "accounting_valid": true, + "allocation_gpu_count": 4, + "batch_ended_at_epoch": 1789488613.0, + "batch_minus_outer_end_seconds": 0.0, + "batch_started_at_epoch": 1789488309.0, + "batch_state": "COMPLETED", + "batch_step_gpu_seconds": 1216.0, + "errors": [], + "job_id": "5055037", + "ledger_ended_at_epoch": 1789488612.9040904, + "ledger_minus_outer_end_seconds": -0.09590959548950195, + "observed_allocation_gpu_seconds": 1215.6075954437256, + "pair_id": "est-qwen25-7b-code_heavy_tail-c3-s152", + "pair_ledger_sha256": "cb53fc6413afbaa786f81fe6a3e57cfe1b07e57d577b6beb5f0904935ebade22", + "pair_manifest_sha256": "7c27305530e8fe45b8afd153ce30cfe047562310d8ac0ae5ee4583924a7c85c6", + "raw_record_provenance": { + "batch": { + "line": "5054951_11.batch|5055037.batch|batch|COMPLETED|0:0|2026-09-15T16:05:09|2026-09-15T16:10:13|304|cpu=64,gres/gpu=4,mem=192G,node=1|acn18|", + "line_number": 35, + "sha256": "87f9c86860d88d023bf6b21fab9782dfb365502d6df6460bece1d3c034e9f8e1" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_11|5055037|rlvram-matched-gpu|COMPLETED|0:0|2026-09-15T16:05:09|2026-09-15T16:10:13|304|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn18|JobArrayTaskLimit", + "line_number": 34, + "sha256": "40a709e26f393883dee8624152f24f9d5633504cbb21392f8df02e4c80013347" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_ended_at_epoch": 1789488613.0, + "scheduler_scope": "outer_job_record", + "scheduler_started_at_epoch": 1789488309.0, + "scheduler_state": "COMPLETED", + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1, + "whole_allocation_gpu_seconds": 1216.0 + }, + { + "accounting_valid": true, + "allocation_gpu_count": 4, + "batch_ended_at_epoch": 1789489147.0, + "batch_minus_outer_end_seconds": 0.0, + "batch_started_at_epoch": 1789488586.0, + "batch_state": "OUT_OF_MEMORY", + "batch_step_gpu_seconds": 2244.0, + "errors": [], + "job_id": "5055038", + "ledger_ended_at_epoch": 1789489146.8906772, + "ledger_minus_outer_end_seconds": -0.10932278633117676, + "observed_allocation_gpu_seconds": 2243.552254676819, + "pair_id": "est-qwen25-7b-code_heavy_tail-c2-s152", + "pair_ledger_sha256": "b9f49fb1c9f629d7d00cc59e9558b0bcc1ab86462a8ecc48b357cc36086b0564", + "pair_manifest_sha256": "e537f9a68f04d79dfc3db97a988dd61656824cab1d1b99587a263c448b225dd6", + "raw_record_provenance": { + "batch": { + "line": "5054951_12.batch|5055038.batch|batch|OUT_OF_MEMORY|0:125|2026-09-15T16:09:46|2026-09-15T16:19:07|561|cpu=64,gres/gpu=4,mem=192G,node=1|acn01|", + "line_number": 38, + "sha256": "9e932163e94f3699d0ba7c3354310c866704ac2fb2f0f8da04d43862a9575a27" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_12|5055038|rlvram-matched-gpu|OUT_OF_MEMORY|0:125|2026-09-15T16:09:46|2026-09-15T16:19:07|561|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn01|JobArrayTaskLimit", + "line_number": 37, + "sha256": "b5366a5c233c938ec7d8edc0d9d53e281bf30003247fe3a175c4f1f1dbd41cba" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_ended_at_epoch": 1789489147.0, + "scheduler_scope": "outer_job_record", + "scheduler_started_at_epoch": 1789488586.0, + "scheduler_state": "OUT_OF_MEMORY", + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1, + "whole_allocation_gpu_seconds": 2244.0 + }, + { + "accounting_valid": true, + "allocation_gpu_count": 4, + "batch_ended_at_epoch": 1789488898.0, + "batch_minus_outer_end_seconds": 0.0, + "batch_started_at_epoch": 1789488613.0, + "batch_state": "COMPLETED", + "batch_step_gpu_seconds": 1140.0, + "errors": [], + "job_id": "5055039", + "ledger_ended_at_epoch": 1789488898.7489774, + "ledger_minus_outer_end_seconds": 0.7489774227142334, + "observed_allocation_gpu_seconds": 1142.9871644973755, + "pair_id": "est-qwen25-7b-gsm8k-c3-s151", + "pair_ledger_sha256": "1974529d05d8aa3adf94448d49f6178553bb12da672dd9ffbe8f9d1f283f06b8", + "pair_manifest_sha256": "019720c3288d38821c459ca30347a1829315b6cbfc8a267aa2bd35f81fa32b92", + "raw_record_provenance": { + "batch": { + "line": "5054951_13.batch|5055039.batch|batch|COMPLETED|0:0|2026-09-15T16:10:13|2026-09-15T16:14:58|285|cpu=64,gres/gpu=4,mem=192G,node=1|acn08|", + "line_number": 41, + "sha256": "d974f4dfebc7b879f43cebd521c472ace1eff36724bfac10a2e127475d9b8e57" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_13|5055039|rlvram-matched-gpu|COMPLETED|0:0|2026-09-15T16:10:13|2026-09-15T16:14:58|285|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn08|JobArrayTaskLimit", + "line_number": 40, + "sha256": "507f420cee2376c7893155ce734b0cad24363f9baae3fbc2f12d7d7a1992edf4" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_ended_at_epoch": 1789488898.0, + "scheduler_scope": "outer_job_record", + "scheduler_started_at_epoch": 1789488613.0, + "scheduler_state": "COMPLETED", + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1, + "whole_allocation_gpu_seconds": 1140.0 + }, + { + "accounting_valid": true, + "allocation_gpu_count": 4, + "batch_ended_at_epoch": 1789488914.0, + "batch_minus_outer_end_seconds": 0.0, + "batch_started_at_epoch": 1789488625.0, + "batch_state": "COMPLETED", + "batch_step_gpu_seconds": 1156.0, + "errors": [], + "job_id": "5055040", + "ledger_ended_at_epoch": 1789488914.1724546, + "ledger_minus_outer_end_seconds": 0.1724545955657959, + "observed_allocation_gpu_seconds": 1156.6805276870728, + "pair_id": "est-qwen25-7b-gsm8k-c3-s152", + "pair_ledger_sha256": "c92968019ad2d4c7b072941de26e1bc6621c4f980c66e27a304e703d1a3e04a5", + "pair_manifest_sha256": "a66599ba72b406d526e0c736ef1daaee8d5d6f844a26eb38f4d3f5df6b57a4dd", + "raw_record_provenance": { + "batch": { + "line": "5054951_14.batch|5055040.batch|batch|COMPLETED|0:0|2026-09-15T16:10:25|2026-09-15T16:15:14|289|cpu=64,gres/gpu=4,mem=192G,node=1|acn11|", + "line_number": 44, + "sha256": "31150809a0e2fef17253e165e2f185b4d1f73b8501d63735667e5e663f8c0a85" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_14|5055040|rlvram-matched-gpu|COMPLETED|0:0|2026-09-15T16:10:25|2026-09-15T16:15:14|289|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn11|JobArrayTaskLimit", + "line_number": 43, + "sha256": "e4d36cac674c275775e0d51dbb3a4edf4acf3dea0caa600d179b618db8cd458f" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_ended_at_epoch": 1789488914.0, + "scheduler_scope": "outer_job_record", + "scheduler_started_at_epoch": 1789488625.0, + "scheduler_state": "COMPLETED", + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1, + "whole_allocation_gpu_seconds": 1156.0 + }, + { + "accounting_valid": true, + "allocation_gpu_count": 4, + "batch_ended_at_epoch": 1789489176.0, + "batch_minus_outer_end_seconds": 0.0, + "batch_started_at_epoch": 1789488900.0, + "batch_state": "COMPLETED", + "batch_step_gpu_seconds": 1104.0, + "errors": [], + "job_id": "5055042", + "ledger_ended_at_epoch": 1789489176.849242, + "ledger_minus_outer_end_seconds": 0.8492419719696045, + "observed_allocation_gpu_seconds": 1107.3879871368408, + "pair_id": "est-qwen25-7b-math-c3-s153", + "pair_ledger_sha256": "f8bdfa5c80e6aaddc5e45ba364a1fec1f08730ec8ca6149077226468af82fb42", + "pair_manifest_sha256": "4cb4bf4a3c5922f305b91dd0985935fde9f33d62c50b707baeb014d517f9c595", + "raw_record_provenance": { + "batch": { + "line": "5054951_15.batch|5055042.batch|batch|COMPLETED|0:0|2026-09-15T16:15:00|2026-09-15T16:19:36|276|cpu=64,gres/gpu=4,mem=192G,node=1|acn08|", + "line_number": 47, + "sha256": "a6b128dbf87b3d4a01bd0d276f5d45b6f671ce64dd1941643fe632ade5b0ba5c" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_15|5055042|rlvram-matched-gpu|COMPLETED|0:0|2026-09-15T16:15:00|2026-09-15T16:19:36|276|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn08|JobArrayTaskLimit", + "line_number": 46, + "sha256": "36930ee3b0567563e4ac913fea86d244146f1e9a54ae4172fa1f5fd6e0aec5f4" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_ended_at_epoch": 1789489176.0, + "scheduler_scope": "outer_job_record", + "scheduler_started_at_epoch": 1789488900.0, + "scheduler_state": "COMPLETED", + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1, + "whole_allocation_gpu_seconds": 1104.0 + }, + { + "accounting_valid": true, + "allocation_gpu_count": 4, + "batch_ended_at_epoch": 1789489357.0, + "batch_minus_outer_end_seconds": 0.0, + "batch_started_at_epoch": 1789488914.0, + "batch_state": "OUT_OF_MEMORY", + "batch_step_gpu_seconds": 1772.0, + "errors": [], + "job_id": "5055043", + "ledger_ended_at_epoch": 1789489357.2469258, + "ledger_minus_outer_end_seconds": 0.24692583084106445, + "observed_allocation_gpu_seconds": 1772.9767618179321, + "pair_id": "est-qwen25-7b-code_heavy_tail-c2-s153", + "pair_ledger_sha256": "265adb55fb58ffe85ab0f4e5e2802e00235c2654fd8429eba9595177d83aaa09", + "pair_manifest_sha256": "14764ab83644f8cd5912cebea75dae9b704ca48ba0b6bc4cb9007fd7751c599b", + "raw_record_provenance": { + "batch": { + "line": "5054951_16.batch|5055043.batch|batch|OUT_OF_MEMORY|0:125|2026-09-15T16:15:14|2026-09-15T16:22:37|443|cpu=64,gres/gpu=4,mem=192G,node=1|acn11|", + "line_number": 50, + "sha256": "3c6922fbdabfdcf51c2da2c5d0b8d478d77d4fb5ed26d1980bb9c62b966925e0" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_16|5055043|rlvram-matched-gpu|OUT_OF_MEMORY|0:125|2026-09-15T16:15:14|2026-09-15T16:22:37|443|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn11|JobArrayTaskLimit", + "line_number": 49, + "sha256": "9312728a73b1e4286376000c297d8618c5a7151fd09cf9862abcdf830cb5fcf6" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_ended_at_epoch": 1789489357.0, + "scheduler_scope": "outer_job_record", + "scheduler_started_at_epoch": 1789488914.0, + "scheduler_state": "OUT_OF_MEMORY", + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1, + "whole_allocation_gpu_seconds": 1772.0 + }, + { + "accounting_valid": true, + "allocation_gpu_count": 4, + "batch_ended_at_epoch": 1789489158.0, + "batch_minus_outer_end_seconds": 0.0, + "batch_started_at_epoch": 1789489149.0, + "batch_state": "FAILED", + "batch_step_gpu_seconds": 36.0, + "errors": [], + "job_id": "5054951", + "ledger_ended_at_epoch": 1789489158.6663845, + "ledger_minus_outer_end_seconds": 0.6663844585418701, + "observed_allocation_gpu_seconds": null, + "pair_id": "est-qwen25-7b-math-c3-s152", + "pair_ledger_sha256": "85f9da6513e1fe16b8dad1e6fec0d54aabe0af20601fa5883a21c81b4a73a06a", + "pair_manifest_sha256": "f4f5d85031ac044d1db31cb535d93ca010ced3f61d7e8abfae2fe2391dffe85e", + "raw_record_provenance": { + "batch": { + "line": "5054951_17.batch|5054951.batch|batch|FAILED|2:0|2026-09-15T16:19:09|2026-09-15T16:19:18|9|cpu=64,gres/gpu=4,mem=192G,node=1|acn01|", + "line_number": 53, + "sha256": "1d5a6d468589d00cc53bcb2843198f3816dc168daad7326a2b96e106b292aa07" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5054951_17|5054951|rlvram-matched-gpu|FAILED|2:0|2026-09-15T16:19:09|2026-09-15T16:19:18|9|billing=4,cpu=64,gres/gpu=4,mem=192G,node=1|acn01|JobArrayTaskLimit", + "line_number": 52, + "sha256": "2e45ccf5289b1149a87417daced55a3bd75c02dab4c40eea82900a32885ff3cb" + }, + "source_sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_ended_at_epoch": 1789489158.0, + "scheduler_scope": "outer_job_record", + "scheduler_started_at_epoch": 1789489149.0, + "scheduler_state": "FAILED", + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1, + "whole_allocation_gpu_seconds": 36.0 + } + ], + "batch_step_time": { + "complete_total_gpu_seconds": 23308.0, + "known_gpu_seconds": 23308.0, + "known_records": 18, + "unknown_records": 0 + }, + "invocation_reservation_time": { + "complete_total_gpu_seconds": 22139.162671485916, + "known_gpu_seconds": 22139.162671485916, + "known_records": 36, + "unknown_records": 0 + }, + "observed_allocation_time": { + "complete_total_gpu_seconds": null, + "known_gpu_seconds": 23300.264592170715, + "known_records": 17, + "unknown_records": 1 + }, + "scheduler_scope_note": "Whole-allocation totals use reported outer Start/End only. Batch-step totals are separate, overlapping spans and must never be added to outer totals. Timestamp precision validates ledger containment but adds no billed seconds.", + "scope": "Task/invocation time includes failed and unresolved work when timing is validated. Observed allocation time ends at runner finalization; whole-job time requires scheduler accounting. The 55-minute limit is not actual elapsed time. Missing records are unknown, not zero cost. No utilization weighting.", + "task_provisioned_time": { + "complete_total_gpu_seconds": 17882.193041578867, + "known_gpu_seconds": 17882.193041578867, + "known_records": 36, + "unknown_records": 0 + }, + "whole_allocation_time": { + "complete_total_gpu_seconds": 23296.0, + "known_gpu_seconds": 23296.0, + "known_records": 18, + "unknown_records": 0 + } +} diff --git a/benchmark/estimation/results/evaluation.json b/benchmark/estimation/results/evaluation.json new file mode 100644 index 0000000000000000000000000000000000000000..5e6ca7ec50e91b8e8477ece5e30198e00d36c34c --- /dev/null +++ b/benchmark/estimation/results/evaluation.json @@ -0,0 +1,216 @@ +{ + "approvals": { + "component_regression": { + "full_panel_approved_configurations": 3, + "full_panel_approved_ids": [ + "est-qwen25-7b-code_heavy_tail-c2-4gpu", + "est-qwen25-7b-gsm8k-c2-4gpu", + "est-qwen25-7b-math-c2-4gpu" + ], + "scored_approved_configurations": 0, + "unscored_approvals_with_known_failure": [], + "unscored_approvals_with_known_margin_exceedance": [], + "unscored_approved_configurations": 3, + "unscored_approved_ids": [ + "est-qwen25-7b-code_heavy_tail-c2-4gpu", + "est-qwen25-7b-gsm8k-c2-4gpu", + "est-qwen25-7b-math-c2-4gpu" + ] + }, + "donor_copy": { + "full_panel_approved_configurations": 12, + "full_panel_approved_ids": [ + "est-qwen25-7b-code_heavy_tail-c2-2gpu", + "est-qwen25-7b-code_heavy_tail-c2-4gpu", + "est-qwen25-7b-code_heavy_tail-c3-2gpu", + "est-qwen25-7b-code_heavy_tail-c3-4gpu", + "est-qwen25-7b-gsm8k-c2-2gpu", + "est-qwen25-7b-gsm8k-c2-4gpu", + "est-qwen25-7b-gsm8k-c3-2gpu", + "est-qwen25-7b-gsm8k-c3-4gpu", + "est-qwen25-7b-math-c2-2gpu", + "est-qwen25-7b-math-c2-4gpu", + "est-qwen25-7b-math-c3-2gpu", + "est-qwen25-7b-math-c3-4gpu" + ], + "scored_approved_configurations": 6, + "unscored_approvals_with_known_failure": [ + "est-qwen25-7b-math-c2-2gpu", + "est-qwen25-7b-math-c3-2gpu", + "est-qwen25-7b-math-c3-4gpu" + ], + "unscored_approvals_with_known_margin_exceedance": [], + "unscored_approved_configurations": 6, + "unscored_approved_ids": [ + "est-qwen25-7b-code_heavy_tail-c2-4gpu", + "est-qwen25-7b-gsm8k-c2-4gpu", + "est-qwen25-7b-math-c2-2gpu", + "est-qwen25-7b-math-c2-4gpu", + "est-qwen25-7b-math-c3-2gpu", + "est-qwen25-7b-math-c3-4gpu" + ] + }, + "logistic": { + "full_panel_approved_configurations": 0, + "full_panel_approved_ids": [], + "scored_approved_configurations": 0, + "unscored_approvals_with_known_failure": [], + "unscored_approvals_with_known_margin_exceedance": [], + "unscored_approved_configurations": 0, + "unscored_approved_ids": [] + } + }, + "full_panel_coverage": 0.5, + "interpretation": "Unresolved configurations are unscored, not correct predictions or safe labels.", + "metric_scope": "resolved_subset_only", + "planned_configurations": 12, + "resolved_configurations": 6, + "resolved_ids": [ + "est-qwen25-7b-code_heavy_tail-c2-2gpu", + "est-qwen25-7b-code_heavy_tail-c3-2gpu", + "est-qwen25-7b-code_heavy_tail-c3-4gpu", + "est-qwen25-7b-gsm8k-c2-2gpu", + "est-qwen25-7b-gsm8k-c3-2gpu", + "est-qwen25-7b-gsm8k-c3-4gpu" + ], + "resolved_subset_metrics": { + "component_regression": { + "above_margin_among_approved": null, + "approval_rate_on_above_margin_targets": null, + "approval_rate_on_memory_failure_targets": 0.0, + "approved": 0, + "approved_above_margin": 0, + "approved_memory_failure": 0, + "approved_within_margin": 0, + "completed_peak_mae_mib": null, + "completed_peak_mean_signed_error_mib": null, + "completed_peak_prediction_coverage": 0, + "completed_peak_worst_underestimate_mib": null, + "confusion_matrix_predicted_rows_observed_columns": { + "above_margin": { + "above_margin": 0, + "memory_failure": 0, + "within_margin": 0 + }, + "memory_failure": { + "above_margin": 0, + "memory_failure": 6, + "within_margin": 0 + }, + "within_margin": { + "above_margin": 0, + "memory_failure": 0, + "within_margin": 0 + } + }, + "correct_three_state": 6, + "distinct_target_configurations": 6, + "memory_failure_among_approved": null, + "queries": 6, + "rejected_within_margin": 0, + "rejection_rate_on_within_margin_targets": null, + "startup_guard_changed_states": 0, + "target_above_margin": 0, + "target_memory_failure": 6, + "target_within_margin": 0, + "three_state_accuracy": 1.0, + "within_margin_approval_precision": null, + "within_margin_recall": null + }, + "donor_copy": { + "above_margin_among_approved": 0.0, + "approval_rate_on_above_margin_targets": null, + "approval_rate_on_memory_failure_targets": 1.0, + "approved": 6, + "approved_above_margin": 0, + "approved_memory_failure": 6, + "approved_within_margin": 0, + "completed_peak_mae_mib": null, + "completed_peak_mean_signed_error_mib": null, + "completed_peak_prediction_coverage": 0, + "completed_peak_worst_underestimate_mib": null, + "confusion_matrix_predicted_rows_observed_columns": { + "above_margin": { + "above_margin": 0, + "memory_failure": 0, + "within_margin": 0 + }, + "memory_failure": { + "above_margin": 0, + "memory_failure": 0, + "within_margin": 0 + }, + "within_margin": { + "above_margin": 0, + "memory_failure": 6, + "within_margin": 0 + } + }, + "correct_three_state": 0, + "distinct_target_configurations": 6, + "memory_failure_among_approved": 1.0, + "queries": 6, + "rejected_within_margin": 0, + "rejection_rate_on_within_margin_targets": null, + "startup_guard_changed_states": 0, + "target_above_margin": 0, + "target_memory_failure": 6, + "target_within_margin": 0, + "three_state_accuracy": 0.0, + "within_margin_approval_precision": 0.0, + "within_margin_recall": null + }, + "logistic": { + "above_margin_among_approved": null, + "approval_rate_on_above_margin_targets": null, + "approval_rate_on_memory_failure_targets": 0.0, + "approved": 0, + "approved_above_margin": 0, + "approved_memory_failure": 0, + "approved_within_margin": 0, + "completed_peak_mae_mib": null, + "completed_peak_mean_signed_error_mib": null, + "completed_peak_prediction_coverage": 0, + "completed_peak_worst_underestimate_mib": null, + "confusion_matrix_predicted_rows_observed_columns": { + "above_margin": { + "above_margin": 0, + "memory_failure": 0, + "within_margin": 0 + }, + "memory_failure": { + "above_margin": 0, + "memory_failure": 6, + "within_margin": 0 + }, + "within_margin": { + "above_margin": 0, + "memory_failure": 0, + "within_margin": 0 + } + }, + "correct_three_state": 6, + "distinct_target_configurations": 6, + "memory_failure_among_approved": null, + "queries": 6, + "rejected_within_margin": 0, + "rejection_rate_on_within_margin_targets": null, + "startup_guard_changed_states": 0, + "target_above_margin": 0, + "target_memory_failure": 6, + "target_within_margin": 0, + "three_state_accuracy": 1.0, + "within_margin_approval_precision": null, + "within_margin_recall": null + } + }, + "unresolved_configurations": 6, + "unresolved_ids": [ + "est-qwen25-7b-code_heavy_tail-c2-4gpu", + "est-qwen25-7b-gsm8k-c2-4gpu", + "est-qwen25-7b-math-c2-2gpu", + "est-qwen25-7b-math-c2-4gpu", + "est-qwen25-7b-math-c3-2gpu", + "est-qwen25-7b-math-c3-4gpu" + ] +} diff --git a/benchmark/estimation/results/output_manifest.json b/benchmark/estimation/results/output_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..49d8f6782c3395e61d5520b13550d3e49eacd91b --- /dev/null +++ b/benchmark/estimation/results/output_manifest.json @@ -0,0 +1,50 @@ +{ + "configurations.csv": { + "sha256": "9d37d28c0204df5d45d4a4f5ffecd8290800004f9d4084996c86c5e1b155a0bd", + "size_bytes": 19245 + }, + "configurations.json": { + "sha256": "afd92cca168dc114138d0551933bb673160c686e3b781503b262a0c8bd03b773", + "size_bytes": 24620 + }, + "configurations_flat.csv": { + "sha256": "18dd5f6ff5a123ae69b9f5ce3ef3c72d05ba51055816a0c49b02a163b6b371e7", + "size_bytes": 3146 + }, + "configurations_flat.json": { + "sha256": "04271cb6baffce9523fbed16f5620bcc184a2b52c9baa18936dfdacb82b09bf8", + "size_bytes": 12494 + }, + "costs.json": { + "sha256": "b3f4e98408d4afac76271fc22815beb7e1699f06c79f0774d360de23d31d1017", + "size_bytes": 40993 + }, + "evaluation.json": { + "sha256": "5a2f2b4677d9d620c1a190edc9d66b61d070d77fef0e09af4eeadf137321ed22", + "size_bytes": 7320 + }, + "pairs.csv": { + "sha256": "040e9d135c0a132debc315d62dd6824b5bd826d42ec85056c722b058995d7cd1", + "size_bytes": 1810 + }, + "pairs.json": { + "sha256": "f1fc7a443ef39e20bdf22ca8bc7429f7fa7a860c364c471d43bdd52e01805165", + "size_bytes": 7197 + }, + "processes.csv": { + "sha256": "9a4e03a34a51a287afb983b93244662f9ac7abf366ba97bd717ba9d371fefdc2", + "size_bytes": 17770 + }, + "processes.json": { + "sha256": "d41a8d9a2a24dc753e242e1e29aa3039171beaed8f61d57639b624f39c6f81d6", + "size_bytes": 47060 + }, + "provenance.json": { + "sha256": "504d883a6ded17d5c56fd7a05e41378305207e5e331cce1b455095fea8621282", + "size_bytes": 160742 + }, + "summary.json": { + "sha256": "ce1fc65b68d51a4b5e5fc8f72e49c0f383694fbc02fadc944c03b4639c741768", + "size_bytes": 321 + } +} diff --git a/benchmark/estimation/results/pairs.csv b/benchmark/estimation/results/pairs.csv new file mode 100644 index 0000000000000000000000000000000000000000..ad5c571848f4514cdc01e475b2ee87d343fdebc7 --- /dev/null +++ b/benchmark/estimation/results/pairs.csv @@ -0,0 +1,19 @@ +both_completed,both_eligible,configuration_level,dataset,first_task_gpu_count,four_gpu_state,pair_id,peak_difference_4_minus_2_mib,runtime_difference_4_minus_2_seconds,training_seed,two_gpu_state +False,False,c2,math,2,unresolved,est-qwen25-7b-math-c2-s153,,,153,unresolved +False,False,c2,math,4,unresolved,est-qwen25-7b-math-c2-s151,,,151,unresolved +False,True,c3,gsm8k,2,memory_failure,est-qwen25-7b-gsm8k-c3-s153,,,153,memory_failure +False,True,c3,math,4,memory_failure,est-qwen25-7b-math-c3-s151,,,151,memory_failure +False,False,c2,gsm8k,2,unresolved,est-qwen25-7b-gsm8k-c2-s152,,,152,memory_failure +False,True,c3,code_heavy_tail,4,memory_failure,est-qwen25-7b-code_heavy_tail-c3-s153,,,153,memory_failure +False,False,c2,gsm8k,2,unresolved,est-qwen25-7b-gsm8k-c2-s151,,,151,memory_failure +False,False,c2,math,2,unresolved,est-qwen25-7b-math-c2-s152,,,152,memory_failure +False,True,c3,code_heavy_tail,4,memory_failure,est-qwen25-7b-code_heavy_tail-c3-s151,,,151,memory_failure +False,False,c2,code_heavy_tail,2,unresolved,est-qwen25-7b-code_heavy_tail-c2-s151,,,151,memory_failure +False,False,c2,gsm8k,4,unresolved,est-qwen25-7b-gsm8k-c2-s153,,,153,memory_failure +False,True,c3,code_heavy_tail,2,memory_failure,est-qwen25-7b-code_heavy_tail-c3-s152,,,152,memory_failure +False,False,c2,code_heavy_tail,2,unresolved,est-qwen25-7b-code_heavy_tail-c2-s152,,,152,memory_failure +False,True,c3,gsm8k,4,memory_failure,est-qwen25-7b-gsm8k-c3-s151,,,151,memory_failure +False,True,c3,gsm8k,4,memory_failure,est-qwen25-7b-gsm8k-c3-s152,,,152,memory_failure +False,True,c3,math,2,memory_failure,est-qwen25-7b-math-c3-s153,,,153,memory_failure +False,False,c2,code_heavy_tail,4,unresolved,est-qwen25-7b-code_heavy_tail-c2-s153,,,153,memory_failure +False,False,c3,math,4,unresolved,est-qwen25-7b-math-c3-s152,,,152,unresolved diff --git a/benchmark/estimation/results/pairs.json b/benchmark/estimation/results/pairs.json new file mode 100644 index 0000000000000000000000000000000000000000..0188cff85d0ae0a6f8031829cb21ab105bb5c109 --- /dev/null +++ b/benchmark/estimation/results/pairs.json @@ -0,0 +1,236 @@ +[ + { + "both_completed": false, + "both_eligible": false, + "configuration_level": "c2", + "dataset": "math", + "first_task_gpu_count": 2, + "four_gpu_state": "unresolved", + "pair_id": "est-qwen25-7b-math-c2-s153", + "peak_difference_4_minus_2_mib": null, + "runtime_difference_4_minus_2_seconds": null, + "training_seed": 153, + "two_gpu_state": "unresolved" + }, + { + "both_completed": false, + "both_eligible": false, + "configuration_level": "c2", + "dataset": "math", + "first_task_gpu_count": 4, + "four_gpu_state": "unresolved", + "pair_id": "est-qwen25-7b-math-c2-s151", + "peak_difference_4_minus_2_mib": null, + "runtime_difference_4_minus_2_seconds": null, + "training_seed": 151, + "two_gpu_state": "unresolved" + }, + { + "both_completed": false, + "both_eligible": true, + "configuration_level": "c3", + "dataset": "gsm8k", + "first_task_gpu_count": 2, + "four_gpu_state": "memory_failure", + "pair_id": "est-qwen25-7b-gsm8k-c3-s153", + "peak_difference_4_minus_2_mib": null, + "runtime_difference_4_minus_2_seconds": null, + "training_seed": 153, + "two_gpu_state": "memory_failure" + }, + { + "both_completed": false, + "both_eligible": true, + "configuration_level": "c3", + "dataset": "math", + "first_task_gpu_count": 4, + "four_gpu_state": "memory_failure", + "pair_id": "est-qwen25-7b-math-c3-s151", + "peak_difference_4_minus_2_mib": null, + "runtime_difference_4_minus_2_seconds": null, + "training_seed": 151, + "two_gpu_state": "memory_failure" + }, + { + "both_completed": false, + "both_eligible": false, + "configuration_level": "c2", + "dataset": "gsm8k", + "first_task_gpu_count": 2, + "four_gpu_state": "unresolved", + "pair_id": "est-qwen25-7b-gsm8k-c2-s152", + "peak_difference_4_minus_2_mib": null, + "runtime_difference_4_minus_2_seconds": null, + "training_seed": 152, + "two_gpu_state": "memory_failure" + }, + { + "both_completed": false, + "both_eligible": true, + "configuration_level": "c3", + "dataset": "code_heavy_tail", + "first_task_gpu_count": 4, + "four_gpu_state": "memory_failure", + "pair_id": "est-qwen25-7b-code_heavy_tail-c3-s153", + "peak_difference_4_minus_2_mib": null, + "runtime_difference_4_minus_2_seconds": null, + "training_seed": 153, + "two_gpu_state": "memory_failure" + }, + { + "both_completed": false, + "both_eligible": false, + "configuration_level": "c2", + "dataset": "gsm8k", + "first_task_gpu_count": 2, + "four_gpu_state": "unresolved", + "pair_id": "est-qwen25-7b-gsm8k-c2-s151", + "peak_difference_4_minus_2_mib": null, + "runtime_difference_4_minus_2_seconds": null, + "training_seed": 151, + "two_gpu_state": "memory_failure" + }, + { + "both_completed": false, + "both_eligible": false, + "configuration_level": "c2", + "dataset": "math", + "first_task_gpu_count": 2, + "four_gpu_state": "unresolved", + "pair_id": "est-qwen25-7b-math-c2-s152", + "peak_difference_4_minus_2_mib": null, + "runtime_difference_4_minus_2_seconds": null, + "training_seed": 152, + "two_gpu_state": "memory_failure" + }, + { + "both_completed": false, + "both_eligible": true, + "configuration_level": "c3", + "dataset": "code_heavy_tail", + "first_task_gpu_count": 4, + "four_gpu_state": "memory_failure", + "pair_id": "est-qwen25-7b-code_heavy_tail-c3-s151", + "peak_difference_4_minus_2_mib": null, + "runtime_difference_4_minus_2_seconds": null, + "training_seed": 151, + "two_gpu_state": "memory_failure" + }, + { + "both_completed": false, + "both_eligible": false, + "configuration_level": "c2", + "dataset": "code_heavy_tail", + "first_task_gpu_count": 2, + "four_gpu_state": "unresolved", + "pair_id": "est-qwen25-7b-code_heavy_tail-c2-s151", + "peak_difference_4_minus_2_mib": null, + "runtime_difference_4_minus_2_seconds": null, + "training_seed": 151, + "two_gpu_state": "memory_failure" + }, + { + "both_completed": false, + "both_eligible": false, + "configuration_level": "c2", + "dataset": "gsm8k", + "first_task_gpu_count": 4, + "four_gpu_state": "unresolved", + "pair_id": "est-qwen25-7b-gsm8k-c2-s153", + "peak_difference_4_minus_2_mib": null, + "runtime_difference_4_minus_2_seconds": null, + "training_seed": 153, + "two_gpu_state": "memory_failure" + }, + { + "both_completed": false, + "both_eligible": true, + "configuration_level": "c3", + "dataset": "code_heavy_tail", + "first_task_gpu_count": 2, + "four_gpu_state": "memory_failure", + "pair_id": "est-qwen25-7b-code_heavy_tail-c3-s152", + "peak_difference_4_minus_2_mib": null, + "runtime_difference_4_minus_2_seconds": null, + "training_seed": 152, + "two_gpu_state": "memory_failure" + }, + { + "both_completed": false, + "both_eligible": false, + "configuration_level": "c2", + "dataset": "code_heavy_tail", + "first_task_gpu_count": 2, + "four_gpu_state": "unresolved", + "pair_id": "est-qwen25-7b-code_heavy_tail-c2-s152", + "peak_difference_4_minus_2_mib": null, + "runtime_difference_4_minus_2_seconds": null, + "training_seed": 152, + "two_gpu_state": "memory_failure" + }, + { + "both_completed": false, + "both_eligible": true, + "configuration_level": "c3", + "dataset": "gsm8k", + "first_task_gpu_count": 4, + "four_gpu_state": "memory_failure", + "pair_id": "est-qwen25-7b-gsm8k-c3-s151", + "peak_difference_4_minus_2_mib": null, + "runtime_difference_4_minus_2_seconds": null, + "training_seed": 151, + "two_gpu_state": "memory_failure" + }, + { + "both_completed": false, + "both_eligible": true, + "configuration_level": "c3", + "dataset": "gsm8k", + "first_task_gpu_count": 4, + "four_gpu_state": "memory_failure", + "pair_id": "est-qwen25-7b-gsm8k-c3-s152", + "peak_difference_4_minus_2_mib": null, + "runtime_difference_4_minus_2_seconds": null, + "training_seed": 152, + "two_gpu_state": "memory_failure" + }, + { + "both_completed": false, + "both_eligible": true, + "configuration_level": "c3", + "dataset": "math", + "first_task_gpu_count": 2, + "four_gpu_state": "memory_failure", + "pair_id": "est-qwen25-7b-math-c3-s153", + "peak_difference_4_minus_2_mib": null, + "runtime_difference_4_minus_2_seconds": null, + "training_seed": 153, + "two_gpu_state": "memory_failure" + }, + { + "both_completed": false, + "both_eligible": false, + "configuration_level": "c2", + "dataset": "code_heavy_tail", + "first_task_gpu_count": 4, + "four_gpu_state": "unresolved", + "pair_id": "est-qwen25-7b-code_heavy_tail-c2-s153", + "peak_difference_4_minus_2_mib": null, + "runtime_difference_4_minus_2_seconds": null, + "training_seed": 153, + "two_gpu_state": "memory_failure" + }, + { + "both_completed": false, + "both_eligible": false, + "configuration_level": "c3", + "dataset": "math", + "first_task_gpu_count": 4, + "four_gpu_state": "unresolved", + "pair_id": "est-qwen25-7b-math-c3-s152", + "peak_difference_4_minus_2_mib": null, + "runtime_difference_4_minus_2_seconds": null, + "training_seed": 152, + "two_gpu_state": "unresolved" + } +] diff --git a/benchmark/estimation/results/processes.csv b/benchmark/estimation/results/processes.csv new file mode 100644 index 0000000000000000000000000000000000000000..1b0f2e0c438bba8c808821e913ee7dbc70f15d39 --- /dev/null +++ b/benchmark/estimation/results/processes.csv @@ -0,0 +1,37 @@ +allocation_gpu_count,artifact_path,completed,completed_peak_mib,configuration_id,configuration_level,dataset,dispatched,durable_terminal_record_present,eligible,expected_execution_commit,experiment_id,failure_kind,failure_stage,invocation_elapsed_seconds,invocation_id,invocation_reservation_gpu_seconds,job_id,known_margin_exceedance,known_memory_failure,launched,observed_task_peak_mib,online_status,pair_id,period,reason,state,task_gpu_count,task_gpu_seconds,task_runtime_seconds,training_seed,validation_errors,worker_initialization +4,output/prospective_estimation/est-qwen25-7b-math-c2-s153-2gpu/attempt-5054339,False,,est-qwen25-7b-math-c2-2gpu,c2,math,True,True,False,0bd659b3f69e2f779dbea84bb2da221bd0ff52db,est-qwen25-7b-math-c2-s153-2gpu,configuration,training,44.64338662382215,est-qwen25-7b-math-c2-s153-2gpu-j5054339,178.5735464952886,5054339,False,False,True,,infrastructure_invalid,est-qwen25-7b-math-c2-s153,1,,unresolved,2,89.2867732476443,,153,"[""ValueError: ValueError: Non-memory failure""]", +4,output/prospective_estimation/est-qwen25-7b-math-c2-s153-4gpu/attempt-5054339,False,,est-qwen25-7b-math-c2-4gpu,c2,math,True,True,False,0bd659b3f69e2f779dbea84bb2da221bd0ff52db,est-qwen25-7b-math-c2-s153-4gpu,configuration,training,31.34275609999895,est-qwen25-7b-math-c2-s153-4gpu-j5054339,125.3710243999958,5054339,False,False,True,,infrastructure_invalid,est-qwen25-7b-math-c2-s153,2,,unresolved,4,125.3710243999958,,153,"[""ValueError: ValueError: Invocation/contamination/cleanup evidence is invalid""]", +4,output/prospective_estimation/est-qwen25-7b-math-c2-s151-4gpu/attempt-5054345,False,,est-qwen25-7b-math-c2-4gpu,c2,math,True,True,False,0bd659b3f69e2f779dbea84bb2da221bd0ff52db,est-qwen25-7b-math-c2-s151-4gpu,configuration,training,31.415447635110468,est-qwen25-7b-math-c2-s151-4gpu-j5054345,125.66179054044187,5054345,False,False,True,,infrastructure_invalid,est-qwen25-7b-math-c2-s151,1,,unresolved,4,125.66179054044187,,151,"[""ValueError: ValueError: Non-memory failure""]", +4,output/prospective_estimation/est-qwen25-7b-math-c2-s151-2gpu/attempt-5054345,False,,est-qwen25-7b-math-c2-2gpu,c2,math,True,True,False,0bd659b3f69e2f779dbea84bb2da221bd0ff52db,est-qwen25-7b-math-c2-s151-2gpu,,device_preflight,10.315737758763134,est-qwen25-7b-math-c2-s151-2gpu-j5054345,41.26295103505254,5054345,False,False,True,,infrastructure_invalid,est-qwen25-7b-math-c2-s151,2,,unresolved,2,20.63147551752627,,151,"[""ValueError: ValueError: Invocation/contamination/cleanup evidence is invalid""]", +4,output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-2gpu/attempt-5054952,False,,est-qwen25-7b-gsm8k-c3-2gpu,c3,gsm8k,True,True,True,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-gsm8k-c3-s153-2gpu,rollout_init_memory,training,164.04821749590337,est-qwen25-7b-gsm8k-c3-s153-2gpu-j5054952,656.1928699836135,5054952,False,True,True,28882.0,memory_failure,est-qwen25-7b-gsm8k-c3-s153,1,,memory_failure,2,328.09643499180675,158.0,153,[],complete +4,output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-4gpu/attempt-5054952,False,,est-qwen25-7b-gsm8k-c3-4gpu,c3,gsm8k,True,True,True,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-gsm8k-c3-s153-4gpu,rollout_init_memory,training,133.82517697988078,est-qwen25-7b-gsm8k-c3-s153-4gpu-j5054952,535.3007079195231,5054952,False,True,True,25512.0,memory_failure,est-qwen25-7b-gsm8k-c3-s153,2,,memory_failure,4,535.3007079195231,126.0,153,[],complete +4,output/prospective_estimation/est-qwen25-7b-math-c3-s151-4gpu/attempt-5054953,False,,est-qwen25-7b-math-c3-4gpu,c3,math,True,True,True,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-math-c3-s151-4gpu,rollout_init_memory,training,165.7348702265881,est-qwen25-7b-math-c3-s151-4gpu-j5054953,662.9394809063524,5054953,False,True,True,25512.0,memory_failure,est-qwen25-7b-math-c3-s151,1,,memory_failure,4,662.9394809063524,158.0,151,[],complete +4,output/prospective_estimation/est-qwen25-7b-math-c3-s151-2gpu/attempt-5054953,False,,est-qwen25-7b-math-c3-2gpu,c3,math,True,True,True,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-math-c3-s151-2gpu,rollout_init_memory,training,127.3300099670887,est-qwen25-7b-math-c3-s151-2gpu-j5054953,509.3200398683548,5054953,False,True,True,28882.0,memory_failure,est-qwen25-7b-math-c3-s151,2,,memory_failure,2,254.6600199341774,121.0,151,[],complete +4,output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-2gpu/attempt-5054954,False,,est-qwen25-7b-gsm8k-c2-2gpu,c2,gsm8k,True,True,True,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-gsm8k-c2-s152-2gpu,rollout_init_memory,training,165.29123209230602,est-qwen25-7b-gsm8k-c2-s152-2gpu-j5054954,661.1649283692241,5054954,False,True,True,28882.0,memory_failure,est-qwen25-7b-gsm8k-c2-s152,1,,memory_failure,2,330.58246418461204,158.0,152,[],complete +4,output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-4gpu/attempt-5054954,False,,est-qwen25-7b-gsm8k-c2-4gpu,c2,gsm8k,True,True,False,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-gsm8k-c2-s152-4gpu,vllm_executor_failure,training,339.9678740478121,est-qwen25-7b-gsm8k-c2-s152-4gpu-j5054954,1359.8714961912483,5054954,False,False,True,,infrastructure_invalid,est-qwen25-7b-gsm8k-c2-s152,2,,unresolved,4,1359.8714961912483,,152,"[""ValueError: ValueError: Non-memory failure""]", +4,output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-4gpu/attempt-5054977,False,,est-qwen25-7b-code_heavy_tail-c3-4gpu,c3,code_heavy_tail,True,True,True,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-code_heavy_tail-c3-s153-4gpu,rollout_init_memory,training,140.9758697557263,est-qwen25-7b-code_heavy_tail-c3-s153-4gpu-j5054977,563.9034790229052,5054977,False,True,True,24580.0,memory_failure,est-qwen25-7b-code_heavy_tail-c3-s153,1,,memory_failure,4,563.9034790229052,133.0,153,[],complete +4,output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-2gpu/attempt-5054977,False,,est-qwen25-7b-code_heavy_tail-c3-2gpu,c3,code_heavy_tail,True,True,True,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-code_heavy_tail-c3-s153-2gpu,rollout_init_memory,training,123.39779567485675,est-qwen25-7b-code_heavy_tail-c3-s153-2gpu-j5054977,493.591182699427,5054977,False,True,True,28882.0,memory_failure,est-qwen25-7b-code_heavy_tail-c3-s153,2,,memory_failure,2,246.7955913497135,116.0,153,[],complete +4,output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-2gpu/attempt-5054981,False,,est-qwen25-7b-gsm8k-c2-2gpu,c2,gsm8k,True,True,True,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-gsm8k-c2-s151-2gpu,rollout_init_memory,training,131.12328019738197,est-qwen25-7b-gsm8k-c2-s151-2gpu-j5054981,524.4931207895279,5054981,False,True,True,28882.0,memory_failure,est-qwen25-7b-gsm8k-c2-s151,1,,memory_failure,2,262.24656039476395,125.0,151,[],complete +4,output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-4gpu/attempt-5054981,False,,est-qwen25-7b-gsm8k-c2-4gpu,c2,gsm8k,True,True,False,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-gsm8k-c2-s151-4gpu,vllm_executor_failure,training,363.27709137881175,est-qwen25-7b-gsm8k-c2-s151-4gpu-j5054981,1453.108365515247,5054981,False,False,True,,infrastructure_invalid,est-qwen25-7b-gsm8k-c2-s151,2,,unresolved,4,1453.108365515247,,151,"[""ValueError: ValueError: Non-memory failure""]", +4,output/prospective_estimation/est-qwen25-7b-math-c2-s152-2gpu/attempt-5054998,False,,est-qwen25-7b-math-c2-2gpu,c2,math,True,True,True,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-math-c2-s152-2gpu,rollout_init_memory,training,141.53408316290006,est-qwen25-7b-math-c2-s152-2gpu-j5054998,566.1363326516002,5054998,False,True,True,28882.0,memory_failure,est-qwen25-7b-math-c2-s152,1,,memory_failure,2,283.0681663258001,134.0,152,[],complete +4,output/prospective_estimation/est-qwen25-7b-math-c2-s152-4gpu/attempt-5054998,False,,est-qwen25-7b-math-c2-4gpu,c2,math,True,True,False,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-math-c2-s152-4gpu,vllm_executor_failure,training,296.0151568907313,est-qwen25-7b-math-c2-s152-4gpu-j5054998,1184.0606275629252,5054998,False,False,True,,infrastructure_invalid,est-qwen25-7b-math-c2-s152,2,,unresolved,4,1184.0606275629252,,152,"[""ValueError: ValueError: Non-memory failure""]", +4,output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-4gpu/attempt-5055003,False,,est-qwen25-7b-code_heavy_tail-c3-4gpu,c3,code_heavy_tail,True,True,True,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-code_heavy_tail-c3-s151-4gpu,rollout_init_memory,training,128.93927961122245,est-qwen25-7b-code_heavy_tail-c3-s151-4gpu-j5055003,515.7571184448898,5055003,False,True,True,25512.0,memory_failure,est-qwen25-7b-code_heavy_tail-c3-s151,1,,memory_failure,4,515.7571184448898,121.0,151,[],complete +4,output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-2gpu/attempt-5055003,False,,est-qwen25-7b-code_heavy_tail-c3-2gpu,c3,code_heavy_tail,True,True,True,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-code_heavy_tail-c3-s151-2gpu,rollout_init_memory,training,127.72932072822005,est-qwen25-7b-code_heavy_tail-c3-s151-2gpu-j5055003,510.9172829128802,5055003,False,True,True,28882.0,memory_failure,est-qwen25-7b-code_heavy_tail-c3-s151,2,,memory_failure,2,255.4586414564401,121.0,151,[],complete +4,output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-2gpu/attempt-5055021,False,,est-qwen25-7b-code_heavy_tail-c2-2gpu,c2,code_heavy_tail,True,True,True,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-code_heavy_tail-c2-s151-2gpu,rollout_init_memory,training,127.96348864305764,est-qwen25-7b-code_heavy_tail-c2-s151-2gpu-j5055021,511.8539545722306,5055021,False,True,True,28882.0,memory_failure,est-qwen25-7b-code_heavy_tail-c2-s151,1,,memory_failure,2,255.9269772861153,122.0,151,[],complete +4,output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-4gpu/attempt-5055021,False,,est-qwen25-7b-code_heavy_tail-c2-4gpu,c2,code_heavy_tail,True,True,False,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-code_heavy_tail-c2-s151-4gpu,vllm_executor_failure,training,284.9775267057121,est-qwen25-7b-code_heavy_tail-c2-s151-4gpu-j5055021,1139.9101068228483,5055021,False,False,True,,infrastructure_invalid,est-qwen25-7b-code_heavy_tail-c2-s151,2,,unresolved,4,1139.9101068228483,,151,"[""ValueError: ValueError: Non-memory failure""]", +4,output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-4gpu/attempt-5055027,False,,est-qwen25-7b-gsm8k-c2-4gpu,c2,gsm8k,True,True,False,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-gsm8k-c2-s153-4gpu,vllm_executor_failure,training,287.0094680399634,est-qwen25-7b-gsm8k-c2-s153-4gpu-j5055027,1148.0378721598536,5055027,False,False,True,,infrastructure_invalid,est-qwen25-7b-gsm8k-c2-s153,1,,unresolved,4,1148.0378721598536,,153,"[""ValueError: ValueError: Non-memory failure""]", +4,output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-2gpu/attempt-5055027,False,,est-qwen25-7b-gsm8k-c2-2gpu,c2,gsm8k,True,True,True,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-gsm8k-c2-s153-2gpu,rollout_init_memory,training,123.93428335664794,est-qwen25-7b-gsm8k-c2-s153-2gpu-j5055027,495.73713342659175,5055027,False,True,True,28882.0,memory_failure,est-qwen25-7b-gsm8k-c2-s153,2,,memory_failure,2,247.86856671329588,117.0,153,[],complete +4,output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-2gpu/attempt-5055037,False,,est-qwen25-7b-code_heavy_tail-c3-2gpu,c3,code_heavy_tail,True,True,True,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-code_heavy_tail-c3-s152-2gpu,rollout_init_memory,training,149.4319053296931,est-qwen25-7b-code_heavy_tail-c3-s152-2gpu-j5055037,597.7276213187724,5055037,False,True,True,28882.0,memory_failure,est-qwen25-7b-code_heavy_tail-c3-s152,1,,memory_failure,2,298.8638106593862,142.0,152,[],complete +4,output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-4gpu/attempt-5055037,False,,est-qwen25-7b-code_heavy_tail-c3-4gpu,c3,code_heavy_tail,True,True,True,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-code_heavy_tail-c3-s152-4gpu,rollout_init_memory,training,139.8543253610842,est-qwen25-7b-code_heavy_tail-c3-s152-4gpu-j5055037,559.4173014443368,5055037,False,True,True,25512.0,memory_failure,est-qwen25-7b-code_heavy_tail-c3-s152,2,,memory_failure,4,559.4173014443368,133.0,152,[],complete +4,output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-2gpu/attempt-5055038,False,,est-qwen25-7b-code_heavy_tail-c2-2gpu,c2,code_heavy_tail,True,True,True,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-code_heavy_tail-c2-s152-2gpu,rollout_init_memory,training,171.11857842793688,est-qwen25-7b-code_heavy_tail-c2-s152-2gpu-j5055038,684.4743137117475,5055038,False,True,True,28882.0,memory_failure,est-qwen25-7b-code_heavy_tail-c2-s152,1,,memory_failure,2,342.23715685587376,162.0,152,[],complete +4,output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-4gpu/attempt-5055038,False,,est-qwen25-7b-code_heavy_tail-c2-4gpu,c2,code_heavy_tail,True,True,False,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-code_heavy_tail-c2-s152-4gpu,vllm_executor_failure,training,356.01408890401945,est-qwen25-7b-code_heavy_tail-c2-s152-4gpu-j5055038,1424.0563556160778,5055038,False,False,True,,infrastructure_invalid,est-qwen25-7b-code_heavy_tail-c2-s152,2,,unresolved,4,1424.0563556160778,,152,"[""ValueError: ValueError: Non-memory failure""]", +4,output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-4gpu/attempt-5055039,False,,est-qwen25-7b-gsm8k-c3-4gpu,c3,gsm8k,True,True,True,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-gsm8k-c3-s151-4gpu,rollout_init_memory,training,142.68559470400214,est-qwen25-7b-gsm8k-c3-s151-4gpu-j5055039,570.7423788160086,5055039,False,True,True,25512.0,memory_failure,est-qwen25-7b-gsm8k-c3-s151,1,,memory_failure,4,570.7423788160086,135.0,151,[],complete +4,output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-2gpu/attempt-5055039,False,,est-qwen25-7b-gsm8k-c3-2gpu,c3,gsm8k,True,True,True,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-gsm8k-c3-s151-2gpu,rollout_init_memory,training,126.38375357398763,est-qwen25-7b-gsm8k-c3-s151-2gpu-j5055039,505.53501429595053,5055039,False,True,True,28882.0,memory_failure,est-qwen25-7b-gsm8k-c3-s151,2,,memory_failure,2,252.76750714797527,119.0,151,[],complete +4,output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-4gpu/attempt-5055040,False,,est-qwen25-7b-gsm8k-c3-4gpu,c3,gsm8k,True,True,True,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-gsm8k-c3-s152-4gpu,rollout_init_memory,training,133.53233256097883,est-qwen25-7b-gsm8k-c3-s152-4gpu-j5055040,534.1293302439153,5055040,False,True,True,25512.0,memory_failure,est-qwen25-7b-gsm8k-c3-s152,1,,memory_failure,4,534.1293302439153,126.0,152,[],complete +4,output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-2gpu/attempt-5055040,False,,est-qwen25-7b-gsm8k-c3-2gpu,c3,gsm8k,True,True,True,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-gsm8k-c3-s152-2gpu,rollout_init_memory,training,136.92342145182192,est-qwen25-7b-gsm8k-c3-s152-2gpu-j5055040,547.6936858072877,5055040,False,True,True,28882.0,memory_failure,est-qwen25-7b-gsm8k-c3-s152,2,,memory_failure,2,273.84684290364385,125.0,152,[],complete +4,output/prospective_estimation/est-qwen25-7b-math-c3-s153-2gpu/attempt-5055042,False,,est-qwen25-7b-math-c3-2gpu,c3,math,True,True,True,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-math-c3-s153-2gpu,rollout_init_memory,training,128.12196724722162,est-qwen25-7b-math-c3-s153-2gpu-j5055042,512.4878689888865,5055042,False,True,True,28882.0,memory_failure,est-qwen25-7b-math-c3-s153,1,,memory_failure,2,256.24393449444324,121.0,153,[],complete +4,output/prospective_estimation/est-qwen25-7b-math-c3-s153-4gpu/attempt-5055042,False,,est-qwen25-7b-math-c3-4gpu,c3,math,True,True,True,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-math-c3-s153-4gpu,rollout_init_memory,training,133.34723173687235,est-qwen25-7b-math-c3-s153-4gpu-j5055042,533.3889269474894,5055042,False,True,True,25512.0,memory_failure,est-qwen25-7b-math-c3-s153,2,,memory_failure,4,533.3889269474894,126.0,153,[],complete +4,output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-4gpu/attempt-5055043,False,,est-qwen25-7b-code_heavy_tail-c2-4gpu,c2,code_heavy_tail,True,True,False,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-code_heavy_tail-c2-s153-4gpu,vllm_executor_failure,training,297.3917622794397,est-qwen25-7b-code_heavy_tail-c2-s153-4gpu-j5055043,1189.5670491177589,5055043,False,False,True,,infrastructure_invalid,est-qwen25-7b-code_heavy_tail-c2-s153,1,,unresolved,4,1189.5670491177589,,153,"[""ValueError: ValueError: Non-memory failure""]", +4,output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-2gpu/attempt-5055043,False,,est-qwen25-7b-code_heavy_tail-c2-2gpu,c2,code_heavy_tail,True,True,True,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-code_heavy_tail-c2-s153-2gpu,rollout_init_memory,training,129.19435322191566,est-qwen25-7b-code_heavy_tail-c2-s153-2gpu-j5055043,516.7774128876626,5055043,False,True,True,28882.0,memory_failure,est-qwen25-7b-code_heavy_tail-c2-s153,2,,memory_failure,2,258.3887064438313,122.0,153,[],complete +4,,False,,est-qwen25-7b-math-c3-4gpu,c3,math,False,True,False,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-math-c3-s152-4gpu,,not_started,0.0,,0.0,,False,False,False,,not_launched,est-qwen25-7b-math-c3-s152,1,ValueError: Scheduler returned a different allocation,unresolved,4,0.0,,152,[], +4,,False,,est-qwen25-7b-math-c3-2gpu,c3,math,False,True,False,d8e0c3551d8572fb64e3b7126fbd7569a987973f,est-qwen25-7b-math-c3-s152-2gpu,,not_started,0.0,,0.0,,False,False,False,,not_launched,est-qwen25-7b-math-c3-s152,2,ValueError: Scheduler returned a different allocation,unresolved,2,0.0,,152,[], diff --git a/benchmark/estimation/results/processes.json b/benchmark/estimation/results/processes.json new file mode 100644 index 0000000000000000000000000000000000000000..17e9d180e4cd11cabc048343b081d59eb08aeed6 --- /dev/null +++ b/benchmark/estimation/results/processes.json @@ -0,0 +1,1244 @@ +[ + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-math-c2-s153-2gpu/attempt-5054339", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-math-c2-2gpu", + "configuration_level": "c2", + "dataset": "math", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": false, + "expected_execution_commit": "0bd659b3f69e2f779dbea84bb2da221bd0ff52db", + "experiment_id": "est-qwen25-7b-math-c2-s153-2gpu", + "failure_kind": "configuration", + "failure_stage": "training", + "invocation_elapsed_seconds": 44.64338662382215, + "invocation_id": "est-qwen25-7b-math-c2-s153-2gpu-j5054339", + "invocation_reservation_gpu_seconds": 178.5735464952886, + "job_id": "5054339", + "known_margin_exceedance": false, + "known_memory_failure": false, + "launched": true, + "observed_task_peak_mib": null, + "online_status": "infrastructure_invalid", + "pair_id": "est-qwen25-7b-math-c2-s153", + "period": 1, + "state": "unresolved", + "task_gpu_count": 2, + "task_gpu_seconds": 89.2867732476443, + "task_runtime_seconds": null, + "training_seed": 153, + "validation_errors": [ + "ValueError: ValueError: Non-memory failure" + ], + "worker_initialization": null + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-math-c2-s153-4gpu/attempt-5054339", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-math-c2-4gpu", + "configuration_level": "c2", + "dataset": "math", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": false, + "expected_execution_commit": "0bd659b3f69e2f779dbea84bb2da221bd0ff52db", + "experiment_id": "est-qwen25-7b-math-c2-s153-4gpu", + "failure_kind": "configuration", + "failure_stage": "training", + "invocation_elapsed_seconds": 31.34275609999895, + "invocation_id": "est-qwen25-7b-math-c2-s153-4gpu-j5054339", + "invocation_reservation_gpu_seconds": 125.3710243999958, + "job_id": "5054339", + "known_margin_exceedance": false, + "known_memory_failure": false, + "launched": true, + "observed_task_peak_mib": null, + "online_status": "infrastructure_invalid", + "pair_id": "est-qwen25-7b-math-c2-s153", + "period": 2, + "state": "unresolved", + "task_gpu_count": 4, + "task_gpu_seconds": 125.3710243999958, + "task_runtime_seconds": null, + "training_seed": 153, + "validation_errors": [ + "ValueError: ValueError: Invocation/contamination/cleanup evidence is invalid" + ], + "worker_initialization": null + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-math-c2-s151-4gpu/attempt-5054345", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-math-c2-4gpu", + "configuration_level": "c2", + "dataset": "math", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": false, + "expected_execution_commit": "0bd659b3f69e2f779dbea84bb2da221bd0ff52db", + "experiment_id": "est-qwen25-7b-math-c2-s151-4gpu", + "failure_kind": "configuration", + "failure_stage": "training", + "invocation_elapsed_seconds": 31.415447635110468, + "invocation_id": "est-qwen25-7b-math-c2-s151-4gpu-j5054345", + "invocation_reservation_gpu_seconds": 125.66179054044187, + "job_id": "5054345", + "known_margin_exceedance": false, + "known_memory_failure": false, + "launched": true, + "observed_task_peak_mib": null, + "online_status": "infrastructure_invalid", + "pair_id": "est-qwen25-7b-math-c2-s151", + "period": 1, + "state": "unresolved", + "task_gpu_count": 4, + "task_gpu_seconds": 125.66179054044187, + "task_runtime_seconds": null, + "training_seed": 151, + "validation_errors": [ + "ValueError: ValueError: Non-memory failure" + ], + "worker_initialization": null + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-math-c2-s151-2gpu/attempt-5054345", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-math-c2-2gpu", + "configuration_level": "c2", + "dataset": "math", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": false, + "expected_execution_commit": "0bd659b3f69e2f779dbea84bb2da221bd0ff52db", + "experiment_id": "est-qwen25-7b-math-c2-s151-2gpu", + "failure_kind": null, + "failure_stage": "device_preflight", + "invocation_elapsed_seconds": 10.315737758763134, + "invocation_id": "est-qwen25-7b-math-c2-s151-2gpu-j5054345", + "invocation_reservation_gpu_seconds": 41.26295103505254, + "job_id": "5054345", + "known_margin_exceedance": false, + "known_memory_failure": false, + "launched": true, + "observed_task_peak_mib": null, + "online_status": "infrastructure_invalid", + "pair_id": "est-qwen25-7b-math-c2-s151", + "period": 2, + "state": "unresolved", + "task_gpu_count": 2, + "task_gpu_seconds": 20.63147551752627, + "task_runtime_seconds": null, + "training_seed": 151, + "validation_errors": [ + "ValueError: ValueError: Invocation/contamination/cleanup evidence is invalid" + ], + "worker_initialization": null + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-2gpu/attempt-5054952", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-gsm8k-c3-2gpu", + "configuration_level": "c3", + "dataset": "gsm8k", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-gsm8k-c3-s153-2gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 164.04821749590337, + "invocation_id": "est-qwen25-7b-gsm8k-c3-s153-2gpu-j5054952", + "invocation_reservation_gpu_seconds": 656.1928699836135, + "job_id": "5054952", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 28882.0, + "online_status": "memory_failure", + "pair_id": "est-qwen25-7b-gsm8k-c3-s153", + "period": 1, + "state": "memory_failure", + "task_gpu_count": 2, + "task_gpu_seconds": 328.09643499180675, + "task_runtime_seconds": 158.0, + "training_seed": 153, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-4gpu/attempt-5054952", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-gsm8k-c3-4gpu", + "configuration_level": "c3", + "dataset": "gsm8k", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-gsm8k-c3-s153-4gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 133.82517697988078, + "invocation_id": "est-qwen25-7b-gsm8k-c3-s153-4gpu-j5054952", + "invocation_reservation_gpu_seconds": 535.3007079195231, + "job_id": "5054952", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 25512.0, + "online_status": "memory_failure", + "pair_id": "est-qwen25-7b-gsm8k-c3-s153", + "period": 2, + "state": "memory_failure", + "task_gpu_count": 4, + "task_gpu_seconds": 535.3007079195231, + "task_runtime_seconds": 126.0, + "training_seed": 153, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-math-c3-s151-4gpu/attempt-5054953", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-math-c3-4gpu", + "configuration_level": "c3", + "dataset": "math", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-math-c3-s151-4gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 165.7348702265881, + "invocation_id": "est-qwen25-7b-math-c3-s151-4gpu-j5054953", + "invocation_reservation_gpu_seconds": 662.9394809063524, + "job_id": "5054953", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 25512.0, + "online_status": "memory_failure", + "pair_id": "est-qwen25-7b-math-c3-s151", + "period": 1, + "state": "memory_failure", + "task_gpu_count": 4, + "task_gpu_seconds": 662.9394809063524, + "task_runtime_seconds": 158.0, + "training_seed": 151, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-math-c3-s151-2gpu/attempt-5054953", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-math-c3-2gpu", + "configuration_level": "c3", + "dataset": "math", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-math-c3-s151-2gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 127.3300099670887, + "invocation_id": "est-qwen25-7b-math-c3-s151-2gpu-j5054953", + "invocation_reservation_gpu_seconds": 509.3200398683548, + "job_id": "5054953", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 28882.0, + "online_status": "memory_failure", + "pair_id": "est-qwen25-7b-math-c3-s151", + "period": 2, + "state": "memory_failure", + "task_gpu_count": 2, + "task_gpu_seconds": 254.6600199341774, + "task_runtime_seconds": 121.0, + "training_seed": 151, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-2gpu/attempt-5054954", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-gsm8k-c2-2gpu", + "configuration_level": "c2", + "dataset": "gsm8k", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-gsm8k-c2-s152-2gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 165.29123209230602, + "invocation_id": "est-qwen25-7b-gsm8k-c2-s152-2gpu-j5054954", + "invocation_reservation_gpu_seconds": 661.1649283692241, + "job_id": "5054954", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 28882.0, + "online_status": "memory_failure", + "pair_id": "est-qwen25-7b-gsm8k-c2-s152", + "period": 1, + "state": "memory_failure", + "task_gpu_count": 2, + "task_gpu_seconds": 330.58246418461204, + "task_runtime_seconds": 158.0, + "training_seed": 152, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-4gpu/attempt-5054954", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-gsm8k-c2-4gpu", + "configuration_level": "c2", + "dataset": "gsm8k", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": false, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-gsm8k-c2-s152-4gpu", + "failure_kind": "vllm_executor_failure", + "failure_stage": "training", + "invocation_elapsed_seconds": 339.9678740478121, + "invocation_id": "est-qwen25-7b-gsm8k-c2-s152-4gpu-j5054954", + "invocation_reservation_gpu_seconds": 1359.8714961912483, + "job_id": "5054954", + "known_margin_exceedance": false, + "known_memory_failure": false, + "launched": true, + "observed_task_peak_mib": null, + "online_status": "infrastructure_invalid", + "pair_id": "est-qwen25-7b-gsm8k-c2-s152", + "period": 2, + "state": "unresolved", + "task_gpu_count": 4, + "task_gpu_seconds": 1359.8714961912483, + "task_runtime_seconds": null, + "training_seed": 152, + "validation_errors": [ + "ValueError: ValueError: Non-memory failure" + ], + "worker_initialization": null + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-4gpu/attempt-5054977", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-code_heavy_tail-c3-4gpu", + "configuration_level": "c3", + "dataset": "code_heavy_tail", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-code_heavy_tail-c3-s153-4gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 140.9758697557263, + "invocation_id": "est-qwen25-7b-code_heavy_tail-c3-s153-4gpu-j5054977", + "invocation_reservation_gpu_seconds": 563.9034790229052, + "job_id": "5054977", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 24580.0, + "online_status": "memory_failure", + "pair_id": "est-qwen25-7b-code_heavy_tail-c3-s153", + "period": 1, + "state": "memory_failure", + "task_gpu_count": 4, + "task_gpu_seconds": 563.9034790229052, + "task_runtime_seconds": 133.0, + "training_seed": 153, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-2gpu/attempt-5054977", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-code_heavy_tail-c3-2gpu", + "configuration_level": "c3", + "dataset": "code_heavy_tail", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-code_heavy_tail-c3-s153-2gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 123.39779567485675, + "invocation_id": "est-qwen25-7b-code_heavy_tail-c3-s153-2gpu-j5054977", + "invocation_reservation_gpu_seconds": 493.591182699427, + "job_id": "5054977", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 28882.0, + "online_status": "memory_failure", + "pair_id": "est-qwen25-7b-code_heavy_tail-c3-s153", + "period": 2, + "state": "memory_failure", + "task_gpu_count": 2, + "task_gpu_seconds": 246.7955913497135, + "task_runtime_seconds": 116.0, + "training_seed": 153, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-2gpu/attempt-5054981", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-gsm8k-c2-2gpu", + "configuration_level": "c2", + "dataset": "gsm8k", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-gsm8k-c2-s151-2gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 131.12328019738197, + "invocation_id": "est-qwen25-7b-gsm8k-c2-s151-2gpu-j5054981", + "invocation_reservation_gpu_seconds": 524.4931207895279, + "job_id": "5054981", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 28882.0, + "online_status": "memory_failure", + "pair_id": "est-qwen25-7b-gsm8k-c2-s151", + "period": 1, + "state": "memory_failure", + "task_gpu_count": 2, + "task_gpu_seconds": 262.24656039476395, + "task_runtime_seconds": 125.0, + "training_seed": 151, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-4gpu/attempt-5054981", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-gsm8k-c2-4gpu", + "configuration_level": "c2", + "dataset": "gsm8k", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": false, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-gsm8k-c2-s151-4gpu", + "failure_kind": "vllm_executor_failure", + "failure_stage": "training", + "invocation_elapsed_seconds": 363.27709137881175, + "invocation_id": "est-qwen25-7b-gsm8k-c2-s151-4gpu-j5054981", + "invocation_reservation_gpu_seconds": 1453.108365515247, + "job_id": "5054981", + "known_margin_exceedance": false, + "known_memory_failure": false, + "launched": true, + "observed_task_peak_mib": null, + "online_status": "infrastructure_invalid", + "pair_id": "est-qwen25-7b-gsm8k-c2-s151", + "period": 2, + "state": "unresolved", + "task_gpu_count": 4, + "task_gpu_seconds": 1453.108365515247, + "task_runtime_seconds": null, + "training_seed": 151, + "validation_errors": [ + "ValueError: ValueError: Non-memory failure" + ], + "worker_initialization": null + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-math-c2-s152-2gpu/attempt-5054998", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-math-c2-2gpu", + "configuration_level": "c2", + "dataset": "math", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-math-c2-s152-2gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 141.53408316290006, + "invocation_id": "est-qwen25-7b-math-c2-s152-2gpu-j5054998", + "invocation_reservation_gpu_seconds": 566.1363326516002, + "job_id": "5054998", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 28882.0, + "online_status": "memory_failure", + "pair_id": "est-qwen25-7b-math-c2-s152", + "period": 1, + "state": "memory_failure", + "task_gpu_count": 2, + "task_gpu_seconds": 283.0681663258001, + "task_runtime_seconds": 134.0, + "training_seed": 152, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-math-c2-s152-4gpu/attempt-5054998", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-math-c2-4gpu", + "configuration_level": "c2", + "dataset": "math", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": false, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-math-c2-s152-4gpu", + "failure_kind": "vllm_executor_failure", + "failure_stage": "training", + "invocation_elapsed_seconds": 296.0151568907313, + "invocation_id": "est-qwen25-7b-math-c2-s152-4gpu-j5054998", + "invocation_reservation_gpu_seconds": 1184.0606275629252, + "job_id": "5054998", + "known_margin_exceedance": false, + "known_memory_failure": false, + "launched": true, + "observed_task_peak_mib": null, + "online_status": "infrastructure_invalid", + "pair_id": "est-qwen25-7b-math-c2-s152", + "period": 2, + "state": "unresolved", + "task_gpu_count": 4, + "task_gpu_seconds": 1184.0606275629252, + "task_runtime_seconds": null, + "training_seed": 152, + "validation_errors": [ + "ValueError: ValueError: Non-memory failure" + ], + "worker_initialization": null + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-4gpu/attempt-5055003", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-code_heavy_tail-c3-4gpu", + "configuration_level": "c3", + "dataset": "code_heavy_tail", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-code_heavy_tail-c3-s151-4gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 128.93927961122245, + "invocation_id": "est-qwen25-7b-code_heavy_tail-c3-s151-4gpu-j5055003", + "invocation_reservation_gpu_seconds": 515.7571184448898, + "job_id": "5055003", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 25512.0, + "online_status": "memory_failure", + "pair_id": "est-qwen25-7b-code_heavy_tail-c3-s151", + "period": 1, + "state": "memory_failure", + "task_gpu_count": 4, + "task_gpu_seconds": 515.7571184448898, + "task_runtime_seconds": 121.0, + "training_seed": 151, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-2gpu/attempt-5055003", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-code_heavy_tail-c3-2gpu", + "configuration_level": "c3", + "dataset": "code_heavy_tail", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-code_heavy_tail-c3-s151-2gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 127.72932072822005, + "invocation_id": "est-qwen25-7b-code_heavy_tail-c3-s151-2gpu-j5055003", + "invocation_reservation_gpu_seconds": 510.9172829128802, + "job_id": "5055003", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 28882.0, + "online_status": "memory_failure", + "pair_id": "est-qwen25-7b-code_heavy_tail-c3-s151", + "period": 2, + "state": "memory_failure", + "task_gpu_count": 2, + "task_gpu_seconds": 255.4586414564401, + "task_runtime_seconds": 121.0, + "training_seed": 151, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-2gpu/attempt-5055021", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-code_heavy_tail-c2-2gpu", + "configuration_level": "c2", + "dataset": "code_heavy_tail", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-code_heavy_tail-c2-s151-2gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 127.96348864305764, + "invocation_id": "est-qwen25-7b-code_heavy_tail-c2-s151-2gpu-j5055021", + "invocation_reservation_gpu_seconds": 511.8539545722306, + "job_id": "5055021", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 28882.0, + "online_status": "memory_failure", + "pair_id": "est-qwen25-7b-code_heavy_tail-c2-s151", + "period": 1, + "state": "memory_failure", + "task_gpu_count": 2, + "task_gpu_seconds": 255.9269772861153, + "task_runtime_seconds": 122.0, + "training_seed": 151, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-4gpu/attempt-5055021", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-code_heavy_tail-c2-4gpu", + "configuration_level": "c2", + "dataset": "code_heavy_tail", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": false, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-code_heavy_tail-c2-s151-4gpu", + "failure_kind": "vllm_executor_failure", + "failure_stage": "training", + "invocation_elapsed_seconds": 284.9775267057121, + "invocation_id": "est-qwen25-7b-code_heavy_tail-c2-s151-4gpu-j5055021", + "invocation_reservation_gpu_seconds": 1139.9101068228483, + "job_id": "5055021", + "known_margin_exceedance": false, + "known_memory_failure": false, + "launched": true, + "observed_task_peak_mib": null, + "online_status": "infrastructure_invalid", + "pair_id": "est-qwen25-7b-code_heavy_tail-c2-s151", + "period": 2, + "state": "unresolved", + "task_gpu_count": 4, + "task_gpu_seconds": 1139.9101068228483, + "task_runtime_seconds": null, + "training_seed": 151, + "validation_errors": [ + "ValueError: ValueError: Non-memory failure" + ], + "worker_initialization": null + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-4gpu/attempt-5055027", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-gsm8k-c2-4gpu", + "configuration_level": "c2", + "dataset": "gsm8k", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": false, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-gsm8k-c2-s153-4gpu", + "failure_kind": "vllm_executor_failure", + "failure_stage": "training", + "invocation_elapsed_seconds": 287.0094680399634, + "invocation_id": "est-qwen25-7b-gsm8k-c2-s153-4gpu-j5055027", + "invocation_reservation_gpu_seconds": 1148.0378721598536, + "job_id": "5055027", + "known_margin_exceedance": false, + "known_memory_failure": false, + "launched": true, + "observed_task_peak_mib": null, + "online_status": "infrastructure_invalid", + "pair_id": "est-qwen25-7b-gsm8k-c2-s153", + "period": 1, + "state": "unresolved", + "task_gpu_count": 4, + "task_gpu_seconds": 1148.0378721598536, + "task_runtime_seconds": null, + "training_seed": 153, + "validation_errors": [ + "ValueError: ValueError: Non-memory failure" + ], + "worker_initialization": null + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-2gpu/attempt-5055027", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-gsm8k-c2-2gpu", + "configuration_level": "c2", + "dataset": "gsm8k", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-gsm8k-c2-s153-2gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 123.93428335664794, + "invocation_id": "est-qwen25-7b-gsm8k-c2-s153-2gpu-j5055027", + "invocation_reservation_gpu_seconds": 495.73713342659175, + "job_id": "5055027", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 28882.0, + "online_status": "memory_failure", + "pair_id": "est-qwen25-7b-gsm8k-c2-s153", + "period": 2, + "state": "memory_failure", + "task_gpu_count": 2, + "task_gpu_seconds": 247.86856671329588, + "task_runtime_seconds": 117.0, + "training_seed": 153, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-2gpu/attempt-5055037", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-code_heavy_tail-c3-2gpu", + "configuration_level": "c3", + "dataset": "code_heavy_tail", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-code_heavy_tail-c3-s152-2gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 149.4319053296931, + "invocation_id": "est-qwen25-7b-code_heavy_tail-c3-s152-2gpu-j5055037", + "invocation_reservation_gpu_seconds": 597.7276213187724, + "job_id": "5055037", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 28882.0, + "online_status": "memory_failure", + "pair_id": "est-qwen25-7b-code_heavy_tail-c3-s152", + "period": 1, + "state": "memory_failure", + "task_gpu_count": 2, + "task_gpu_seconds": 298.8638106593862, + "task_runtime_seconds": 142.0, + "training_seed": 152, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-4gpu/attempt-5055037", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-code_heavy_tail-c3-4gpu", + "configuration_level": "c3", + "dataset": "code_heavy_tail", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-code_heavy_tail-c3-s152-4gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 139.8543253610842, + "invocation_id": "est-qwen25-7b-code_heavy_tail-c3-s152-4gpu-j5055037", + "invocation_reservation_gpu_seconds": 559.4173014443368, + "job_id": "5055037", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 25512.0, + "online_status": "memory_failure", + "pair_id": "est-qwen25-7b-code_heavy_tail-c3-s152", + "period": 2, + "state": "memory_failure", + "task_gpu_count": 4, + "task_gpu_seconds": 559.4173014443368, + "task_runtime_seconds": 133.0, + "training_seed": 152, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-2gpu/attempt-5055038", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-code_heavy_tail-c2-2gpu", + "configuration_level": "c2", + "dataset": "code_heavy_tail", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-code_heavy_tail-c2-s152-2gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 171.11857842793688, + "invocation_id": "est-qwen25-7b-code_heavy_tail-c2-s152-2gpu-j5055038", + "invocation_reservation_gpu_seconds": 684.4743137117475, + "job_id": "5055038", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 28882.0, + "online_status": "memory_failure", + "pair_id": "est-qwen25-7b-code_heavy_tail-c2-s152", + "period": 1, + "state": "memory_failure", + "task_gpu_count": 2, + "task_gpu_seconds": 342.23715685587376, + "task_runtime_seconds": 162.0, + "training_seed": 152, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-4gpu/attempt-5055038", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-code_heavy_tail-c2-4gpu", + "configuration_level": "c2", + "dataset": "code_heavy_tail", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": false, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-code_heavy_tail-c2-s152-4gpu", + "failure_kind": "vllm_executor_failure", + "failure_stage": "training", + "invocation_elapsed_seconds": 356.01408890401945, + "invocation_id": "est-qwen25-7b-code_heavy_tail-c2-s152-4gpu-j5055038", + "invocation_reservation_gpu_seconds": 1424.0563556160778, + "job_id": "5055038", + "known_margin_exceedance": false, + "known_memory_failure": false, + "launched": true, + "observed_task_peak_mib": null, + "online_status": "infrastructure_invalid", + "pair_id": "est-qwen25-7b-code_heavy_tail-c2-s152", + "period": 2, + "state": "unresolved", + "task_gpu_count": 4, + "task_gpu_seconds": 1424.0563556160778, + "task_runtime_seconds": null, + "training_seed": 152, + "validation_errors": [ + "ValueError: ValueError: Non-memory failure" + ], + "worker_initialization": null + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-4gpu/attempt-5055039", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-gsm8k-c3-4gpu", + "configuration_level": "c3", + "dataset": "gsm8k", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-gsm8k-c3-s151-4gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 142.68559470400214, + "invocation_id": "est-qwen25-7b-gsm8k-c3-s151-4gpu-j5055039", + "invocation_reservation_gpu_seconds": 570.7423788160086, + "job_id": "5055039", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 25512.0, + "online_status": "memory_failure", + "pair_id": "est-qwen25-7b-gsm8k-c3-s151", + "period": 1, + "state": "memory_failure", + "task_gpu_count": 4, + "task_gpu_seconds": 570.7423788160086, + "task_runtime_seconds": 135.0, + "training_seed": 151, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-2gpu/attempt-5055039", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-gsm8k-c3-2gpu", + "configuration_level": "c3", + "dataset": "gsm8k", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-gsm8k-c3-s151-2gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 126.38375357398763, + "invocation_id": "est-qwen25-7b-gsm8k-c3-s151-2gpu-j5055039", + "invocation_reservation_gpu_seconds": 505.53501429595053, + "job_id": "5055039", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 28882.0, + "online_status": "memory_failure", + "pair_id": "est-qwen25-7b-gsm8k-c3-s151", + "period": 2, + "state": "memory_failure", + "task_gpu_count": 2, + "task_gpu_seconds": 252.76750714797527, + "task_runtime_seconds": 119.0, + "training_seed": 151, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-4gpu/attempt-5055040", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-gsm8k-c3-4gpu", + "configuration_level": "c3", + "dataset": "gsm8k", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-gsm8k-c3-s152-4gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 133.53233256097883, + "invocation_id": "est-qwen25-7b-gsm8k-c3-s152-4gpu-j5055040", + "invocation_reservation_gpu_seconds": 534.1293302439153, + "job_id": "5055040", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 25512.0, + "online_status": "memory_failure", + "pair_id": "est-qwen25-7b-gsm8k-c3-s152", + "period": 1, + "state": "memory_failure", + "task_gpu_count": 4, + "task_gpu_seconds": 534.1293302439153, + "task_runtime_seconds": 126.0, + "training_seed": 152, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-2gpu/attempt-5055040", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-gsm8k-c3-2gpu", + "configuration_level": "c3", + "dataset": "gsm8k", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-gsm8k-c3-s152-2gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 136.92342145182192, + "invocation_id": "est-qwen25-7b-gsm8k-c3-s152-2gpu-j5055040", + "invocation_reservation_gpu_seconds": 547.6936858072877, + "job_id": "5055040", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 28882.0, + "online_status": "memory_failure", + "pair_id": "est-qwen25-7b-gsm8k-c3-s152", + "period": 2, + "state": "memory_failure", + "task_gpu_count": 2, + "task_gpu_seconds": 273.84684290364385, + "task_runtime_seconds": 125.0, + "training_seed": 152, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-math-c3-s153-2gpu/attempt-5055042", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-math-c3-2gpu", + "configuration_level": "c3", + "dataset": "math", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-math-c3-s153-2gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 128.12196724722162, + "invocation_id": "est-qwen25-7b-math-c3-s153-2gpu-j5055042", + "invocation_reservation_gpu_seconds": 512.4878689888865, + "job_id": "5055042", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 28882.0, + "online_status": "memory_failure", + "pair_id": "est-qwen25-7b-math-c3-s153", + "period": 1, + "state": "memory_failure", + "task_gpu_count": 2, + "task_gpu_seconds": 256.24393449444324, + "task_runtime_seconds": 121.0, + "training_seed": 153, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-math-c3-s153-4gpu/attempt-5055042", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-math-c3-4gpu", + "configuration_level": "c3", + "dataset": "math", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-math-c3-s153-4gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 133.34723173687235, + "invocation_id": "est-qwen25-7b-math-c3-s153-4gpu-j5055042", + "invocation_reservation_gpu_seconds": 533.3889269474894, + "job_id": "5055042", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 25512.0, + "online_status": "memory_failure", + "pair_id": "est-qwen25-7b-math-c3-s153", + "period": 2, + "state": "memory_failure", + "task_gpu_count": 4, + "task_gpu_seconds": 533.3889269474894, + "task_runtime_seconds": 126.0, + "training_seed": 153, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-4gpu/attempt-5055043", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-code_heavy_tail-c2-4gpu", + "configuration_level": "c2", + "dataset": "code_heavy_tail", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": false, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-code_heavy_tail-c2-s153-4gpu", + "failure_kind": "vllm_executor_failure", + "failure_stage": "training", + "invocation_elapsed_seconds": 297.3917622794397, + "invocation_id": "est-qwen25-7b-code_heavy_tail-c2-s153-4gpu-j5055043", + "invocation_reservation_gpu_seconds": 1189.5670491177589, + "job_id": "5055043", + "known_margin_exceedance": false, + "known_memory_failure": false, + "launched": true, + "observed_task_peak_mib": null, + "online_status": "infrastructure_invalid", + "pair_id": "est-qwen25-7b-code_heavy_tail-c2-s153", + "period": 1, + "state": "unresolved", + "task_gpu_count": 4, + "task_gpu_seconds": 1189.5670491177589, + "task_runtime_seconds": null, + "training_seed": 153, + "validation_errors": [ + "ValueError: ValueError: Non-memory failure" + ], + "worker_initialization": null + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-2gpu/attempt-5055043", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-code_heavy_tail-c2-2gpu", + "configuration_level": "c2", + "dataset": "code_heavy_tail", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-code_heavy_tail-c2-s153-2gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 129.19435322191566, + "invocation_id": "est-qwen25-7b-code_heavy_tail-c2-s153-2gpu-j5055043", + "invocation_reservation_gpu_seconds": 516.7774128876626, + "job_id": "5055043", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 28882.0, + "online_status": "memory_failure", + "pair_id": "est-qwen25-7b-code_heavy_tail-c2-s153", + "period": 2, + "state": "memory_failure", + "task_gpu_count": 2, + "task_gpu_seconds": 258.3887064438313, + "task_runtime_seconds": 122.0, + "training_seed": 153, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-math-c3-4gpu", + "configuration_level": "c3", + "dataset": "math", + "dispatched": false, + "durable_terminal_record_present": true, + "eligible": false, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-math-c3-s152-4gpu", + "failure_kind": null, + "failure_stage": "not_started", + "invocation_elapsed_seconds": 0.0, + "invocation_reservation_gpu_seconds": 0.0, + "known_margin_exceedance": false, + "known_memory_failure": false, + "launched": false, + "observed_task_peak_mib": null, + "online_status": "not_launched", + "pair_id": "est-qwen25-7b-math-c3-s152", + "period": 1, + "reason": "ValueError: Scheduler returned a different allocation", + "state": "unresolved", + "task_gpu_count": 4, + "task_gpu_seconds": 0.0, + "task_runtime_seconds": null, + "training_seed": 152, + "validation_errors": [], + "worker_initialization": null + }, + { + "allocation_gpu_count": 4, + "completed": false, + "completed_peak_mib": null, + "configuration_id": "est-qwen25-7b-math-c3-2gpu", + "configuration_level": "c3", + "dataset": "math", + "dispatched": false, + "durable_terminal_record_present": true, + "eligible": false, + "expected_execution_commit": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "experiment_id": "est-qwen25-7b-math-c3-s152-2gpu", + "failure_kind": null, + "failure_stage": "not_started", + "invocation_elapsed_seconds": 0.0, + "invocation_reservation_gpu_seconds": 0.0, + "known_margin_exceedance": false, + "known_memory_failure": false, + "launched": false, + "observed_task_peak_mib": null, + "online_status": "not_launched", + "pair_id": "est-qwen25-7b-math-c3-s152", + "period": 2, + "reason": "ValueError: Scheduler returned a different allocation", + "state": "unresolved", + "task_gpu_count": 2, + "task_gpu_seconds": 0.0, + "task_runtime_seconds": null, + "training_seed": 152, + "validation_errors": [], + "worker_initialization": null + } +] diff --git a/benchmark/estimation/results/provenance.json b/benchmark/estimation/results/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..7c213f491ad8fa8af6d1d2e0d3e1290dd93c0dd3 --- /dev/null +++ b/benchmark/estimation/results/provenance.json @@ -0,0 +1,2780 @@ +{ + "accounting_validation_implementation_sha256": "9b00de319d8ac76d32897322bfbdfde47248737436854111f1e1808f156d4d08", + "allocation_accounting_sha256": "1a91571d5bc83040c72a507602c8640faac7d6197608fcac4d17596c7ffc8483", + "amendments_sha256": { + "amendment-01.json": "b02fe0419f865c4cc3d77fb35bb4205a54f3aa9605fdadc7b86a9a11b172d25e" + }, + "analysis_implementation_sha256": "8326859efb69b0c0f7003932d1b491efb80618671437e87d77edbb3ec5f484ba", + "contamination_scope": "All-four allocation samples, excluded-device process membership, and recorded live-child ownership annotations. This is not continuous monitoring or an independent offline reconstruction of PID ancestry.", + "data_verification_scope": "Environment-captured dataset file hashes compared with frozen protocol; no external dataset/cache paths are opened.", + "execution_commits_expected": { + "est-qwen25-7b-code_heavy_tail-c2-s151": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-code_heavy_tail-c2-s152": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-code_heavy_tail-c2-s153": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-code_heavy_tail-c3-s151": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-code_heavy_tail-c3-s152": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-code_heavy_tail-c3-s153": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-gsm8k-c2-s151": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-gsm8k-c2-s152": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-gsm8k-c2-s153": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-gsm8k-c3-s151": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-gsm8k-c3-s152": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-gsm8k-c3-s153": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-math-c2-s151": "0bd659b3f69e2f779dbea84bb2da221bd0ff52db", + "est-qwen25-7b-math-c2-s152": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-math-c2-s153": "0bd659b3f69e2f779dbea84bb2da221bd0ff52db", + "est-qwen25-7b-math-c3-s151": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-math-c3-s152": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-math-c3-s153": "d8e0c3551d8572fb64e3b7126fbd7569a987973f" + }, + "execution_commits_file_sha256": "cf70aa7c361f114b2d06a5da5fbbbf5021a13b7d19c24a55023647f5b3c8301a", + "input_evidence": { + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-2gpu/attempt-5055021/device-evidence/probe-exit-idle.json": { + "sha256": "59ac6704b23ff5aca33cfe234516b279eb35438f0f57de2f53fc6296a6a0822f", + "size_bytes": 3699 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-2gpu/attempt-5055021/device-evidence/ray-resources.json": { + "sha256": "240ed32057f9ac3215820c4f668a9b749a8b74366e1d7110705f45f15d235540", + "size_bytes": 1568 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-2gpu/attempt-5055021/device-evidence/trainer-cuda.json": { + "sha256": "6db5d30339a7a103e282a65773db986a21c668aca74509b53d187e94612dfa05", + "size_bytes": 584 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-2gpu/attempt-5055021/device-evidence/trainer-exec.json": { + "sha256": "2e34663623ab8886b276d2f026545b1d5a91ae997c7fd4dd5799e1e5a2b004c4", + "size_bytes": 376 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-2gpu/attempt-5055021/device-evidence/trainer-terminal.json": { + "sha256": "98c6af2df44a4ed720f4ecbf79a698f228cc0e2fcb246c465b134c95234eb14f", + "size_bytes": 174 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-2gpu/attempt-5055021/device-evidence/worker-actor_rollout-0-2847646.json": { + "sha256": "c59714f3e36decf24003ab844dc2f4e28fb19742410e53f9a23a5482522258bc", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-2gpu/attempt-5055021/device-evidence/worker-actor_rollout-1-2847647.json": { + "sha256": "c68ba7e3610020755bd923894624daf1f92785f78d7680b6cbcf3e93593c8901", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-2gpu/attempt-5055021/dispatch.json": { + "sha256": "e271fc82bb39ce803f6b2907bf7c9449c6b671f5f942cade9b27835b61776392", + "size_bytes": 4624 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-2gpu/attempt-5055021/environment-5055021.json": { + "sha256": "dc9523edefc96ab630e2e890f1a7e7a9d233ec3137a3f5729471dab8ea817655", + "size_bytes": 4477 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-2gpu/attempt-5055021/gpu-device-map.json": { + "sha256": "aab4742f53515eb078dc9017769673b5359f9ae32057b1b453ee7fb7d37a893f", + "size_bytes": 480 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-2gpu/attempt-5055021/gpu-memory-5055021.csv": { + "sha256": "8eee692f090582e0996a4bb5a3c0a41dc9ca915d3a85ea8e01ddfe04681b6961", + "size_bytes": 35780 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-2gpu/attempt-5055021/gpu-telemetry-5055021.csv": { + "sha256": "124ec9b8654220701700f7632af6f3573a602dae76bb35317b354c361d9d03d0", + "size_bytes": 106440 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-2gpu/attempt-5055021/launcher.log": { + "sha256": "544226afd6c0563acec4bc8803f024df112934fd695044a2b2ece22863925763", + "size_bytes": 152195 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-2gpu/attempt-5055021/phase-memory-5055021.csv": { + "sha256": "50895000b239cde5c35211c9dec3e7c0dbadb1acc6dbec4d06e63215e38ad7f2", + "size_bytes": 54005 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-2gpu/attempt-5055021/training-5055021.log": { + "sha256": "6cba7640bf4ad67f263ef7bcdfde785b8ceddaafb925adb9e5e0cdc2e0b29c3d", + "size_bytes": 152021 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-2gpu/attempt-5055021/trial-5055021.json": { + "sha256": "7adccf2c8a206a647c52dcaee6b38f2038cbe733af5092aac876dfa17edf76be", + "size_bytes": 1484 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-4gpu/attempt-5055021/device-evidence/probe-exit-idle.json": { + "sha256": "c3895c62afadb8531a52e370c3c4547dc1a2f897213c382798d17bd6b5fa1a84", + "size_bytes": 3699 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-4gpu/attempt-5055021/device-evidence/ray-resources.json": { + "sha256": "1d1645a40094b2dbf8ca8873c537f72e9877c468ffae3e217b103c6447345ef1", + "size_bytes": 1568 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-4gpu/attempt-5055021/device-evidence/trainer-cuda.json": { + "sha256": "f15c100be8acaeac00f9f1323da3e459d2e96c757f089e1884bc1e84509547ad", + "size_bytes": 866 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-4gpu/attempt-5055021/device-evidence/trainer-exec.json": { + "sha256": "46279d7151c77515200baef466903bc6f79d84cc7895684458183ab63a62f9cb", + "size_bytes": 458 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-4gpu/attempt-5055021/device-evidence/trainer-terminal.json": { + "sha256": "2f400eeb5b4d8de17cd8666e7134c42e24395773ca546eab44ee13f0e1c2063f", + "size_bytes": 174 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-4gpu/attempt-5055021/device-evidence/worker-actor_rollout-0-2863714.json": { + "sha256": "5b46dbe63b9680b17004cdcd4602ad57ecd32d279d6f0fcb1463d4dfd31cd965", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-4gpu/attempt-5055021/device-evidence/worker-actor_rollout-1-2863716.json": { + "sha256": "8da2c0e12d30d97d21f03bfc7b5313d7f58f7ea818bedcd26758dde64637661a", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-4gpu/attempt-5055021/device-evidence/worker-actor_rollout-2-2863717.json": { + "sha256": "eac450361359b14778d7244f270a78194889e9417390d0f07c35a46fac0ddbc3", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-4gpu/attempt-5055021/device-evidence/worker-actor_rollout-3-2863718.json": { + "sha256": "9217ac7676b4b665509e9b3cbc31cc7a6df7a02d51821a1eb5f2b296fde5b864", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-4gpu/attempt-5055021/dispatch.json": { + "sha256": "9711833b6e17e33129847fbe1585f56ed91246d3f895acc5cf4fe53ea3e8be5c", + "size_bytes": 4806 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-4gpu/attempt-5055021/environment-5055021.json": { + "sha256": "41e270bd8c1b602ffa4f3264d62028501a787ffb8e269b0f703e2989c944a632", + "size_bytes": 4959 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-4gpu/attempt-5055021/gpu-device-map.json": { + "sha256": "9c4a4ab9c1b6e40323aea6b42a5bb0ff656e2a90d7efef7ea63b285d78723e01", + "size_bytes": 820 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-4gpu/attempt-5055021/gpu-memory-5055021.csv": { + "sha256": "2e4f46e888ea1e0a502288ec02b1d0185797181adb6856dd0cea10035aa91bb0", + "size_bytes": 115142 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-4gpu/attempt-5055021/gpu-telemetry-5055021.csv": { + "sha256": "3a26241b95a96a5bdc53bfa6a72419f40bcaad0b0a86f2e30884db2562adb141", + "size_bytes": 495499 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-4gpu/attempt-5055021/launcher.log": { + "sha256": "483fe40e09949a0060c50d6cad25db0eb5ac222e85db9317ca2619c813da629b", + "size_bytes": 156335 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-4gpu/attempt-5055021/phase-memory-5055021.csv": { + "sha256": "53421fc3d0d6c14065ee3d25f88527187bece56450a1d4a54f95eca60d9572bf", + "size_bytes": 156827 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-4gpu/attempt-5055021/training-5055021.log": { + "sha256": "e8989c94e4c9cc40a77bc06629623c89976d44394b1625536cd417d54cc962d7", + "size_bytes": 151931 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s151-4gpu/attempt-5055021/trial-5055021.json": { + "sha256": "e5b4c52cd1ea74a4fbed17679cb89719c29e6943d7d654af7703b17fee9aa691", + "size_bytes": 1486 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-2gpu/attempt-5055038/device-evidence/probe-exit-idle.json": { + "sha256": "06788be8e4ab4a66d84a5f888b3ae266d10a31823619c95085e8f0fb204e8e18", + "size_bytes": 3699 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-2gpu/attempt-5055038/device-evidence/ray-resources.json": { + "sha256": "d10c6b1cc8ad3606becb69efc89cdfd2f38cce6ca77a7f440db3c5d5adefe9e2", + "size_bytes": 1564 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-2gpu/attempt-5055038/device-evidence/trainer-cuda.json": { + "sha256": "7b60df8097081d1305f443ea80c96a0f8e6eca8509a256374ce81fe6411dc15c", + "size_bytes": 584 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-2gpu/attempt-5055038/device-evidence/trainer-exec.json": { + "sha256": "b746028b9da71eeedea63d2497242524579802310ce90fe20d18472edf1529a1", + "size_bytes": 376 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-2gpu/attempt-5055038/device-evidence/trainer-terminal.json": { + "sha256": "a6e0d870c7828428215a0b4c944d5419b9a18f73c3821cb627279fae5cebc364", + "size_bytes": 174 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-2gpu/attempt-5055038/device-evidence/worker-actor_rollout-0-3863206.json": { + "sha256": "6f65586995d73b33f284570b9090c2e1ef92dc2c65477e994a8c0461a1ec9faa", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-2gpu/attempt-5055038/device-evidence/worker-actor_rollout-1-3863207.json": { + "sha256": "9dc729aead9de893b70950852d573b1cc502bb87730dadfa2ec5c04e20feb2a9", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-2gpu/attempt-5055038/dispatch.json": { + "sha256": "0a9cc1883db8c345fab254b01fe8205c708390b10710a7d19e9c3af42d7ecbb1", + "size_bytes": 4624 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-2gpu/attempt-5055038/environment-5055038.json": { + "sha256": "7769820322233f09377224ce5b5e30be0049d0e13096e981a62964e6b883a602", + "size_bytes": 4478 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-2gpu/attempt-5055038/gpu-device-map.json": { + "sha256": "2bab05f7ebdbf272d9beac042b2b80296e85a6bc9b42ea0e0a6fad16f9459f9a", + "size_bytes": 480 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-2gpu/attempt-5055038/gpu-memory-5055038.csv": { + "sha256": "ebebd6645dba5d56e28a05beffde1f83035a8e61252612e04ff9f76d97f5624e", + "size_bytes": 47314 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-2gpu/attempt-5055038/gpu-telemetry-5055038.csv": { + "sha256": "4fa73da0d4251d5831f0860f051c780d3dd2ffaec39d6a320e09f70ef26e4e53", + "size_bytes": 141666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-2gpu/attempt-5055038/launcher.log": { + "sha256": "7b79b4c7e8db395bdc28a398cdd1790de8f8e8fe37d16a3bab4ed8bbd533acb4", + "size_bytes": 152465 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-2gpu/attempt-5055038/phase-memory-5055038.csv": { + "sha256": "7ed12b25686f800f6dcf0bc33b5ce7f2b79c63b6a20fa4a751dfb3fd76d67407", + "size_bytes": 71674 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-2gpu/attempt-5055038/training-5055038.log": { + "sha256": "0cc2bfe9bae03072afba2ccb12de2aa21de0ca058da66539d35577b8c7265108", + "size_bytes": 152291 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-2gpu/attempt-5055038/trial-5055038.json": { + "sha256": "6744ce597a88246ab51bd713b1ed6614a8fde76ff50ea98d9fdeec0f3fbe9eda", + "size_bytes": 1484 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-4gpu/attempt-5055038/device-evidence/probe-exit-idle.json": { + "sha256": "8b1ccc04fdd4e1f26d062c82befdcbfd8e9ea4b72855a2282bdbf6393c5f9347", + "size_bytes": 3699 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-4gpu/attempt-5055038/device-evidence/ray-resources.json": { + "sha256": "70f32aef4e5ab2d3b120af5277f34771bd135be7a7e80457f6ffb34b6040aa8b", + "size_bytes": 1564 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-4gpu/attempt-5055038/device-evidence/trainer-cuda.json": { + "sha256": "a7892dd36619cc3e5b5b410313408327106b61c83717e17d343018addad1215d", + "size_bytes": 866 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-4gpu/attempt-5055038/device-evidence/trainer-exec.json": { + "sha256": "e0c4c2ae64f028e41ddfe337ada5001dddb8d3f0bf1f48b4f6c172c2831de620", + "size_bytes": 458 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-4gpu/attempt-5055038/device-evidence/trainer-terminal.json": { + "sha256": "70ff4c3db2624624d0b6fba9f0b4a3fcd2a63b1c4518b20c056a0727c74f2359", + "size_bytes": 174 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-4gpu/attempt-5055038/device-evidence/worker-actor_rollout-0-3880160.json": { + "sha256": "aa43321177a6b725c2403c667027e4ed695d4afe115c5fc36bf1906dec254d38", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-4gpu/attempt-5055038/device-evidence/worker-actor_rollout-1-3880161.json": { + "sha256": "ea802e402fc79d4afc381104c32f2deb99b4a33192470254b0bd55b78d168c90", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-4gpu/attempt-5055038/device-evidence/worker-actor_rollout-2-3880162.json": { + "sha256": "6b3b94932f581857f1a346734a5362e32341c1667f9e303d3514dcdd600b3fce", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-4gpu/attempt-5055038/device-evidence/worker-actor_rollout-3-3880163.json": { + "sha256": "154af79171e521caa6ac463595a6f2bdd7f18ab73dbb5715b0951903cdacb675", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-4gpu/attempt-5055038/dispatch.json": { + "sha256": "bc7647952f6a372a3122de9fab3398c11177d586c12a0791ef3f42591d064c72", + "size_bytes": 4806 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-4gpu/attempt-5055038/environment-5055038.json": { + "sha256": "7ed940b09897c2c22d9b258faf30a7140e677a7d35414f97475d135bfa2e7f90", + "size_bytes": 4960 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-4gpu/attempt-5055038/gpu-device-map.json": { + "sha256": "feda07913dcfd858a305c32c66fb3ccb71d3f756ea6a1459c01a818917775b90", + "size_bytes": 820 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-4gpu/attempt-5055038/gpu-memory-5055038.csv": { + "sha256": "f72a4b3e83039ded6131e1411d2321fa02e0c0e7fc6db388e67e9fedf11d91a3", + "size_bytes": 144748 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-4gpu/attempt-5055038/gpu-telemetry-5055038.csv": { + "sha256": "dd18c56d53c5443f509c4a2c322338dcb9a137dba3cf3197140001ae0e59e991", + "size_bytes": 623657 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-4gpu/attempt-5055038/launcher.log": { + "sha256": "c551e284334c9a8f62873b12e09af98ebdcc33161c82d875a768ace9b3bf81fb", + "size_bytes": 157394 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-4gpu/attempt-5055038/phase-memory-5055038.csv": { + "sha256": "ef7f2487dfb44ef51488729d67960e1581edda80886c1b1f7eb707136bb31344", + "size_bytes": 196768 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-4gpu/attempt-5055038/training-5055038.log": { + "sha256": "0bb0940597c79ceacb042cad25784ff994cde287562bbcee3ea9f37f2b02d28f", + "size_bytes": 152990 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s152-4gpu/attempt-5055038/trial-5055038.json": { + "sha256": "d33afb7d2165c3b58f339954a726bb22cce8c5bdc50372f944ae5f9f9bb7fbdd", + "size_bytes": 1486 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-2gpu/attempt-5055043/device-evidence/probe-exit-idle.json": { + "sha256": "cd369e377d41866b76297480596ae3d3eae8ce0b6b29ae7ae37afd3e26af4d96", + "size_bytes": 3699 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-2gpu/attempt-5055043/device-evidence/ray-resources.json": { + "sha256": "9ba191429c00f916d630eb15dd9328f03b509b1fad66608a57f7cc966fc7c013", + "size_bytes": 1568 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-2gpu/attempt-5055043/device-evidence/trainer-cuda.json": { + "sha256": "1292ce78c459c1ca4c89febb68f5eb302bbcbab3c552310322ce8b6982b599a7", + "size_bytes": 584 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-2gpu/attempt-5055043/device-evidence/trainer-exec.json": { + "sha256": "625509c4097be0ab2cb2148eb2cd28de620eaa8379d84813b6663cc813a89164", + "size_bytes": 376 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-2gpu/attempt-5055043/device-evidence/trainer-terminal.json": { + "sha256": "8d1b85b4998ba51db4b89b1562337f1bcfa2a82c71b63ad82e51373e17b9de0f", + "size_bytes": 174 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-2gpu/attempt-5055043/device-evidence/worker-actor_rollout-0-2946112.json": { + "sha256": "123c71f974d50d86276b8fe8eafef84e564472281bf33e3bd586054d1063d059", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-2gpu/attempt-5055043/device-evidence/worker-actor_rollout-1-2946113.json": { + "sha256": "b6f16d2d19ccd6671d4eedc73c014fc17042a3484582017f41320dfa459a0bc9", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-2gpu/attempt-5055043/dispatch.json": { + "sha256": "398811eda2fae3f01c9dca1785e46c0c6ecb372b4588234cbe2b5b1b30241868", + "size_bytes": 4624 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-2gpu/attempt-5055043/environment-5055043.json": { + "sha256": "d5d218033f04a8e86ab72b1884addc7607df0d9796cad718400df827de4a6a6b", + "size_bytes": 4478 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-2gpu/attempt-5055043/gpu-device-map.json": { + "sha256": "519cf45cd70e1e21e20bddfd45b884855b1d761d4eea9d42bd6ce535b208f5dc", + "size_bytes": 480 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-2gpu/attempt-5055043/gpu-memory-5055043.csv": { + "sha256": "f6e255d7300c8839f9451f8c902383582d1d420b2abc0c101a945cd2017672c9", + "size_bytes": 36103 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-2gpu/attempt-5055043/gpu-telemetry-5055043.csv": { + "sha256": "3444736f71d1b39795fdf2233920f3701d947b6f2b2477ac1aff72049800306d", + "size_bytes": 107332 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-2gpu/attempt-5055043/launcher.log": { + "sha256": "05cfb24ddcb97a8471f8e5a1864bf7c4979443aa19364858a60a3f314ca39a22", + "size_bytes": 152207 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-2gpu/attempt-5055043/phase-memory-5055043.csv": { + "sha256": "bebd44ed58f397851a83ea3cad2ef0bc421fe59de299eaff0205ec9eb713604b", + "size_bytes": 54478 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-2gpu/attempt-5055043/training-5055043.log": { + "sha256": "d0ac49f27c0ab58b5c1c877a0788c62693b6be669b8ad3d53035ec618ff2ccfc", + "size_bytes": 152033 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-2gpu/attempt-5055043/trial-5055043.json": { + "sha256": "27caa17a29a90fe2e69ca0136b06f12db9a39a3b925a4e680a1cc800d7850b53", + "size_bytes": 1484 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-4gpu/attempt-5055043/device-evidence/probe-exit-idle.json": { + "sha256": "9a66db948d472b8046476007fcdad2d432a533f52838311102d420dda3901286", + "size_bytes": 3699 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-4gpu/attempt-5055043/device-evidence/ray-resources.json": { + "sha256": "1ee986794bd3c8d6b5b344fef16a70a94b1e0ca1ffb6bf8e6778ae757a798bb2", + "size_bytes": 1568 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-4gpu/attempt-5055043/device-evidence/trainer-cuda.json": { + "sha256": "9b4a22503317da374d43a23f4b53201852d9c964cd40d4c7fc959acbc648903d", + "size_bytes": 866 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-4gpu/attempt-5055043/device-evidence/trainer-exec.json": { + "sha256": "3a0df9680b509f4a0e162e4f19391d5e4f65ad3bed023fad7f1c8d16b8eb1907", + "size_bytes": 458 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-4gpu/attempt-5055043/device-evidence/trainer-terminal.json": { + "sha256": "6e1097ea5a58807f6e86e675145bd73626b937be72eeada2e26ebcad3f7b206a", + "size_bytes": 174 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-4gpu/attempt-5055043/device-evidence/worker-actor_rollout-0-2919667.json": { + "sha256": "ab56e033abbf5177755a2a1c82f5f12ae26a83b2e7e706d9c80403a8430f4290", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-4gpu/attempt-5055043/device-evidence/worker-actor_rollout-1-2919668.json": { + "sha256": "5950acfb5bdb1528655956453b08969ba18f612361c83e893ac947e0b192ac6b", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-4gpu/attempt-5055043/device-evidence/worker-actor_rollout-2-2919669.json": { + "sha256": "033e6db4c21e7adcf3dc7699e6747bf5368030f5d643e57d32c1f768d060c828", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-4gpu/attempt-5055043/device-evidence/worker-actor_rollout-3-2919670.json": { + "sha256": "360ed25da26bbfb53e39345ad17af6566d2df8ada71fd55ce8972740d722082f", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-4gpu/attempt-5055043/dispatch.json": { + "sha256": "09eeb0682b65b65d4a66fabf8d267f87956d6a3f4d3d740655b5929207ae7aa6", + "size_bytes": 4806 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-4gpu/attempt-5055043/environment-5055043.json": { + "sha256": "1b095bc4465e2ce59df24f2f28bf774e06e661d0897470eb18f1a46f420d445a", + "size_bytes": 4960 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-4gpu/attempt-5055043/gpu-device-map.json": { + "sha256": "de46fa8a4b591ab86f3e20d8ae39b52015ce63f73899ddf0934a779a32e8aa90", + "size_bytes": 820 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-4gpu/attempt-5055043/gpu-memory-5055043.csv": { + "sha256": "c28b0f36f440bfa757a26c60b04d51948918aae3844f79000e228e09f79a453b", + "size_bytes": 120246 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-4gpu/attempt-5055043/gpu-telemetry-5055043.csv": { + "sha256": "14f1ca6d04ebf29d746d7a669eca8bfafd8cfa94f2e40e163fcb3949f3521b9f", + "size_bytes": 517516 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-4gpu/attempt-5055043/launcher.log": { + "sha256": "96c0081f7748058983ef221b12d11f75ce458a55336f6fca87b01e4335aab59c", + "size_bytes": 155993 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-4gpu/attempt-5055043/phase-memory-5055043.csv": { + "sha256": "5e745ccc5158ec741e13afb1dfa5f9a98b832d78017b1f151d8676746838307a", + "size_bytes": 163731 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-4gpu/attempt-5055043/training-5055043.log": { + "sha256": "94732240e47f1f55f7b40f3dc33e8e5a790c28bb242dc85b8a461b4c7dbbd039", + "size_bytes": 151589 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c2-s153-4gpu/attempt-5055043/trial-5055043.json": { + "sha256": "55c6b063d7d740884dbd12abcd1e3e78a696134a581a8ec7eeef7417a0a56ba1", + "size_bytes": 1486 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-2gpu/attempt-5055003/device-evidence/probe-exit-idle.json": { + "sha256": "1dbca815c0ca64dfd29c17b92c98942af39216801130a3cf99a3ce25beff73cb", + "size_bytes": 3699 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-2gpu/attempt-5055003/device-evidence/ray-resources.json": { + "sha256": "f2302720a3a7d6169ee524e86cc7650d26e87c29678fd46e92d83582d6837cef", + "size_bytes": 1568 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-2gpu/attempt-5055003/device-evidence/trainer-cuda.json": { + "sha256": "425152808096f4aa55bf92085af2dc9ae2eeb8a4c6eeaeb8a3e530b8326a5ba6", + "size_bytes": 584 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-2gpu/attempt-5055003/device-evidence/trainer-exec.json": { + "sha256": "f3788a0989a0c53e94b14ac6909ea408aa54cbfc88072ba83b7279d0b1650241", + "size_bytes": 376 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-2gpu/attempt-5055003/device-evidence/trainer-terminal.json": { + "sha256": "7a026ea04526d8a6b0d6e671e4f58fd7175f132da97ce8131c3c26327695ab73", + "size_bytes": 174 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-2gpu/attempt-5055003/device-evidence/worker-actor_rollout-0-3775650.json": { + "sha256": "036022d50beb4e158213b551bacf2d74893cb4d5fee2b1853a80bbbead0aff37", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-2gpu/attempt-5055003/device-evidence/worker-actor_rollout-1-3775651.json": { + "sha256": "8a4e4feb3d04879bd4b6a8f84c8196588600730a9b91ad96e7bc84c62a15fadc", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-2gpu/attempt-5055003/dispatch.json": { + "sha256": "66cce611f2dacfc11f526cc1f651a8b34967264e97e99a750f75be5d14e3a0f9", + "size_bytes": 4626 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-2gpu/attempt-5055003/environment-5055003.json": { + "sha256": "7c6ed8206a056b2827bfcac1648aa7a45927dbe1cbeaac63fcbf4c53759e0740", + "size_bytes": 4477 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-2gpu/attempt-5055003/gpu-device-map.json": { + "sha256": "2927b900f40bf89cdeba06b93541e9938d7c190192982dec71487651e5dc1081", + "size_bytes": 480 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-2gpu/attempt-5055003/gpu-memory-5055003.csv": { + "sha256": "3d281d09002977b58be3d3d9d4aeea7c485e6b765006dd68369d835be3351b28", + "size_bytes": 35521 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-2gpu/attempt-5055003/gpu-telemetry-5055003.csv": { + "sha256": "7c31209872628d4650f44d2e068e62c72ea082b18eb4eb15f37b512e18559518", + "size_bytes": 105985 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-2gpu/attempt-5055003/launcher.log": { + "sha256": "43187a548d50712f6321fba2eaae6d0ca0b55dad038cf2f48f433aff25b54d6c", + "size_bytes": 151470 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-2gpu/attempt-5055003/phase-memory-5055003.csv": { + "sha256": "0d6c4d80d13e43d6a06d584d434219f34f4cfea7e982c4ae54e14b6fbd258731", + "size_bytes": 53701 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-2gpu/attempt-5055003/training-5055003.log": { + "sha256": "3f2f637634fc1063bb356c604b358c22487f976a9b70ad63c244cc74a4a5d016", + "size_bytes": 151296 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-2gpu/attempt-5055003/trial-5055003.json": { + "sha256": "2296fd8267a00e1bef5aac9eec0aed83db8e07491f9da9599b203f71a21f33d4", + "size_bytes": 1485 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-4gpu/attempt-5055003/device-evidence/probe-exit-idle.json": { + "sha256": "c05919680cb026fd0bb30d782fbf36bdfc232495871e622afb9001efd2e2751f", + "size_bytes": 3699 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-4gpu/attempt-5055003/device-evidence/ray-resources.json": { + "sha256": "a4b4fa60dd0b3245023f79b76de467605833c72cbad72f8940a5ffb42cda639d", + "size_bytes": 1568 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-4gpu/attempt-5055003/device-evidence/trainer-cuda.json": { + "sha256": "eeca73034df046a0157ca4ee233d695dec5578235338694c31f167a31e6915c1", + "size_bytes": 866 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-4gpu/attempt-5055003/device-evidence/trainer-exec.json": { + "sha256": "4088f7c7db31dca42e32de1b5d3f8935ff164475069c11d88183eb2dd9cedf7d", + "size_bytes": 458 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-4gpu/attempt-5055003/device-evidence/trainer-terminal.json": { + "sha256": "57f4e131503f015e788d8613412898fabb23ae8ef1ccc21ad10ca68323bafee2", + "size_bytes": 174 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-4gpu/attempt-5055003/device-evidence/worker-actor_rollout-0-3759124.json": { + "sha256": "46d595d5eea54b45f95461c4289663aee2058b51b71f3e5a49b2852807852f40", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-4gpu/attempt-5055003/device-evidence/worker-actor_rollout-1-3759125.json": { + "sha256": "14953a3ed01d4ea153adf9ff04e86f512240cbf1da62c6d0bea846a3f13f5ebd", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-4gpu/attempt-5055003/device-evidence/worker-actor_rollout-2-3759126.json": { + "sha256": "6dfa91d43e218f2ce43c2f072064b22216b0b0903cb04793fa3dbf81b84b07be", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-4gpu/attempt-5055003/device-evidence/worker-actor_rollout-3-3759127.json": { + "sha256": "cbfa10eb46c5b41110cd61b671c80f6a3bf5f735290b55f87d854e0142b31309", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-4gpu/attempt-5055003/dispatch.json": { + "sha256": "49a2eac7c803535585c4d2e290936650057d84d6e736d32f797daeea0ace0542", + "size_bytes": 4808 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-4gpu/attempt-5055003/environment-5055003.json": { + "sha256": "52eec9d31096ff784035edb6430343c8708d31b33a85090ab7bb4c803705bad1", + "size_bytes": 4959 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-4gpu/attempt-5055003/gpu-device-map.json": { + "sha256": "37d5331e002fd0857e302da14b4f0a67523f88ffc4b079147142c7377db4f86d", + "size_bytes": 820 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-4gpu/attempt-5055003/gpu-memory-5055003.csv": { + "sha256": "4e5699d4d404f1e65cb930cf3ac4d3589cae54f8b48d3ed01f213f64d81b098a", + "size_bytes": 47035 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-4gpu/attempt-5055003/gpu-telemetry-5055003.csv": { + "sha256": "d28687e50cf18cfcfa6aba6ffa83a19977d2094cacac803af5745df042680048", + "size_bytes": 212412 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-4gpu/attempt-5055003/launcher.log": { + "sha256": "82d3bc1640be7a05984ca17e5e201cf791dcd68888fd57c2c55a39eb76b7e530", + "size_bytes": 153481 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-4gpu/attempt-5055003/phase-memory-5055003.csv": { + "sha256": "c4c90a218525062d83c7c4961d00f3e487ff7c2cbe45af4ca71b4cb9afc4db0b", + "size_bytes": 65245 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-4gpu/attempt-5055003/training-5055003.log": { + "sha256": "fa3747ff929ebf99fadc14fb094422278a4a55d854b4cbc87574c65856985e36", + "size_bytes": 153307 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s151-4gpu/attempt-5055003/trial-5055003.json": { + "sha256": "0f3e3c195fe013b88f153e9338163c8603c5ca70c66031150581163ce2a670c7", + "size_bytes": 1485 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-2gpu/attempt-5055037/device-evidence/probe-exit-idle.json": { + "sha256": "4dddd6f43a859cd78c98af4fd5c6750cd8d6f6bf2e91d21e3d126b1e345a7fb5", + "size_bytes": 3697 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-2gpu/attempt-5055037/device-evidence/ray-resources.json": { + "sha256": "74866920e335bf455aee1a7f1316122724863c64ce45eb8457d656cbf0c8afe3", + "size_bytes": 1565 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-2gpu/attempt-5055037/device-evidence/trainer-cuda.json": { + "sha256": "b4d0b4527db57f795d9b1aa92533a7bb20e2f558d7706cefc6cdde44c9a844d7", + "size_bytes": 582 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-2gpu/attempt-5055037/device-evidence/trainer-exec.json": { + "sha256": "9938b206b9be5b69e7c7028f16e470db2b928ad31b67381537ec4fe32a4cdcfd", + "size_bytes": 375 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-2gpu/attempt-5055037/device-evidence/trainer-terminal.json": { + "sha256": "dff1f8e389a02941b4ec9e1754dc28949832d8ba0c03b3872befe8db64cbd717", + "size_bytes": 173 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-2gpu/attempt-5055037/device-evidence/worker-actor_rollout-0-491385.json": { + "sha256": "349131ff049ff2a5bad62f9d18292fdacbff524648703738b7966336af81d512", + "size_bytes": 665 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-2gpu/attempt-5055037/device-evidence/worker-actor_rollout-1-491386.json": { + "sha256": "4b8cfeae5cf93c1ec9dfcd2b93c20de749aeba98e2e25ef4248c8e84da39e393", + "size_bytes": 665 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-2gpu/attempt-5055037/dispatch.json": { + "sha256": "b6261583974e56d2d13c8ea39ea3065c93b1adf8e8072176a712cda589a00cd9", + "size_bytes": 4626 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-2gpu/attempt-5055037/environment-5055037.json": { + "sha256": "b41f6b3060e57d01afba52dc76d18186e52bce80573afcc5ea9ab53547fa8b6b", + "size_bytes": 4478 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-2gpu/attempt-5055037/gpu-device-map.json": { + "sha256": "ea0b9571a2378c3216b581a93c4287010f50bd11138f3e8e29a2c19e28aa87bb", + "size_bytes": 480 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-2gpu/attempt-5055037/gpu-memory-5055037.csv": { + "sha256": "299caa31a771c09a7e5d7a9c49aa1385debfb087e2ebf6081dc01eae02cbbd7a", + "size_bytes": 41497 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-2gpu/attempt-5055037/gpu-telemetry-5055037.csv": { + "sha256": "a8c7aa6ae4941da21b4a961c68d1f608614e32ebc6124ce7e3df07b2a5d7a061", + "size_bytes": 123871 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-2gpu/attempt-5055037/launcher.log": { + "sha256": "9c4b445036c304b05e0e0acebd49ab621b8f2fddf90b37b8e19fa089c2599256", + "size_bytes": 150890 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-2gpu/attempt-5055037/phase-memory-5055037.csv": { + "sha256": "150fe7a2ac73efb233b06be8082ca8d8f15c0cd11877dd8b14a99e11587ddcdb", + "size_bytes": 62752 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-2gpu/attempt-5055037/training-5055037.log": { + "sha256": "ccb5821387ec2ffc3f48d4649893763b1ed5b25568f55ee0f0c2d2c9002e7dee", + "size_bytes": 150716 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-2gpu/attempt-5055037/trial-5055037.json": { + "sha256": "dd79d25853552b59ccf59cebee27dde61a7d331aa8e0aedc8aef2f0451d216cc", + "size_bytes": 1485 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-4gpu/attempt-5055037/device-evidence/probe-exit-idle.json": { + "sha256": "2d45bea356f7df91ee8451791346e7dff9732e302082cd44c864e9992f1e3248", + "size_bytes": 3697 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-4gpu/attempt-5055037/device-evidence/ray-resources.json": { + "sha256": "e6107c10b10b2719393ed551a50964432b9767250463057733117dea58dd3642", + "size_bytes": 1565 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-4gpu/attempt-5055037/device-evidence/trainer-cuda.json": { + "sha256": "f11c4ce68ec319e3081fa242d8915b83e8685064709f2231bcc8f90954609581", + "size_bytes": 864 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-4gpu/attempt-5055037/device-evidence/trainer-exec.json": { + "sha256": "00d59ed2aa6499c0d63a09bf2382de01010943312a9ce4d029b3d3205972b9c1", + "size_bytes": 457 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-4gpu/attempt-5055037/device-evidence/trainer-terminal.json": { + "sha256": "2cb33ca007c8eb5c220102fcec672cdf785b48da834a0fef9aaf2ea4321d6044", + "size_bytes": 173 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-4gpu/attempt-5055037/device-evidence/worker-actor_rollout-0-506520.json": { + "sha256": "a4671b3883acc42c00c5703e057c1cb15f9b0516f5b82fd4c8a2eababd8ed128", + "size_bytes": 665 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-4gpu/attempt-5055037/device-evidence/worker-actor_rollout-1-506521.json": { + "sha256": "087489b7296a73ed81a2682cf6f60e391f40abb0ed449127d2e6fe90c5bba425", + "size_bytes": 665 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-4gpu/attempt-5055037/device-evidence/worker-actor_rollout-2-506522.json": { + "sha256": "c00392cd88f11271e001f3cbaa3c6b7129fe13ab26016a65ff716556aca8916f", + "size_bytes": 665 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-4gpu/attempt-5055037/device-evidence/worker-actor_rollout-3-506523.json": { + "sha256": "a32cb5e57dbf602e6a07d4427824af1c011de75188b310440a8fa614056bd05a", + "size_bytes": 665 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-4gpu/attempt-5055037/dispatch.json": { + "sha256": "fa60d7e32a7c1c67cd0b22be34ef47a2302bf8b931662065373877d26a90b177", + "size_bytes": 4808 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-4gpu/attempt-5055037/environment-5055037.json": { + "sha256": "8fa11cd57bafd85216c491025cbab112fb145c376ceecc7576243b37f8c949d5", + "size_bytes": 4960 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-4gpu/attempt-5055037/gpu-device-map.json": { + "sha256": "3fc9e4cce9959d4d5a73aefd1eff6387d1fa7025cde4a67b24ffc232b8b1be3c", + "size_bytes": 820 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-4gpu/attempt-5055037/gpu-memory-5055037.csv": { + "sha256": "ec6e33aebb9acb4027f88550bef74a22ad7dca593d4674065f71b1d52453fca3", + "size_bytes": 51276 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-4gpu/attempt-5055037/gpu-telemetry-5055037.csv": { + "sha256": "bd7f8494bc81561a4d089555630854ca3f946f47ccce271c84aa3b5e87a9a78a", + "size_bytes": 231627 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-4gpu/attempt-5055037/launcher.log": { + "sha256": "d0cd4a803cfd0f696ff08828d1e89c43d5f3211f5f8f74e71d03202298561d4f", + "size_bytes": 145218 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-4gpu/attempt-5055037/phase-memory-5055037.csv": { + "sha256": "cccf9069f246c4132130cffbee009c1dc9a500c0f4f84c6299430421ffc1bf20", + "size_bytes": 71136 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-4gpu/attempt-5055037/training-5055037.log": { + "sha256": "ef5cfb905ad581e903cef0383c9969696a07071fe4304000ce6af7896142553f", + "size_bytes": 145044 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s152-4gpu/attempt-5055037/trial-5055037.json": { + "sha256": "af0ebe13249914ddcced0693be9486881f2ed569ebd308dfa12bb18309ae3355", + "size_bytes": 1485 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-2gpu/attempt-5054977/device-evidence/probe-exit-idle.json": { + "sha256": "c7ea57595cd681d4804cd50d48de65e77cdc7154d5dfd2d0f00dfcee46b231e8", + "size_bytes": 3699 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-2gpu/attempt-5054977/device-evidence/ray-resources.json": { + "sha256": "979091ff34b7646c0e4f6345a72d02d4ce7fac2ef474de75d12a3426938f2d1a", + "size_bytes": 1568 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-2gpu/attempt-5054977/device-evidence/trainer-cuda.json": { + "sha256": "838ddd90bf7ef6893ad319ad189f98834528492119fc05c61069db76278a61ac", + "size_bytes": 584 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-2gpu/attempt-5054977/device-evidence/trainer-exec.json": { + "sha256": "f0e25344c5d2247073cfb12ed546f21999602e165e28aac966e1c69b83dd2bc7", + "size_bytes": 376 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-2gpu/attempt-5054977/device-evidence/trainer-terminal.json": { + "sha256": "14eb55652badaa6b507ea35ff6a13c26da740cd556338e0de934717c3a445481", + "size_bytes": 174 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-2gpu/attempt-5054977/device-evidence/worker-actor_rollout-0-3743555.json": { + "sha256": "f327f3b0ad3b34b927709f7caed2ea5a05eaa9bbe6b02edcad57e03eb1a70635", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-2gpu/attempt-5054977/device-evidence/worker-actor_rollout-1-3743556.json": { + "sha256": "7ba8ef82430bb9cb06d38905c0b5d529d0dc9779ddb2664f067088a3ff299b1a", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-2gpu/attempt-5054977/dispatch.json": { + "sha256": "2a834f169fc8c23cb738b438bf1097757693bb457249c79964dd4bbd98845116", + "size_bytes": 4626 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-2gpu/attempt-5054977/environment-5054977.json": { + "sha256": "25b9859ffb8893d1842eabf162857d746502aa26a76580343cb6c1c17aa5eba7", + "size_bytes": 4477 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-2gpu/attempt-5054977/gpu-device-map.json": { + "sha256": "ea0bc1e29c1fd0dc123b9ae96090692b0396962c53c2b880bbd001bce7c9fbf3", + "size_bytes": 480 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-2gpu/attempt-5054977/gpu-memory-5054977.csv": { + "sha256": "a1dd94c41cec7c30210850bd955d500c4862ee89226afa847f536428f88de1aa", + "size_bytes": 34344 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-2gpu/attempt-5054977/gpu-telemetry-5054977.csv": { + "sha256": "9aff437d423474591294b65cb27c70d23ef83d60a2cf99065cdf1a608b97410d", + "size_bytes": 102139 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-2gpu/attempt-5054977/launcher.log": { + "sha256": "a65fc554b4779680b5afa0e8df03a63dc60bd01c73445ed3ad7a0c6b348ba965", + "size_bytes": 152010 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-2gpu/attempt-5054977/phase-memory-5054977.csv": { + "sha256": "7e1a86266ba5a836ac717fb6735b150873ca1ebfb1f1adca2e3a99edfb4c6205", + "size_bytes": 51834 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-2gpu/attempt-5054977/training-5054977.log": { + "sha256": "f177f9e729edff85dd41b55be183dfcdd4c10d70bdad4f20179712b7405f8185", + "size_bytes": 151836 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-2gpu/attempt-5054977/trial-5054977.json": { + "sha256": "49ac889b4e91adeb4c6b5924d5843d5613d56a6af126fafdb272d038d0fe40f6", + "size_bytes": 1485 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-4gpu/attempt-5054977/device-evidence/probe-exit-idle.json": { + "sha256": "1e087eb0fbba4a0e247b7de9785ecdd86a4fdee87813673eac5e48fd09c24436", + "size_bytes": 3699 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-4gpu/attempt-5054977/device-evidence/ray-resources.json": { + "sha256": "037136fe1967772a27659960a5ecd1351fc84d50946229a1ebd06f7cbecfa995", + "size_bytes": 1568 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-4gpu/attempt-5054977/device-evidence/trainer-cuda.json": { + "sha256": "e6a2e3cac024ef35b611ce2d74831c7a78587d4b893c7ce4c0ce1ff3db97ba72", + "size_bytes": 866 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-4gpu/attempt-5054977/device-evidence/trainer-exec.json": { + "sha256": "df64137971a9c606dea4bbdedf3bedf76049b57a3ef0269dddffff5b9bffebb0", + "size_bytes": 458 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-4gpu/attempt-5054977/device-evidence/trainer-terminal.json": { + "sha256": "059151a4189cd3adf234f5299b6d044974b4ede7eb901800855f4d12e0b9fb94", + "size_bytes": 174 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-4gpu/attempt-5054977/device-evidence/worker-actor_rollout-0-3728119.json": { + "sha256": "af7d609c5056946c78d7e49e2121d5da002d8efa1edf05ca237bbf39ad331785", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-4gpu/attempt-5054977/device-evidence/worker-actor_rollout-1-3728120.json": { + "sha256": "4e21ba3d6ce1c507cd42ce10942e542f16a4bd79ca7dac4d54155e33ba20a8cb", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-4gpu/attempt-5054977/device-evidence/worker-actor_rollout-2-3728121.json": { + "sha256": "16c39326f48ac4786af80c726fd023d240d7b6f7da25505c28b095399c253a6a", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-4gpu/attempt-5054977/device-evidence/worker-actor_rollout-3-3728122.json": { + "sha256": "38cd6dfb455d39aa4ce8c303aee2df25ffe1062d09e0d131151e43f45d902fd7", + "size_bytes": 666 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-4gpu/attempt-5054977/dispatch.json": { + "sha256": "559e2725e3024e1a24e9d26c2fa0b621f1f0cde7fe939624c2aee024b1f35b7e", + "size_bytes": 4808 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-4gpu/attempt-5054977/environment-5054977.json": { + "sha256": "2f8de95ea29c1049c8bba2e05682492dd40bf1f56de09197548fe84ed20a5285", + "size_bytes": 4959 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-4gpu/attempt-5054977/gpu-device-map.json": { + "sha256": "d617289cda88bb19f23fa678bda283096f719bb51dbb79f3a575e4c9b9bb80d6", + "size_bytes": 820 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-4gpu/attempt-5054977/gpu-memory-5054977.csv": { + "sha256": "f9aec4bbf519828277fad7c7a25cee84f5c345318bedaa5c62f50fd645f6f2b0", + "size_bytes": 51283 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-4gpu/attempt-5054977/gpu-telemetry-5054977.csv": { + "sha256": "9f4d16122a232b98b2c37d4e61148e3bed3f9f4646e1a1777794d4d9d585ae02", + "size_bytes": 232805 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-4gpu/attempt-5054977/launcher.log": { + "sha256": "5828c1974b390835d9a09635ab2877b4c594a576f300dd645d2e3467c7d3d864", + "size_bytes": 152307 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-4gpu/attempt-5054977/phase-memory-5054977.csv": { + "sha256": "3111c35961a1cf85081605c2cb0b55bf39e005fe223f774a8b66701163955514", + "size_bytes": 71278 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-4gpu/attempt-5054977/training-5054977.log": { + "sha256": "37a43e59a8ada4e16c298ce7f0143564efc191c533f3e3927ebb80da124fd872", + "size_bytes": 152133 + }, + "output/prospective_estimation/est-qwen25-7b-code_heavy_tail-c3-s153-4gpu/attempt-5054977/trial-5054977.json": { + "sha256": "6144b3b428ca1c7d089cdbc97d76b73aafc64cb6abe9f92054e78dc0890b5412", + "size_bytes": 1485 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-2gpu/attempt-5054981/device-evidence/probe-exit-idle.json": { + "sha256": "90fa0e4b1a52340e8c4c98ff7cdda0bfa41db592bb24a1ce6f8fa7c2e77e0db1", + "size_bytes": 3689 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-2gpu/attempt-5054981/device-evidence/ray-resources.json": { + "sha256": "379306cc2fc1befcba3cff1a6b52b9b896d8048ea4b172cb0772626d95a14ed4", + "size_bytes": 1558 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-2gpu/attempt-5054981/device-evidence/trainer-cuda.json": { + "sha256": "959e67c1120b0b8d729f78c4588be2eefce7fc7969923e247fd6539a1622da5e", + "size_bytes": 574 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-2gpu/attempt-5054981/device-evidence/trainer-exec.json": { + "sha256": "6eefc78e8732ded6413708812b52170f89d81d24ba367da7181b8c03762d7aa5", + "size_bytes": 366 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-2gpu/attempt-5054981/device-evidence/trainer-terminal.json": { + "sha256": "2affc3b0e9e4c8aaf79e72f5f5e194dd43391e8499d6ada32c502ced7556908e", + "size_bytes": 164 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-2gpu/attempt-5054981/device-evidence/worker-actor_rollout-0-2806307.json": { + "sha256": "2bd7411c693d1f5e88ac7660f6f381f9add9db1b6739d1365c0b0960f65e6ada", + "size_bytes": 656 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-2gpu/attempt-5054981/device-evidence/worker-actor_rollout-1-2806308.json": { + "sha256": "e42937dd13e8751db5b0c8584f4f12a7dee51995f28d626ec6628aa58a34cb14", + "size_bytes": 656 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-2gpu/attempt-5054981/dispatch.json": { + "sha256": "a745b8f334d678244c170d5976bb6c97f25dfd70ac4b74bcfbf3fd04bfdec5de", + "size_bytes": 4321 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-2gpu/attempt-5054981/environment-5054981.json": { + "sha256": "4f23124f2bc26ef778e3540cd85cf909a1eb71eb5955a33f99a20bcd905204e4", + "size_bytes": 4415 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-2gpu/attempt-5054981/gpu-device-map.json": { + "sha256": "0b7d5c09e2e71adacea6ad1c12a389e9a3f506b224badbd21f481993bc3e6119", + "size_bytes": 470 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-2gpu/attempt-5054981/gpu-memory-5054981.csv": { + "sha256": "594b29129675b1d8aaad73f0b7f7bba385ea25a41808ec1bfa316708153365dd", + "size_bytes": 36703 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-2gpu/attempt-5054981/gpu-telemetry-5054981.csv": { + "sha256": "2f10422e6e524eb6eee6a9945b83b984a39b1147812566585d7d9ede92b4e44a", + "size_bytes": 109117 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-2gpu/attempt-5054981/launcher.log": { + "sha256": "25d890d270752afebcf6ce0b4cb69c0a47c0f0807d67f5b49f7b9aa77bfcdbb7", + "size_bytes": 145143 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-2gpu/attempt-5054981/phase-memory-5054981.csv": { + "sha256": "cc9ecf092643bc201917289cbcf6efdc4cc3cd04d0efac19063eab808ccc4795", + "size_bytes": 55378 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-2gpu/attempt-5054981/training-5054981.log": { + "sha256": "7e0c204cf0fb81c5de0b75b7423f4784870918609c53f5f9f62a5f4c60e05e79", + "size_bytes": 144979 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-2gpu/attempt-5054981/trial-5054981.json": { + "sha256": "a0e64245e03316933ae4a918d3a5b7042bf12391f2b755ebdfaed1abd2fb3574", + "size_bytes": 1416 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-4gpu/attempt-5054981/allocator-traces/allocator-2821514.jsonl": { + "sha256": "d712ede802469e2fc3d9ce82a40a0e759cb9c9e4a6a2cabafecded378ef63f97", + "size_bytes": 282 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-4gpu/attempt-5054981/device-evidence/probe-exit-idle.json": { + "sha256": "804525c85e678aa8486d51856d2a0c86a0d398d928e4d987ecea9f461d1acc43", + "size_bytes": 3689 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-4gpu/attempt-5054981/device-evidence/ray-resources.json": { + "sha256": "e125f16005583c083bfbdf37f8b02d66ffb66ded09892d9973396469cea11e48", + "size_bytes": 1558 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-4gpu/attempt-5054981/device-evidence/trainer-cuda.json": { + "sha256": "2113a7f63fe8fdb43df0a1c7093667f0828f72de9e715be77c787975b8d92d9e", + "size_bytes": 856 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-4gpu/attempt-5054981/device-evidence/trainer-exec.json": { + "sha256": "f07158d34fc4f35d2bfc5d755f3635f101bee2e95c01804a878a81966e32e094", + "size_bytes": 448 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-4gpu/attempt-5054981/device-evidence/trainer-terminal.json": { + "sha256": "61d00347814e1a6efbf91bac154fdfc7a7b82d476d0cc01f5141c3ddaba5246a", + "size_bytes": 164 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-4gpu/attempt-5054981/device-evidence/worker-actor_rollout-0-2821511.json": { + "sha256": "a29ce4602e2cb85e6bde07e0920fc634b78212dec9e28220fcc1b36894ca66df", + "size_bytes": 656 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-4gpu/attempt-5054981/device-evidence/worker-actor_rollout-1-2821512.json": { + "sha256": "631f3c3ed615605dcecc5e381c9efa5261c6cc2c0001dd3eaf9af2f89b9f2560", + "size_bytes": 656 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-4gpu/attempt-5054981/device-evidence/worker-actor_rollout-2-2821513.json": { + "sha256": "8a514ce8a8f9a171747de1879c0b6082070d966f4ffcb432d037f6f441cf42a2", + "size_bytes": 656 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-4gpu/attempt-5054981/device-evidence/worker-actor_rollout-3-2821514.json": { + "sha256": "88b41ad98e9c0f255cc91589b3dceb7723888dac04e03e2fa2eb4b19a58cd255", + "size_bytes": 656 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-4gpu/attempt-5054981/dispatch.json": { + "sha256": "7f32c0f73e6b8a5387a064ffd544a62f41ed40b83b8207ae89af865c517e4af2", + "size_bytes": 4503 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-4gpu/attempt-5054981/environment-5054981.json": { + "sha256": "87c427e3d01a09ab48ea3fc289f2fa27ca1f5607a39d68cf49d37650c74c07f3", + "size_bytes": 4897 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-4gpu/attempt-5054981/gpu-device-map.json": { + "sha256": "c2b62990f941295d01f7cbe5c12620400b43d1e5931388ba743caa1080524382", + "size_bytes": 810 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-4gpu/attempt-5054981/gpu-memory-5054981.csv": { + "sha256": "fedee8ab4ea4d93f43862a55f65364fb4b723dcbe4b60091723a51742e96cc18", + "size_bytes": 148795 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-4gpu/attempt-5054981/gpu-telemetry-5054981.csv": { + "sha256": "77ae93c83f0b2b9ed0503695fab0824223ce939879ddea126e1e44443477642b", + "size_bytes": 638968 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-4gpu/attempt-5054981/launcher.log": { + "sha256": "e5ef1d05b8ad1452bba3822d12a8c2198a15e4f4ed2401bfa0592ec8dc4ab05f", + "size_bytes": 151411 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-4gpu/attempt-5054981/phase-memory-5054981.csv": { + "sha256": "137e9f609fe7871e3a15149734c439fb3ee6a8ca63d80ed5fccbfb9d8ce2808c", + "size_bytes": 202165 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-4gpu/attempt-5054981/training-5054981.log": { + "sha256": "7e6ecd4d14d60f69fee1aa2c049ed80450ed690984240cb645effd038ce2775f", + "size_bytes": 147017 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s151-4gpu/attempt-5054981/trial-5054981.json": { + "sha256": "e4c49ff774d87acd2707f9919cc4e7e38cac180f272510ab07b7ce4d521176bb", + "size_bytes": 1418 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-2gpu/attempt-5054954/device-evidence/probe-exit-idle.json": { + "sha256": "4107ab6efd8cf9f913e0e9ce2ad869009bf653be72f96668ab572930c398d554", + "size_bytes": 3687 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-2gpu/attempt-5054954/device-evidence/ray-resources.json": { + "sha256": "fb599bb3d368204b6f67460c0f2edfb47e810fc63ea5892ef70cd38a4f1c687d", + "size_bytes": 1555 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-2gpu/attempt-5054954/device-evidence/trainer-cuda.json": { + "sha256": "71ba7599e62fcc02f34215cb6797d1670f90f64f5b7ec21f9ec1f6cc4b442b94", + "size_bytes": 572 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-2gpu/attempt-5054954/device-evidence/trainer-exec.json": { + "sha256": "2901ce44563e23a22630c972c1f7ff5b4fedb484b86cb6e8c3f985490d6b5c89", + "size_bytes": 365 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-2gpu/attempt-5054954/device-evidence/trainer-terminal.json": { + "sha256": "76938616f3b0ecdf8dd0bb130c841148738c9a8cbfeb3a5b1e759161ca5e0ac9", + "size_bytes": 163 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-2gpu/attempt-5054954/device-evidence/worker-actor_rollout-0-410106.json": { + "sha256": "d54d55e5cfd50fbfe7a6dda907bc9df188ddb97e3849fd6e4b45e9905093a341", + "size_bytes": 655 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-2gpu/attempt-5054954/device-evidence/worker-actor_rollout-1-410107.json": { + "sha256": "38328e22b423781ce6a9dd70e913733e3906739e18aedb7b123b94f5e7afd192", + "size_bytes": 655 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-2gpu/attempt-5054954/dispatch.json": { + "sha256": "66245783253fe1702d869a83ebf5dce6f7de90cdc4479b6b6888ec88cbaed8d1", + "size_bytes": 4321 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-2gpu/attempt-5054954/environment-5054954.json": { + "sha256": "c43b67c49ac56dbaa3e87669b8f172784d60fc55f5a0cff1cf48c56559828d80", + "size_bytes": 4415 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-2gpu/attempt-5054954/gpu-device-map.json": { + "sha256": "2aa119121505ecf0b1f143a601f830a8a2b658abeba9a854cbb6e5cd2896281f", + "size_bytes": 470 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-2gpu/attempt-5054954/gpu-memory-5054954.csv": { + "sha256": "49c1874add9eec467618a448372ca0683061d9bd42a6997e61e78ea9062ef750", + "size_bytes": 46060 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-2gpu/attempt-5054954/gpu-telemetry-5054954.csv": { + "sha256": "c8ba1477ae0cb483f1b8eccc6eb379b23ff5d78a9c0e8bc4a88b0e43beb06dbd", + "size_bytes": 137886 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-2gpu/attempt-5054954/launcher.log": { + "sha256": "0936366d54213dddf49d9a5658b0faab622209e8d5f457c3a859004969c50ca1", + "size_bytes": 151336 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-2gpu/attempt-5054954/phase-memory-5054954.csv": { + "sha256": "18edaf82dc8332757ab1cdd3019e941a5f24b8f7bd75ce09139dd99994bb8a90", + "size_bytes": 69760 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-2gpu/attempt-5054954/training-5054954.log": { + "sha256": "85ba10cc4807c211269cb392b762776278df81a3486b771518405ecc4bf5facb", + "size_bytes": 151172 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-2gpu/attempt-5054954/trial-5054954.json": { + "sha256": "3061299ae20effdca2058e6ec3f2c5e366edd76d89e0441a03b3bd93eaca7315", + "size_bytes": 1416 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-4gpu/attempt-5054954/device-evidence/probe-exit-idle.json": { + "sha256": "a0840a56e6ed3999c2df3a1ca4128c92106f6d375b68a2987aaaaf79a6ea9989", + "size_bytes": 3687 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-4gpu/attempt-5054954/device-evidence/ray-resources.json": { + "sha256": "54b57283b9d7b599b79c480e9fbf4f3866c6eea5236db157288ad489294ef486", + "size_bytes": 1555 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-4gpu/attempt-5054954/device-evidence/trainer-cuda.json": { + "sha256": "a00dded3334bf4fb4a146e469ae097e978036d44c7a0597176aaca40d1e7c0ae", + "size_bytes": 854 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-4gpu/attempt-5054954/device-evidence/trainer-exec.json": { + "sha256": "83b2c66f451c32a52bb6bddcf2a7080205e7376c64fcecb11d77266ad0c384a6", + "size_bytes": 447 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-4gpu/attempt-5054954/device-evidence/trainer-terminal.json": { + "sha256": "17b5d4d261d6069ba8b0490a1fe91ab677a35b1aed9892797e3bb65d42c2af28", + "size_bytes": 163 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-4gpu/attempt-5054954/device-evidence/worker-actor_rollout-0-425449.json": { + "sha256": "1fe0df63775786fe2e683c38e0227d3e772bee605caeedaff5b362cd18f186f2", + "size_bytes": 655 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-4gpu/attempt-5054954/device-evidence/worker-actor_rollout-1-425450.json": { + "sha256": "453244a6ddfb3cfa1ecf697649ac9469cada3af2dd5b74478afd8c9ac1efe70c", + "size_bytes": 655 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-4gpu/attempt-5054954/device-evidence/worker-actor_rollout-2-425451.json": { + "sha256": "b8e8310bf3a481032ed6686fa43875321c0420f080a9e9ae5599aa14375c8b8f", + "size_bytes": 655 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-4gpu/attempt-5054954/device-evidence/worker-actor_rollout-3-425452.json": { + "sha256": "196ae6067fbb4d0da5ad61f8047f5d689c5027aded996f1450f50c78007913f7", + "size_bytes": 655 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-4gpu/attempt-5054954/dispatch.json": { + "sha256": "d8b404e1ce721be1b75cb9a521417393f7fe94aa4501cf7aee4b557f2cb36a5e", + "size_bytes": 4503 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-4gpu/attempt-5054954/environment-5054954.json": { + "sha256": "de422397ce8452971821b11615c76883c1becf77095bd17df516dccb884521fb", + "size_bytes": 4897 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-4gpu/attempt-5054954/gpu-device-map.json": { + "sha256": "d71a8d5d4de49d6eeb98973e2a16438376da0a09a4c3cabc744a345cc53e2b55", + "size_bytes": 810 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-4gpu/attempt-5054954/gpu-memory-5054954.csv": { + "sha256": "cb0ac66c6a7db36de316dc325916d9a6b891ce569391fa77513bac760d531767", + "size_bytes": 139033 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-4gpu/attempt-5054954/gpu-telemetry-5054954.csv": { + "sha256": "4365fb4ad0ab3f5dd2b5d7c7495ca020fa1a5b9b8de729ada6cfdafdc60f14e1", + "size_bytes": 597830 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-4gpu/attempt-5054954/launcher.log": { + "sha256": "c1028c2500fcd9c27d42910fbc1936c8788718ca70196f842b9c52f3e3c60987", + "size_bytes": 154490 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-4gpu/attempt-5054954/phase-memory-5054954.csv": { + "sha256": "69755c74c35ff5140c4eec63a72729ed2844df34ff7d16d517f50c15bd176541", + "size_bytes": 188893 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-4gpu/attempt-5054954/training-5054954.log": { + "sha256": "d474369fe8dd11d3d56229fdf74edd99de482ddb07e2a715074602c2936e6113", + "size_bytes": 150098 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s152-4gpu/attempt-5054954/trial-5054954.json": { + "sha256": "bd51bcbcfa5a66cd9704a7f0b3832631fb64b7571119d5e54162b72e382117cb", + "size_bytes": 1418 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-2gpu/attempt-5055027/device-evidence/probe-exit-idle.json": { + "sha256": "1740e788d8ca31f8d485b02f312ae133e8984d2ab8ec2b0352f04d6aba4b79f0", + "size_bytes": 3689 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-2gpu/attempt-5055027/device-evidence/ray-resources.json": { + "sha256": "73ac142c2c64255ecdc645e3ad386685cfca2df4ea9af8b47d1b70255443e2b5", + "size_bytes": 1558 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-2gpu/attempt-5055027/device-evidence/trainer-cuda.json": { + "sha256": "b51154ea81e3bffcb63d92fdcfe1e91c99cf702178476c6c5b6246a02a403631", + "size_bytes": 574 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-2gpu/attempt-5055027/device-evidence/trainer-exec.json": { + "sha256": "49b64b8bb3ec47badefab7e6456c4b59c4845f91fe005cdbea42916b5169e1da", + "size_bytes": 366 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-2gpu/attempt-5055027/device-evidence/trainer-terminal.json": { + "sha256": "2b3a128aab444c9a43aa92da219fb3d66e1d4ffa29e8b91adf36c732d2faa77e", + "size_bytes": 164 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-2gpu/attempt-5055027/device-evidence/worker-actor_rollout-0-3813590.json": { + "sha256": "696bd08fcbef4fb038266d69ccc2d83c595d76dca9c0d376e0c2f4c51763a77b", + "size_bytes": 656 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-2gpu/attempt-5055027/device-evidence/worker-actor_rollout-1-3813591.json": { + "sha256": "66d2f3537f94d99974072439ebceb226a6c3e848f1b13b99caaa4b91e7a55b50", + "size_bytes": 656 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-2gpu/attempt-5055027/dispatch.json": { + "sha256": "ee3fe7b2450d0a8fc5126ba05bb05831b045fd30eabd91a240ac3831e2935c6b", + "size_bytes": 4321 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-2gpu/attempt-5055027/environment-5055027.json": { + "sha256": "1b08b8218f41f5c1682af48e0309e7459376d7aec57a9515e55c4faacea05b99", + "size_bytes": 4416 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-2gpu/attempt-5055027/gpu-device-map.json": { + "sha256": "a0779d6df7a6ece5304bf455b9311d601db75272721242c06992e9485f4cbbef", + "size_bytes": 470 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-2gpu/attempt-5055027/gpu-memory-5055027.csv": { + "sha256": "b482a38cf025c6f19926530e0337b07c9caa4118a74561d96d80c2080f90d35c", + "size_bytes": 34465 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-2gpu/attempt-5055027/gpu-telemetry-5055027.csv": { + "sha256": "ccb5f5e6182ce8dc290c557b40e90753f4a57b431caeb0880d1c7527d6be2814", + "size_bytes": 102542 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-2gpu/attempt-5055027/launcher.log": { + "sha256": "ff4e042e2bba908035d947e1ed2c37f118572761693337ea2597b6b022dcfb5a", + "size_bytes": 151202 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-2gpu/attempt-5055027/phase-memory-5055027.csv": { + "sha256": "4df2331347bbd97373729430a1f7e794144d5c158b610e4a0ed6fa135c2569ec", + "size_bytes": 52030 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-2gpu/attempt-5055027/training-5055027.log": { + "sha256": "3591d0732d8cd8ddf2bdb13b4619d2cb653bea94f7bac683e00b63cd6f37fd22", + "size_bytes": 151038 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-2gpu/attempt-5055027/trial-5055027.json": { + "sha256": "2b1eadcb97b5b5d95d2b6da97b29c93b1bb45aef50876db0459fd66535046e8a", + "size_bytes": 1416 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-4gpu/attempt-5055027/device-evidence/probe-exit-idle.json": { + "sha256": "9396260cb7ae11e8401f6eceec19a742a58b7cb54d658fee0571370163045007", + "size_bytes": 3689 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-4gpu/attempt-5055027/device-evidence/ray-resources.json": { + "sha256": "762d7f4b31f962551471ccdc53bcf1d2ed3d7d4acd3cac1e0bdf26ec4b03b5ae", + "size_bytes": 1558 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-4gpu/attempt-5055027/device-evidence/trainer-cuda.json": { + "sha256": "1f101fded9d3db22b50f2240f614880fda3442bba550fdc0cd6101a99a102148", + "size_bytes": 856 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-4gpu/attempt-5055027/device-evidence/trainer-exec.json": { + "sha256": "d30090c7beebfba68cb21ae77b7013b1d3eee3ac36632ff514442a98a11e8030", + "size_bytes": 448 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-4gpu/attempt-5055027/device-evidence/trainer-terminal.json": { + "sha256": "8ef37cefc5d55eb8bffebcb3894dc9118c29b3e1cee70cbb6b6984d1521ed7de", + "size_bytes": 164 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-4gpu/attempt-5055027/device-evidence/worker-actor_rollout-0-3791141.json": { + "sha256": "cf926a2b5d7815be8edfa86adb1ebe72a74a29336abf8d6263845c5e6d818dd2", + "size_bytes": 656 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-4gpu/attempt-5055027/device-evidence/worker-actor_rollout-1-3791142.json": { + "sha256": "2096638c193ca5b78bce7ae334e40047d13fb8bfd82fafad05b2dbb68f041b8d", + "size_bytes": 656 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-4gpu/attempt-5055027/device-evidence/worker-actor_rollout-2-3791143.json": { + "sha256": "c834ec491bc157a4716f57f47b77aa61e9c43386bf02072704c6b3faa051d888", + "size_bytes": 656 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-4gpu/attempt-5055027/device-evidence/worker-actor_rollout-3-3791144.json": { + "sha256": "ef8911f8ec15a89654550067fd1cbf31c88cf1710e02a2634cb9172abee32cbf", + "size_bytes": 656 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-4gpu/attempt-5055027/dispatch.json": { + "sha256": "0ee85ca91b25aefd41d19426a356b6d4bfbf8bb458b77d79cc1bebf3289cb8cf", + "size_bytes": 4503 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-4gpu/attempt-5055027/environment-5055027.json": { + "sha256": "c7bca2b0b46488551114bdc3e5f426caaa54e62650468f9aeac7819ef75c4440", + "size_bytes": 4898 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-4gpu/attempt-5055027/gpu-device-map.json": { + "sha256": "5e60dbc55d5622911b43fb84013bdd919157e11fdb296d5f0439444b7a519bf9", + "size_bytes": 810 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-4gpu/attempt-5055027/gpu-memory-5055027.csv": { + "sha256": "f113e972e4ea6055560fabc80f7164d30ac18338d46b54a5decf26288784688e", + "size_bytes": 116232 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-4gpu/attempt-5055027/gpu-telemetry-5055027.csv": { + "sha256": "690105f46069a92e5690a69e7a7289372e9ca257df6e01a4668dbfd00abef3d3", + "size_bytes": 499267 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-4gpu/attempt-5055027/launcher.log": { + "sha256": "044bf81654363aa7c6a5ffeba166a652c2f37d78b6d18f1c06cffafef27e81ce", + "size_bytes": 155291 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-4gpu/attempt-5055027/phase-memory-5055027.csv": { + "sha256": "343842d69c393a31990a3313711f8b8ad026724c408f10cda220e56490d26541", + "size_bytes": 158172 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-4gpu/attempt-5055027/training-5055027.log": { + "sha256": "dbaffc5c9c4193221001f79f952d79fe0b64faaf7e9090674d4c5de8ceeb6f4f", + "size_bytes": 150897 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c2-s153-4gpu/attempt-5055027/trial-5055027.json": { + "sha256": "6c1c8661f969a04b0b0ac3e529ba0b0df3221077145a7aac7028a079385b7d91", + "size_bytes": 1418 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-2gpu/attempt-5055039/device-evidence/probe-exit-idle.json": { + "sha256": "ba15bdea21460527e89ea13e69842b0dc631c0c75162128b6cb74a7a3a649cce", + "size_bytes": 3687 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-2gpu/attempt-5055039/device-evidence/ray-resources.json": { + "sha256": "09219e8fb099749b8eb0b5666c1ecbdd30f4111760cd7f96c8ce9fc80f29348f", + "size_bytes": 1551 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-2gpu/attempt-5055039/device-evidence/trainer-cuda.json": { + "sha256": "7eb14600fff6eac6282edc52175b1adebbbe760003a3ef31d614e61875fd8ce2", + "size_bytes": 572 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-2gpu/attempt-5055039/device-evidence/trainer-exec.json": { + "sha256": "7bc938dd09dba8725db54cc59e434f0271dd362dc8d0b1c5102aa7139d4261ea", + "size_bytes": 365 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-2gpu/attempt-5055039/device-evidence/trainer-terminal.json": { + "sha256": "61c643e521dcde021ea1617350721923fbde77850ca37b89fb4db9af50e21bba", + "size_bytes": 163 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-2gpu/attempt-5055039/device-evidence/worker-actor_rollout-0-239632.json": { + "sha256": "3da55a07a3457923c36bfa1ceb7033967479098421a7046c44611b0029fe9b85", + "size_bytes": 655 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-2gpu/attempt-5055039/device-evidence/worker-actor_rollout-1-239633.json": { + "sha256": "bc1b457afa0ef091ae3310cdba84965376acd22aaa00b37214ab0c38e4ba2be9", + "size_bytes": 655 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-2gpu/attempt-5055039/dispatch.json": { + "sha256": "2fae3952ea5c1fde9ed60ada9fccf26483d9d03afabb983d3d42ce6111e5b505", + "size_bytes": 4323 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-2gpu/attempt-5055039/environment-5055039.json": { + "sha256": "87f9dc01bbc55e3f0cc67591cc2c460d71827b5b6402c673b2f6029842b784e0", + "size_bytes": 4416 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-2gpu/attempt-5055039/gpu-device-map.json": { + "sha256": "fae28d32a9df9180c58eda619dff6b5db8df46bb18c233b2ef41234d05e45378", + "size_bytes": 470 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-2gpu/attempt-5055039/gpu-memory-5055039.csv": { + "sha256": "a54ec463961e181c556b9abc43910301f6d6a74ddd35aaaf12b358f4f49a685e", + "size_bytes": 35250 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-2gpu/attempt-5055039/gpu-telemetry-5055039.csv": { + "sha256": "d3bfb6aee8f9eae4d5630661739aaf7113868eccd2aecec1e8b9971a0f9cc08a", + "size_bytes": 104846 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-2gpu/attempt-5055039/launcher.log": { + "sha256": "205092ecb0bc5c914b36ffdbb8870ae2bb5bb09fde8a46653c372f1faaf8b59b", + "size_bytes": 150703 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-2gpu/attempt-5055039/phase-memory-5055039.csv": { + "sha256": "314d55a058f4a8ba1385c8885a2111027035665448f45ec61a236c1e81cfb825", + "size_bytes": 53205 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-2gpu/attempt-5055039/training-5055039.log": { + "sha256": "4a58745127c8523f1c96911dd815e87da77be3b06bd3a770754bee9f7235472d", + "size_bytes": 150539 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-2gpu/attempt-5055039/trial-5055039.json": { + "sha256": "dde907c9c97a7036741ad42b45ff91855d3f3734763eeb795dbca1f2b535b8e7", + "size_bytes": 1417 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-4gpu/attempt-5055039/device-evidence/probe-exit-idle.json": { + "sha256": "b12f8a6ec4e595945f6ce230731b3f086196cbca04c07c0900957dea69528998", + "size_bytes": 3687 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-4gpu/attempt-5055039/device-evidence/ray-resources.json": { + "sha256": "fa8834ac36566e78d4fdabc8a7c2604d47d79cb3c1b73b1d96a66a9aa6f07a61", + "size_bytes": 1551 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-4gpu/attempt-5055039/device-evidence/trainer-cuda.json": { + "sha256": "5fcc43e1ce2e6853aa13c3cb855f5e03da3219469ba0181951cfca7dcb100636", + "size_bytes": 854 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-4gpu/attempt-5055039/device-evidence/trainer-exec.json": { + "sha256": "c0beefaeea52dd3a3c37838d6fc01e73857f7b42c5248145b09a5aebc90fb7fd", + "size_bytes": 447 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-4gpu/attempt-5055039/device-evidence/trainer-terminal.json": { + "sha256": "f1595d9da0d74ab3c803d31eff5276d18f0b1bc657ba6e69e0fc9e7537a49e57", + "size_bytes": 163 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-4gpu/attempt-5055039/device-evidence/worker-actor_rollout-0-223649.json": { + "sha256": "26a408fbf6ae56e0593299fee27efdf253aef44354bf51094d252274f1e3c0de", + "size_bytes": 655 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-4gpu/attempt-5055039/device-evidence/worker-actor_rollout-1-223650.json": { + "sha256": "62f30eefe34dccf5318b4e9ce2d379c7e3474590f900ba0e0968e30cdb3851c6", + "size_bytes": 655 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-4gpu/attempt-5055039/device-evidence/worker-actor_rollout-2-223651.json": { + "sha256": "3f565bde33bbe381432aa553ab98d9a96033f7e946280ccc2d958a0c3a1608ef", + "size_bytes": 655 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-4gpu/attempt-5055039/device-evidence/worker-actor_rollout-3-223652.json": { + "sha256": "9dbfad63684dd6b9c8d4975b31b6e11ae2ddc4b5566bdb5665974e02591fef10", + "size_bytes": 655 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-4gpu/attempt-5055039/dispatch.json": { + "sha256": "e7d5b49fe00aca4ac25fe286e83ce98de47b4c8eef763b7a04708f4571af7f34", + "size_bytes": 4505 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-4gpu/attempt-5055039/environment-5055039.json": { + "sha256": "eeaa99b78ef7e9ff35d6eb0dc05b8ef398fae6e2c7b272fd6f13094bf2dae18a", + "size_bytes": 4898 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-4gpu/attempt-5055039/gpu-device-map.json": { + "sha256": "7559ab85f60fcd5254e2134eda144444bd97cf2e4a3796e449ae3450105d9ca3", + "size_bytes": 810 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-4gpu/attempt-5055039/gpu-memory-5055039.csv": { + "sha256": "b4b4e19dd46e05882f0eb6239d50081bf3fb41128dc9aaf69af544b59f2e6c3c", + "size_bytes": 52234 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-4gpu/attempt-5055039/gpu-telemetry-5055039.csv": { + "sha256": "757a749f5d82b9303b34e6183405244fe0aeb8dc806b0881680f4e5b629ea13b", + "size_bytes": 236803 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-4gpu/attempt-5055039/launcher.log": { + "sha256": "a056539d039a4b064daaffaf2666ebda418123f101f05830c8cdc804ead51466", + "size_bytes": 152029 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-4gpu/attempt-5055039/phase-memory-5055039.csv": { + "sha256": "81cf73993ba8398b64d63d49573d5f2ff00262f13fb0b682be8585b65c82cae3", + "size_bytes": 72559 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-4gpu/attempt-5055039/training-5055039.log": { + "sha256": "33e3c6f3b906271e7850fb9020806adc86d620e389c69be33935a783930dfa3a", + "size_bytes": 151865 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s151-4gpu/attempt-5055039/trial-5055039.json": { + "sha256": "20533f21f4494b5ea2cee41ea77ca60fa4e5e0100993924d2c8806ed88463f18", + "size_bytes": 1417 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-2gpu/attempt-5055040/device-evidence/probe-exit-idle.json": { + "sha256": "feee07cb7f76a70f68b03182ced69610f5d33a29c1527dbc5df8222c07b06ca1", + "size_bytes": 3689 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-2gpu/attempt-5055040/device-evidence/ray-resources.json": { + "sha256": "a1f8aa24e578f0039fb7b9941d269783e257642595cbc4ff0c02fba94317305d", + "size_bytes": 1558 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-2gpu/attempt-5055040/device-evidence/trainer-cuda.json": { + "sha256": "d9559a799fa49f3187916a1af9288c31c535858c83fbb10f6e3147c9a3898879", + "size_bytes": 574 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-2gpu/attempt-5055040/device-evidence/trainer-exec.json": { + "sha256": "561323fbd8b8535cf5418f098f71810fa10ac94ff2d8270afb9ca0cfe411d8a8", + "size_bytes": 366 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-2gpu/attempt-5055040/device-evidence/trainer-terminal.json": { + "sha256": "b8484860c422989e82224921da56c8d9e3a03ad092f2bd1901112036f519310b", + "size_bytes": 164 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-2gpu/attempt-5055040/device-evidence/worker-actor_rollout-0-2902450.json": { + "sha256": "3307be1a667009467159f69e201cad1b6e4beca18ba6d49450159300f1c9eb89", + "size_bytes": 656 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-2gpu/attempt-5055040/device-evidence/worker-actor_rollout-1-2902451.json": { + "sha256": "5f00a77643eceb402ebfdac0447aeaa43f52cc9061569782f7d7ab774295214b", + "size_bytes": 656 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-2gpu/attempt-5055040/dispatch.json": { + "sha256": "5b45b509a807fb6d79cf3d153cd88f037ffadbbcddafe34b1b07be5d2663dcce", + "size_bytes": 4323 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-2gpu/attempt-5055040/environment-5055040.json": { + "sha256": "40c6943fa4388e05558e340c6180cdd13649c60e0c5110a128f7ab5b3789f22e", + "size_bytes": 4416 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-2gpu/attempt-5055040/gpu-device-map.json": { + "sha256": "3495fd0696fd723901acb870a0b34177df4d4972f8312ab7ad7ce9d0220559fa", + "size_bytes": 470 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-2gpu/attempt-5055040/gpu-memory-5055040.csv": { + "sha256": "4163dbdf9f2443113abf7b578e013717cd0b220ad021c948cf5900e56704838f", + "size_bytes": 37007 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-2gpu/attempt-5055040/gpu-telemetry-5055040.csv": { + "sha256": "7548b88fe81f9776870b1d4725252c3e3a8d57b880f862beb68110b297a8e32d", + "size_bytes": 110110 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-2gpu/attempt-5055040/launcher.log": { + "sha256": "5d00afea2962b0949280254f4cc9f35a6f70a8d86d750e6a5ce089cd3e398f9a", + "size_bytes": 140971 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-2gpu/attempt-5055040/phase-memory-5055040.csv": { + "sha256": "22be30796ae66cc4170aa19b4e45feb57a43e09b08a2b2f691661b1915e9f9fa", + "size_bytes": 55862 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-2gpu/attempt-5055040/training-5055040.log": { + "sha256": "369f4d3b0f5a16b625bae5f70fb59c815661c5aaccc5aab4e436c45c6ac09908", + "size_bytes": 140807 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-2gpu/attempt-5055040/trial-5055040.json": { + "sha256": "722c6aa143f3d154e35b863f491771e5f90e1a104293600cfdec795cd064a239", + "size_bytes": 1417 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-4gpu/attempt-5055040/device-evidence/probe-exit-idle.json": { + "sha256": "2bea834a702244208d969f98a894452a750ce7e185afd6b2cb1929bbbc024459", + "size_bytes": 3689 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-4gpu/attempt-5055040/device-evidence/ray-resources.json": { + "sha256": "a7782c6d413819d743caf2110cff21144fda783d9ef9415b1b53eb282e33cbdf", + "size_bytes": 1558 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-4gpu/attempt-5055040/device-evidence/trainer-cuda.json": { + "sha256": "157d8b6700f0a64f914f3042df4a2c46d0c5200fb638375bf7e2cab19b7d0e72", + "size_bytes": 856 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-4gpu/attempt-5055040/device-evidence/trainer-exec.json": { + "sha256": "85b18cc608cf75bcdc9fee8718afa6b98ac3dcb54dce7eaf77b14328aa788f21", + "size_bytes": 448 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-4gpu/attempt-5055040/device-evidence/trainer-terminal.json": { + "sha256": "f7b4ca047049210b34e206639e5e6b47aba848f35ab68a2f1806e5105fe994d5", + "size_bytes": 164 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-4gpu/attempt-5055040/device-evidence/worker-actor_rollout-0-2887019.json": { + "sha256": "ac319652b009e95d853a593b78a6b59b5c3b30364264086e1c07504f88a1cfa3", + "size_bytes": 656 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-4gpu/attempt-5055040/device-evidence/worker-actor_rollout-1-2887020.json": { + "sha256": "b7b2b3c867a4105d8546c2f628a5ade8d80b51b3acb0f64a625619e75ad921ad", + "size_bytes": 656 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-4gpu/attempt-5055040/device-evidence/worker-actor_rollout-2-2887021.json": { + "sha256": "a37e4fd1723a1a62a55e60a44d05775750971f6ce6c7ccb6fe1444c00c01d26d", + "size_bytes": 656 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-4gpu/attempt-5055040/device-evidence/worker-actor_rollout-3-2887022.json": { + "sha256": "2f9f69f3c15fc8d6488ee3976b76dad55b7b0dd0de251f797326f0fd2833aa12", + "size_bytes": 656 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-4gpu/attempt-5055040/dispatch.json": { + "sha256": "d33647238742eedb44dba2d151f03383b6fe1214544214e43bc94ee7cac553ed", + "size_bytes": 4505 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-4gpu/attempt-5055040/environment-5055040.json": { + "sha256": "a70d7d101c1811df6f942a4e0459ed2d5ace97143b31a1fe24e6a56be88d5922", + "size_bytes": 4898 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-4gpu/attempt-5055040/gpu-device-map.json": { + "sha256": "4e003ac25bcb095a3c7e55a45bf816f6bf0351a567d17d1b503d8c9aa4ba00ce", + "size_bytes": 810 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-4gpu/attempt-5055040/gpu-memory-5055040.csv": { + "sha256": "7d3bc7fa380d4544cf96c56437476191a8c9e852b1413b08a785ae4bc2b02e02", + "size_bytes": 49029 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-4gpu/attempt-5055040/gpu-telemetry-5055040.csv": { + "sha256": "80203b36d91d12ebf912cde8acdf05acea63d7b4a29b6f824f7d64692dd4e751", + "size_bytes": 221146 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-4gpu/attempt-5055040/launcher.log": { + "sha256": "b943caad1f40f41905f322485febe7aa8c7764752f70042270409ecbda57348e", + "size_bytes": 155846 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-4gpu/attempt-5055040/phase-memory-5055040.csv": { + "sha256": "b2291867ff38d0a752bf73c9536bf61b8991221e3c600a4b73a824124dac0e7d", + "size_bytes": 67974 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-4gpu/attempt-5055040/training-5055040.log": { + "sha256": "e6e2ca38ce707c7e488c5d5aa8377ba53a4cbe557e3f888efa7ad9005ad7ee6a", + "size_bytes": 155682 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s152-4gpu/attempt-5055040/trial-5055040.json": { + "sha256": "04f5c2ca8313cba072da72603637da301d1b6ca326429b5e3252e220264d6396", + "size_bytes": 1417 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-2gpu/attempt-5054952/device-evidence/probe-exit-idle.json": { + "sha256": "d7c8c0fbd4376432e73b6c8759bc44c2eea73fc84f269d4f8c2f658dcd8f1daa", + "size_bytes": 3689 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-2gpu/attempt-5054952/device-evidence/ray-resources.json": { + "sha256": "403b9de0f449cb25dd5d35dce27083db4f0ea173aa3c7ff72db8bf80e87bbb7e", + "size_bytes": 1558 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-2gpu/attempt-5054952/device-evidence/trainer-cuda.json": { + "sha256": "50c07928d6b0bac5f2e08870c94af2bc53e6a069a63846efb16928c4a8ffa0dd", + "size_bytes": 574 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-2gpu/attempt-5054952/device-evidence/trainer-exec.json": { + "sha256": "b1f85bb98ab0ba953cb1204cf08eb0b2f8fab1d05079c7af16aba51ac1d996af", + "size_bytes": 366 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-2gpu/attempt-5054952/device-evidence/trainer-terminal.json": { + "sha256": "f3b7ee842055fa434b6ea1e099bde1a40635180c7950924fca6ba450c129d724", + "size_bytes": 164 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-2gpu/attempt-5054952/device-evidence/worker-actor_rollout-0-2760700.json": { + "sha256": "0d4d894ac3bf5d9bd8ee6679d9821605e4c0e5c188a13f965388748938cd3f94", + "size_bytes": 656 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-2gpu/attempt-5054952/device-evidence/worker-actor_rollout-1-2760701.json": { + "sha256": "ce131b01419e22cb02e5a85346400046f0ed36865499b757ffc3408a6dd5ee19", + "size_bytes": 656 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-2gpu/attempt-5054952/dispatch.json": { + "sha256": "8043b4b55eece61862ba81169400a18661ff47853c5e653ceb6a92ce53162afa", + "size_bytes": 4323 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-2gpu/attempt-5054952/environment-5054952.json": { + "sha256": "afaa0d70abf3cde2a8c8b9d8d3c38c67dbb1309f7f165b3c9f57788826571d41", + "size_bytes": 4415 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-2gpu/attempt-5054952/gpu-device-map.json": { + "sha256": "bad13271331f2ea11b13d98b4452d396f53c8d430074a9af291b7777d3873af0", + "size_bytes": 470 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-2gpu/attempt-5054952/gpu-memory-5054952.csv": { + "sha256": "5b5f6997e624d8af915c62b812cbbde59c81a65b0eba62382017516ea5261ac0", + "size_bytes": 45872 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-2gpu/attempt-5054952/gpu-telemetry-5054952.csv": { + "sha256": "774662db1ec7c689f3f6bd87dd5f23f77cdf83716c9d3215fe73ec95b435e926", + "size_bytes": 137383 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-2gpu/attempt-5054952/launcher.log": { + "sha256": "863901fa900c31e06f19d77714eeae085f8c8ebb503162931708fdcd7ea39970", + "size_bytes": 152621 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-2gpu/attempt-5054952/phase-memory-5054952.csv": { + "sha256": "d18bd122fa26508a60eb8a326f92a017bde7c74ae9da3a050958bb282be4385b", + "size_bytes": 69497 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-2gpu/attempt-5054952/training-5054952.log": { + "sha256": "85abdbcd12fefe6edf252af0497e53e5d54f2cea23662ea91176b9dc1a0197cf", + "size_bytes": 152457 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-2gpu/attempt-5054952/trial-5054952.json": { + "sha256": "85386c62174686d5201c0834cad0d14e06ecf8d7ed1ae7d71cc8f05ee3f77c8c", + "size_bytes": 1417 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-4gpu/attempt-5054952/device-evidence/probe-exit-idle.json": { + "sha256": "685ed87ce4d077f99653c7c53f90a98655da0c12e7e61decffe4f6092cf6e71d", + "size_bytes": 3689 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-4gpu/attempt-5054952/device-evidence/ray-resources.json": { + "sha256": "44ac9a3817f8d2f976d969f91cddc30685db79ede4cef5d4c7a7d24441463f71", + "size_bytes": 1558 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-4gpu/attempt-5054952/device-evidence/trainer-cuda.json": { + "sha256": "da98dea5b322620b2d2d73aa8e565a6b3e2e5580d28a0f7df491eb63eac78895", + "size_bytes": 856 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-4gpu/attempt-5054952/device-evidence/trainer-exec.json": { + "sha256": "3bd0a6d86437317bf3c94814fad2db0992f31e094adaec7eeeb9b5a07f4e2dae", + "size_bytes": 448 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-4gpu/attempt-5054952/device-evidence/trainer-terminal.json": { + "sha256": "8104be04acdedd4d36e862784fce4690661c6c219d2c8da604be1f70a6c2d6a6", + "size_bytes": 164 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-4gpu/attempt-5054952/device-evidence/worker-actor_rollout-0-2788660.json": { + "sha256": "d67d94856b298c4f5454eb84b109f84900a847a9bc3f541af8a6e916c5cc2eb3", + "size_bytes": 656 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-4gpu/attempt-5054952/device-evidence/worker-actor_rollout-1-2788661.json": { + "sha256": "10e2d8192a8dcc7a2d994dcebf5f60c358a1ec57518383ae6cffff528877a2aa", + "size_bytes": 656 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-4gpu/attempt-5054952/device-evidence/worker-actor_rollout-2-2788662.json": { + "sha256": "d160db7fd34b20bf355b9c65ffc1d796a3c3adc95a83752a2fe413a13d67e5c5", + "size_bytes": 656 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-4gpu/attempt-5054952/device-evidence/worker-actor_rollout-3-2788663.json": { + "sha256": "6e466baf7e0927b9e1688f1d338e981bdea6b8d0dece90fa86570fd129c42b9d", + "size_bytes": 656 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-4gpu/attempt-5054952/dispatch.json": { + "sha256": "8d8d4416de398f8faecb78cfabb5b3430b706f33e6f763b7664363926ffe04b6", + "size_bytes": 4505 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-4gpu/attempt-5054952/environment-5054952.json": { + "sha256": "3e2532255afcffeed84584822d74c424d299bcc70fe81cc45ec3abea5be8122d", + "size_bytes": 4897 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-4gpu/attempt-5054952/gpu-device-map.json": { + "sha256": "345d3ef06279fe77e65068ccecf5bac8f1604fc79803196594fe00054b9dace1", + "size_bytes": 810 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-4gpu/attempt-5054952/gpu-memory-5054952.csv": { + "sha256": "265e77f8d181c60fcc525dc4fcddac206ee733dcac679d9f85a7ed6457a5af31", + "size_bytes": 49132 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-4gpu/attempt-5054952/gpu-telemetry-5054952.csv": { + "sha256": "2cf282e3cf99d9c2a36b305e5a1860e0fe8616bbed7e3d7202df7957c0f7ef9c", + "size_bytes": 221504 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-4gpu/attempt-5054952/launcher.log": { + "sha256": "ef1c1c80ed3b0cc41e0ff28038a6c05e1b44b004fe40e07b895c95b299f30d2e", + "size_bytes": 155733 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-4gpu/attempt-5054952/phase-memory-5054952.csv": { + "sha256": "2e5a994f775670b9576219310f5bb2e68fdd694fbc609aba2c907864697d0701", + "size_bytes": 68107 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-4gpu/attempt-5054952/training-5054952.log": { + "sha256": "2147efcf62abd483d7078a1452c2dbc5049741f0826a3531ecc84e95692c26d5", + "size_bytes": 155569 + }, + "output/prospective_estimation/est-qwen25-7b-gsm8k-c3-s153-4gpu/attempt-5054952/trial-5054952.json": { + "sha256": "dd7a44d1166aff42c546442053606e573bb6692367116507ce2ea5ab5079c2ec", + "size_bytes": 1417 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s151-2gpu/attempt-5054345/device-evidence/trainer-cuda.json": { + "sha256": "90e16c69bc181a9009833b3ddd8160d5a592d64bc97c52918b048924f078e871", + "size_bytes": 571 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s151-2gpu/attempt-5054345/dispatch.json": { + "sha256": "2345925c10892031a4cc309320d3e2643c0acd247d537f7a1acc71160c7ae84d", + "size_bytes": 4309 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s151-2gpu/attempt-5054345/environment-5054345.json": { + "sha256": "b51954a2607845c3cce4463635348d68f0f1132fb758ea0d6ccfaafe2f4d9701", + "size_bytes": 4411 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s151-2gpu/attempt-5054345/gpu-device-map.json": { + "sha256": "7a4216ea83e6a6756cb22c5262d380e288af353ed6510b5a09902433ea47225b", + "size_bytes": 469 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s151-2gpu/attempt-5054345/gpu-memory-5054345.csv": { + "sha256": "1d47fc8ba3e2d468f3c70d188990359d9db1a587e15b13ccf2cdf094d6deaba5", + "size_bytes": 1260 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s151-2gpu/attempt-5054345/gpu-telemetry-5054345.csv": { + "sha256": "87a2440969988c477592b8a48d99f6d6bf0aa11782f5244d3abd20e0c278db6f", + "size_bytes": 3940 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s151-2gpu/attempt-5054345/launcher.log": { + "sha256": "f57a87c23afdfcc37de4193c2ac128cb8e5b1a003b49c849a09381c05c269136", + "size_bytes": 258 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s151-2gpu/attempt-5054345/phase-memory-5054345.csv": { + "sha256": "f530f6baf12d8f31b5d4b73cc7815f498c6e48da68f6b5547f5bda595e845d93", + "size_bytes": 1935 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s151-2gpu/attempt-5054345/training-5054345.log": { + "sha256": "eac54c43cc7f6f3e8e3600f74443e9e8a61bd69e2f995f6628a777e008ed0bdd", + "size_bytes": 95 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s151-4gpu/attempt-5054345/device-evidence/probe-exit-idle.json": { + "sha256": "9461184cd58c9a70e3bd729e7711d65d93a1f9a3a72f00bce0ffc2e01eae62b5", + "size_bytes": 3686 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s151-4gpu/attempt-5054345/device-evidence/trainer-cuda.json": { + "sha256": "a89d064ad40a6b68bf2315cfc8c551a1c62d7d63235a6f7de84fab2ad23307e0", + "size_bytes": 853 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s151-4gpu/attempt-5054345/device-evidence/trainer-exec.json": { + "sha256": "46bd8136afdc19421c361a385359ea9185374d631051f31b018a39b475897765", + "size_bytes": 446 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s151-4gpu/attempt-5054345/device-evidence/trainer-terminal.json": { + "sha256": "d48a2803b8f633a538050fc486e04f0710bf20a061d5131b442bee73f648ee72", + "size_bytes": 162 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s151-4gpu/attempt-5054345/dispatch.json": { + "sha256": "4b077647f91142d2e99676ea21982990e7d55cd7745d815e2b093f4813d67526", + "size_bytes": 4491 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s151-4gpu/attempt-5054345/environment-5054345.json": { + "sha256": "f63005c8a5b9dc5a19e38b602e8ba4ff1a17749916689f2d629b2ae9f9c5a79a", + "size_bytes": 4893 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s151-4gpu/attempt-5054345/gpu-device-map.json": { + "sha256": "3a6325e56d84a0132f2962fca19be36779c10f9de471c6929e4fcde88af3a133", + "size_bytes": 809 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s151-4gpu/attempt-5054345/gpu-memory-5054345.csv": { + "sha256": "dcb3012335a99a5d21820ef66818228848bb412292ff4e1b610f2dada9f8caa2", + "size_bytes": 8820 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s151-4gpu/attempt-5054345/gpu-telemetry-5054345.csv": { + "sha256": "e799d4f61172b95a161d4b1661acbbeb54cd913b41a59dace0502ae7c7f1ad50", + "size_bytes": 42210 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s151-4gpu/attempt-5054345/launcher.log": { + "sha256": "307f1dc04dde6c3fdca42369c832e7079469183ddc4d7ef1736fa54392e654e7", + "size_bytes": 10455 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s151-4gpu/attempt-5054345/phase-memory-5054345.csv": { + "sha256": "384e2104d66b97c634d8fca42e1be39e86ee6f7e40b76395d13478bd3c66b97e", + "size_bytes": 12495 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s151-4gpu/attempt-5054345/training-5054345.log": { + "sha256": "1ff7787dc3f88e6f80d318949ce48c33619476d4bedd24b138669b2e673313a5", + "size_bytes": 10292 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s151-4gpu/attempt-5054345/trial-5054345.json": { + "sha256": "9fa49106720cb9aca024b7b033413838d7f492d7f2df414e2729969ed1a42878", + "size_bytes": 1406 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-2gpu/attempt-5054998/device-evidence/probe-exit-idle.json": { + "sha256": "b99eb0d22eab6d453b09c81a4a9292ae4bfa8df33987c7357a978799e1438dbe", + "size_bytes": 3686 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-2gpu/attempt-5054998/device-evidence/ray-resources.json": { + "sha256": "9bd8b9a2b86c0aa0e2214bdf0459e3f0fb0366e4464d70c1a66b05b4354fcb3a", + "size_bytes": 1554 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-2gpu/attempt-5054998/device-evidence/trainer-cuda.json": { + "sha256": "6e0f8f609672add0a13fd987e36a64c632e433cfea7c96effb6a4fe9379afc29", + "size_bytes": 571 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-2gpu/attempt-5054998/device-evidence/trainer-exec.json": { + "sha256": "3eee4e55c1ac1b6f55e6039bf86de7d35cabeeed5f296592b07d6a7d0a2abec5", + "size_bytes": 364 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-2gpu/attempt-5054998/device-evidence/trainer-terminal.json": { + "sha256": "4fcaf8558ecc49ffc5884d8742c07288d0656437104ddbb9d18c459dc8b6524b", + "size_bytes": 162 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-2gpu/attempt-5054998/device-evidence/worker-actor_rollout-0-451980.json": { + "sha256": "f46a96671843faae2bacfa5801f211f9c077a3f4c74a09136bfb35a4d6ad45f3", + "size_bytes": 654 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-2gpu/attempt-5054998/device-evidence/worker-actor_rollout-1-451981.json": { + "sha256": "788b19e134eea78e486fd3575aef5df18d67c8fae97da0dca293feeb99346f04", + "size_bytes": 654 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-2gpu/attempt-5054998/dispatch.json": { + "sha256": "8feb16464f0d2ca28f3b541361ca7f3270f23b53e8fe1283e44bd597c9da4512", + "size_bytes": 4309 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-2gpu/attempt-5054998/environment-5054998.json": { + "sha256": "34f0ffa07a9e7e7897548aafa9aa92b406331dad4b0ee6a8f8def18a4a60046d", + "size_bytes": 4411 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-2gpu/attempt-5054998/gpu-device-map.json": { + "sha256": "66fb2bfc97ae0fb2fff38f57d1655e69c2d4ff87e566a407448b1191667c6642", + "size_bytes": 469 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-2gpu/attempt-5054998/gpu-memory-5054998.csv": { + "sha256": "0aa37086ab7e9070b081d5aa3bc91b09046ebd89b1b7bb5d36754b0b24ac5882", + "size_bytes": 39421 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-2gpu/attempt-5054998/gpu-telemetry-5054998.csv": { + "sha256": "79419c7b608679323a25f9eb9bc3bd841c9a4cbaa10ac0e10c4a043ffec15d39", + "size_bytes": 117551 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-2gpu/attempt-5054998/launcher.log": { + "sha256": "e1c5cafb971c65d4f32d5d2100fe963ed8bbdb582f2a8af8809c4d2d4356f882", + "size_bytes": 150618 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-2gpu/attempt-5054998/phase-memory-5054998.csv": { + "sha256": "4c6841f02bb6e70398bdc4e2962e740607f761b67c49e19c4ed3c684064d5336", + "size_bytes": 59581 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-2gpu/attempt-5054998/training-5054998.log": { + "sha256": "d7e24d5d98c6bff45277e89937c759eea50f7d3c7fc8549f4936ff92ed12d5e7", + "size_bytes": 150455 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-2gpu/attempt-5054998/trial-5054998.json": { + "sha256": "8f6495ad693eece376505beb764beb1a5ffac9dd68245a3c1e07117c51c8323f", + "size_bytes": 1409 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-4gpu/attempt-5054998/device-evidence/probe-exit-idle.json": { + "sha256": "480897e3332992290eb1bfe8cfb14cb1e618fe9e07be293ca8d9da342dda2e1b", + "size_bytes": 3686 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-4gpu/attempt-5054998/device-evidence/ray-resources.json": { + "sha256": "fbcdf42ed79b27a2fa4876f4c64600f61d8cdc3447a5c9a1544abc0c255ba0d3", + "size_bytes": 1554 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-4gpu/attempt-5054998/device-evidence/trainer-cuda.json": { + "sha256": "068ceaf1f12e7cf0baab78193e53f33ebd120aaa69d5e67032d43bfdc49ed673", + "size_bytes": 853 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-4gpu/attempt-5054998/device-evidence/trainer-exec.json": { + "sha256": "ac50779ae57691b01213b50b92a54b14d0ca75b52476e3a8cedd3c01a8c1de26", + "size_bytes": 446 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-4gpu/attempt-5054998/device-evidence/trainer-terminal.json": { + "sha256": "b546a9a74658f0041d3902c6c47c969cb2f371d477578bbf1ac991e7fd7b8b32", + "size_bytes": 162 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-4gpu/attempt-5054998/device-evidence/worker-actor_rollout-0-468090.json": { + "sha256": "8203cdbd03b5128e28f98377ca3a85d1cf33af2b77875a4dee1d9ca75f1a6e1c", + "size_bytes": 654 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-4gpu/attempt-5054998/device-evidence/worker-actor_rollout-1-468091.json": { + "sha256": "ae9eb18efaada3d17e6bcb30b784311126c4ac644d6eabd3a44f01dce8f96d1f", + "size_bytes": 654 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-4gpu/attempt-5054998/device-evidence/worker-actor_rollout-2-468092.json": { + "sha256": "f7bda721987a42665b53210a77fe64ba1681a5c5e106724c4096b63884939b13", + "size_bytes": 654 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-4gpu/attempt-5054998/device-evidence/worker-actor_rollout-3-468093.json": { + "sha256": "77fc128d253218b05955d2629d59c76c96ee7b6972388b5bed0498958ba97fc6", + "size_bytes": 654 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-4gpu/attempt-5054998/dispatch.json": { + "sha256": "70a85f75536c2d081ffdfff4060c74c8d12716a6da6c0a0b7e8ffd3c080a249e", + "size_bytes": 4491 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-4gpu/attempt-5054998/environment-5054998.json": { + "sha256": "f6b532622d557861a86f4d020dd8c90a2aef1a2b914236e37f2bf5c4f84aee04", + "size_bytes": 4893 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-4gpu/attempt-5054998/gpu-device-map.json": { + "sha256": "ba38781c7d65e96531ad2145a1ac09d048655521309b744f31f0aeff333713f4", + "size_bytes": 809 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-4gpu/attempt-5054998/gpu-memory-5054998.csv": { + "sha256": "f24cade344c4d736fb5c189db544559e44a297d8ac224219d96e91dd70b545d5", + "size_bytes": 120081 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-4gpu/attempt-5054998/gpu-telemetry-5054998.csv": { + "sha256": "2ee05e876c95ff50fbdab8a5d2f8238a4f798938d12a6e4eb21c204fb3de4e37", + "size_bytes": 514727 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-4gpu/attempt-5054998/launcher.log": { + "sha256": "8769792fe6f87d61b318ff8c3d47278a8a6d6369aaf6d44c53fd898e74882983", + "size_bytes": 155006 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-4gpu/attempt-5054998/phase-memory-5054998.csv": { + "sha256": "516f8f2d677a0cdefe574773590f06f5c37140858ea64def3efc16e1089fe225", + "size_bytes": 163326 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-4gpu/attempt-5054998/training-5054998.log": { + "sha256": "f08ddac8dcede8a9cfb583001419cd35efb23bbe0f37602023505e23fcecf89f", + "size_bytes": 150615 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s152-4gpu/attempt-5054998/trial-5054998.json": { + "sha256": "ad7680af13190276e3f550de3e21e20565290b233aa0fd9b9329da07148d8430", + "size_bytes": 1411 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s153-2gpu/attempt-5054339/device-evidence/probe-exit-idle.json": { + "sha256": "9556b7a32f70dfe5fc0fdee601a63f24d377c8f901f55704b16c64b9b37c40f7", + "size_bytes": 3686 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s153-2gpu/attempt-5054339/device-evidence/trainer-cuda.json": { + "sha256": "627f3a672aa5f9906b1a1879ca51d295189df78beae66ced55afa584d265d54e", + "size_bytes": 571 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s153-2gpu/attempt-5054339/device-evidence/trainer-exec.json": { + "sha256": "0512a053e74e7deb268d5b79ac28359f4ccd03ed49cad60a219d007147fdc4ef", + "size_bytes": 364 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s153-2gpu/attempt-5054339/device-evidence/trainer-terminal.json": { + "sha256": "255b902ec26f18f7aac5a48ed6810647a5fc96ca619ada351cfb27d1bc30b0de", + "size_bytes": 162 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s153-2gpu/attempt-5054339/dispatch.json": { + "sha256": "f443e29fcaca20845c206cd1e75f01fda8e9309021558543e68d3602479104d6", + "size_bytes": 4309 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s153-2gpu/attempt-5054339/environment-5054339.json": { + "sha256": "85536a219980d06af0fb3e414fef1e08d77340315cb4b1fcb21400f3e4dd379f", + "size_bytes": 4411 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s153-2gpu/attempt-5054339/gpu-device-map.json": { + "sha256": "c23a51963967ba1c56250dfefbe360f84038ccb2a340ed13f9fd82e65e8c92c5", + "size_bytes": 469 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s153-2gpu/attempt-5054339/gpu-memory-5054339.csv": { + "sha256": "67ad25e15c2c36498138738f5ed66a4b8abfa5a7ae7efcf1b47a10c2c328e2fd", + "size_bytes": 8568 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s153-2gpu/attempt-5054339/gpu-telemetry-5054339.csv": { + "sha256": "39d1bc0c68fd85d7e8771e422d3e21ea25c19f0c59c31d2bfd173796ce6171e7", + "size_bytes": 26386 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s153-2gpu/attempt-5054339/launcher.log": { + "sha256": "7e584aaf5cf66367ff6f878235c2036849c9d9c680bfc87eb6b7e7a4e58e27f8", + "size_bytes": 10455 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s153-2gpu/attempt-5054339/phase-memory-5054339.csv": { + "sha256": "e9183963a2da9aa0cc3ef666762ebb23dfb65ef51377115225b8b4c97fa3337d", + "size_bytes": 13158 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s153-2gpu/attempt-5054339/training-5054339.log": { + "sha256": "a1f531d3332468e2f18de8be8d4a95fd63c97c85f68b45fa1a1a7bb418d74d1b", + "size_bytes": 10292 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s153-2gpu/attempt-5054339/trial-5054339.json": { + "sha256": "9e13c1af0b908f0c2ea40afc9a2e40e057c5d6e699022896f7a0c9ca294d1e4d", + "size_bytes": 1406 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s153-4gpu/attempt-5054339/device-evidence/probe-exit-idle.json": { + "sha256": "85bb241b33744fc554a9470d27ebd8997d8d92a0c7b3538c46f44a6e0dec9f2b", + "size_bytes": 3686 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s153-4gpu/attempt-5054339/device-evidence/trainer-cuda.json": { + "sha256": "8fef48b328cd2dbeae005c2a2a631d1a41c22a13d4ef818a4d9354f28f598279", + "size_bytes": 853 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s153-4gpu/attempt-5054339/device-evidence/trainer-exec.json": { + "sha256": "830f16c01399d463901bf46346f7d056c6445e6d162740725a788cf95fc81f1e", + "size_bytes": 446 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s153-4gpu/attempt-5054339/device-evidence/trainer-terminal.json": { + "sha256": "0cb7630274d2962c1437aecc687238e7535d32dad0bf593893a3445c2328a39d", + "size_bytes": 162 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s153-4gpu/attempt-5054339/dispatch.json": { + "sha256": "a4f5709aca2c38d8b134e11d1b9a9a3e5c87874724e91676e40e03dabdae8a24", + "size_bytes": 4491 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s153-4gpu/attempt-5054339/environment-5054339.json": { + "sha256": "d6fa5cbdb4fb4461ae7d1ed3ec0f70c1fdb13a641e469cb60df47e49fc6dedb8", + "size_bytes": 4893 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s153-4gpu/attempt-5054339/gpu-device-map.json": { + "sha256": "cef4624a72f61ccea59349b6643641f674f07f617ac12af67e9d125e66d41e25", + "size_bytes": 809 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s153-4gpu/attempt-5054339/gpu-memory-5054339.csv": { + "sha256": "abc3721de8b9d42a4a628edfc908136493eeb983f55c201e0724cf8e9455c77a", + "size_bytes": 8676 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s153-4gpu/attempt-5054339/gpu-telemetry-5054339.csv": { + "sha256": "d3c90732ec74ee2fd0c15c57a9de5da8a688a831335443b8804bea860b8d9e17", + "size_bytes": 41522 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s153-4gpu/attempt-5054339/launcher.log": { + "sha256": "0286464a61f5b44cf1fac9a343e0a2c7c720aea31c0d5a3d27483f93a9b6abbf", + "size_bytes": 10455 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s153-4gpu/attempt-5054339/phase-memory-5054339.csv": { + "sha256": "80e29952e1e78ec8cf2dfdd16622001329ea334e853cf464c45ff3858c77e5ef", + "size_bytes": 12291 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s153-4gpu/attempt-5054339/training-5054339.log": { + "sha256": "3d9523bc5029e605006be55f02104673f3659f2b203ea391b8167ccccb1fde40", + "size_bytes": 10292 + }, + "output/prospective_estimation/est-qwen25-7b-math-c2-s153-4gpu/attempt-5054339/trial-5054339.json": { + "sha256": "7d682fa15fb1c4d17716c268f3325d56e9aec006e42af1f780428e9a59a04d74", + "size_bytes": 1406 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-2gpu/attempt-5054953/device-evidence/probe-exit-idle.json": { + "sha256": "dc080bb6455b5c27d0eda6020d2a958af63da0d47428e52292a801e191669c9d", + "size_bytes": 3688 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-2gpu/attempt-5054953/device-evidence/ray-resources.json": { + "sha256": "c8f7557d827168a4274f5b704773c8a0dd0ef1a5b0f3a6c513e80128c0844fca", + "size_bytes": 1557 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-2gpu/attempt-5054953/device-evidence/trainer-cuda.json": { + "sha256": "728df0a70dc279a477ec99b3589065df079d8d653bb070537fa56eb1c2ee4497", + "size_bytes": 573 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-2gpu/attempt-5054953/device-evidence/trainer-exec.json": { + "sha256": "d6f94c33136dae78f148d76906efcbd5463c37aa8f95d7d7b3a46d09797b7ea5", + "size_bytes": 365 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-2gpu/attempt-5054953/device-evidence/trainer-terminal.json": { + "sha256": "11a073586b2c2af528e846f6ea30c4a0fda256fee7b617ad66cd9a2b88efe094", + "size_bytes": 163 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-2gpu/attempt-5054953/device-evidence/worker-actor_rollout-0-2335597.json": { + "sha256": "7f15cd1128e8efae260d741824058b6ecde6b2576cdd4799a59bcf8acc388eef", + "size_bytes": 655 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-2gpu/attempt-5054953/device-evidence/worker-actor_rollout-1-2335598.json": { + "sha256": "ab1818f63a85fc426235e9ff2fd7d23731c94269d817602a61346bae95bf47d7", + "size_bytes": 655 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-2gpu/attempt-5054953/dispatch.json": { + "sha256": "d321cbad33c38be83a730bf724ea8d93f8288d2c15efef21644c7a29ee3eaf30", + "size_bytes": 4311 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-2gpu/attempt-5054953/environment-5054953.json": { + "sha256": "5872fc717498756fb30975e06897bb997113a76c8abe72a4c8628db7a8fee9ff", + "size_bytes": 4411 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-2gpu/attempt-5054953/gpu-device-map.json": { + "sha256": "9c55175ad610ff22efdd27a61e9ced07b78da3a8ad3845d3e58a947fe66b9df5", + "size_bytes": 469 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-2gpu/attempt-5054953/gpu-memory-5054953.csv": { + "sha256": "3e849fcc9be3487d04e31f92eee40c102e3be2f5dfaa69e32d83fab4f760a02f", + "size_bytes": 35549 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-2gpu/attempt-5054953/gpu-telemetry-5054953.csv": { + "sha256": "5a85bcec3e90bbcc73bc05984ed29437e883d1bca15d9dfd1ea0728d9f3c1e07", + "size_bytes": 105743 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-2gpu/attempt-5054953/launcher.log": { + "sha256": "446e29d3a8fe336a7649febe5293bde8b8ffd6b2760cd32057882da5ef03c5b1", + "size_bytes": 152309 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-2gpu/attempt-5054953/phase-memory-5054953.csv": { + "sha256": "659c30c780578bc889aa3042ae3cd140fe169b8922dbd4234cb7e362e07a6639", + "size_bytes": 53654 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-2gpu/attempt-5054953/training-5054953.log": { + "sha256": "4e8f15a53467228b73de03ef2745d9f9ed7e8f9ad8630f7293eb500b32d683f4", + "size_bytes": 152146 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-2gpu/attempt-5054953/trial-5054953.json": { + "sha256": "12c875a5f2a690b2051c155437768272d20eefa85ddbdc74f397cd21a03e0c70", + "size_bytes": 1410 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-4gpu/attempt-5054953/device-evidence/probe-exit-idle.json": { + "sha256": "1b68671ff81fd691046739262b384d0639678fca68575a84c33e347882337395", + "size_bytes": 3688 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-4gpu/attempt-5054953/device-evidence/ray-resources.json": { + "sha256": "9aa27652f7589676c6cf38a4d294d6e5aed1c4560f5091e17c74849acb2eed25", + "size_bytes": 1557 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-4gpu/attempt-5054953/device-evidence/trainer-cuda.json": { + "sha256": "75f045bae702f5c4ed4bb19b9eb664122c9cbbc15428e6d4198b4d0393536c87", + "size_bytes": 855 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-4gpu/attempt-5054953/device-evidence/trainer-exec.json": { + "sha256": "a16ec95a7ba4dc7c134bf14383f5a19db90ffaa13f13fedb77d371e9b8432915", + "size_bytes": 447 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-4gpu/attempt-5054953/device-evidence/trainer-terminal.json": { + "sha256": "b9953b5d0c5b6eb7f4f64bb222ad77c22e65086c47e368b3d48602866d747675", + "size_bytes": 163 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-4gpu/attempt-5054953/device-evidence/worker-actor_rollout-0-2319770.json": { + "sha256": "33195e5fdb45b6905f7a786aef09677f688b4afd699d57b0ed859be03a98e750", + "size_bytes": 655 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-4gpu/attempt-5054953/device-evidence/worker-actor_rollout-1-2319772.json": { + "sha256": "29b75cca43746c50b361dea870f2c96c74c72452b415337257d027cc84caca60", + "size_bytes": 655 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-4gpu/attempt-5054953/device-evidence/worker-actor_rollout-2-2319776.json": { + "sha256": "0d60c15ce3da29ace1eb5323a300bcde899cc8da98383d75665f96fa319f8539", + "size_bytes": 655 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-4gpu/attempt-5054953/device-evidence/worker-actor_rollout-3-2319777.json": { + "sha256": "d94a7dd302bf69ebee7df1c9532e48d89832e198c23f2f22fe4fd5bd6eba4e62", + "size_bytes": 655 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-4gpu/attempt-5054953/dispatch.json": { + "sha256": "de01f7d0e1e75cdbab5e13f72ed1b8c62709a7727eb421eae30cfeb89247790e", + "size_bytes": 4493 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-4gpu/attempt-5054953/environment-5054953.json": { + "sha256": "f6d71510fb687c6b20af9baf45dbe9849905c249f74b5b65749e50a8453df4f2", + "size_bytes": 4893 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-4gpu/attempt-5054953/gpu-device-map.json": { + "sha256": "1095177515b19f8f7400445cbd4f3f2626b687a02ef62865930d79a4985061e1", + "size_bytes": 809 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-4gpu/attempt-5054953/gpu-memory-5054953.csv": { + "sha256": "3f93f3e58278d0f6ddf75cce3b5b23b8efd6dba7cb3aa63e89fd5ffa4ac89e11", + "size_bytes": 60617 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-4gpu/attempt-5054953/gpu-telemetry-5054953.csv": { + "sha256": "f208c89d404a576caa68f1178dddb8900490f8fb23073c079e41a02cc4d92ff8", + "size_bytes": 276311 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-4gpu/attempt-5054953/launcher.log": { + "sha256": "5f8861e1eae0f29616ebeb6e2d1b3949fd61dc935fb1ce01e1dcb411adc04014", + "size_bytes": 154767 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-4gpu/attempt-5054953/phase-memory-5054953.csv": { + "sha256": "d66080b583636b67e2a6018c1726c8620000a860de3b216bb508fd54f0cbc0e6", + "size_bytes": 84377 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-4gpu/attempt-5054953/training-5054953.log": { + "sha256": "c9194693783b900807f42728d2576d93194397371168f7cd2a935501ba8d533b", + "size_bytes": 154604 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s151-4gpu/attempt-5054953/trial-5054953.json": { + "sha256": "b21bf8eb7e0c6fd7c674192106dc31137667fc5eb9e9ee9998a528ab53180ac9", + "size_bytes": 1410 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-2gpu/attempt-5055042/device-evidence/probe-exit-idle.json": { + "sha256": "3a5a01204c8543baa9e97497197a61d568dc803091b25526a01f154ccd46b831", + "size_bytes": 3686 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-2gpu/attempt-5055042/device-evidence/ray-resources.json": { + "sha256": "9094ac4d4c1b98478a0cc1af1ce741ccc57f5ebc32b52853ec20335507c34b1a", + "size_bytes": 1550 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-2gpu/attempt-5055042/device-evidence/trainer-cuda.json": { + "sha256": "4e710b39490af687bea7ef5d00e1295c67a57e67f3c0242c7d34849ddd06808c", + "size_bytes": 571 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-2gpu/attempt-5055042/device-evidence/trainer-exec.json": { + "sha256": "59ff0e9d2cc60581af8b691256193958fee52dc1616e96637e19b9513ca43af0", + "size_bytes": 364 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-2gpu/attempt-5055042/device-evidence/trainer-terminal.json": { + "sha256": "6504c3731f00a06446447d9037f55d0e68098fb98d60da8f54b2c5d635ecc1dd", + "size_bytes": 162 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-2gpu/attempt-5055042/device-evidence/worker-actor_rollout-0-256435.json": { + "sha256": "cbee17ce5e575b9016a475ac4dabeda3595a32d3c433fee2a46101c6db061144", + "size_bytes": 654 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-2gpu/attempt-5055042/device-evidence/worker-actor_rollout-1-256436.json": { + "sha256": "9d93ab01540a69389e859485a551484fc620b2425bc8de62cba5caddd62b107f", + "size_bytes": 654 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-2gpu/attempt-5055042/dispatch.json": { + "sha256": "a9191b57fb9c325f8cc4e946bf649a509544cbb4d992e3eb6cb22f73de435393", + "size_bytes": 4311 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-2gpu/attempt-5055042/environment-5055042.json": { + "sha256": "4a90b0b811ac3f504daa59ea0140f3313590fc3240440e7497f86f3a00068181", + "size_bytes": 4412 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-2gpu/attempt-5055042/gpu-device-map.json": { + "sha256": "0235428df2fd0de8018643234456f36ad2d6ec351c273e32225b8d1595c91275", + "size_bytes": 469 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-2gpu/attempt-5055042/gpu-memory-5055042.csv": { + "sha256": "b64fd681c3a6c1c97c4126b0df0848c6cf4fd2c164c68acb9b39ad23b96d0bb0", + "size_bytes": 35818 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-2gpu/attempt-5055042/gpu-telemetry-5055042.csv": { + "sha256": "16db51ad8195093dff45792488347fa38e88198a4564b5253c78fdb92e4d2740", + "size_bytes": 106540 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-2gpu/attempt-5055042/launcher.log": { + "sha256": "7a2f3a9b6eba94a5200c8eb0ef525fce7d99dda68443906693162a1c2bcf886d", + "size_bytes": 150049 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-2gpu/attempt-5055042/phase-memory-5055042.csv": { + "sha256": "fa7326fc19942091ab5b7eb7270e3da658865b7ff0daffd01e1a4bf1bb5f27e7", + "size_bytes": 54058 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-2gpu/attempt-5055042/training-5055042.log": { + "sha256": "efa31f8c575d34d16d417343eb93361e6b4e0a45a9d8d82312479967e1693c52", + "size_bytes": 149886 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-2gpu/attempt-5055042/trial-5055042.json": { + "sha256": "bd886eb42686b3594abdea833bb85be5b03987d57c452d40af121efd75684f3d", + "size_bytes": 1410 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-4gpu/attempt-5055042/device-evidence/probe-exit-idle.json": { + "sha256": "6d8ab896edab73bb94e793b34a47b047fa9e8e412e3dab38518f37bc777cdfeb", + "size_bytes": 3686 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-4gpu/attempt-5055042/device-evidence/ray-resources.json": { + "sha256": "801627c53c09cff4358b400a759197cd8fc42271435e09d5d433221ff92764cc", + "size_bytes": 1550 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-4gpu/attempt-5055042/device-evidence/trainer-cuda.json": { + "sha256": "889e7d5e46bc399be6fb555e8357c0cda642ea94f16f4942a2e245f758666194", + "size_bytes": 853 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-4gpu/attempt-5055042/device-evidence/trainer-exec.json": { + "sha256": "b45899103b146a3115ab40b665559d767a7a554f22cb1935197a0859d77c2f88", + "size_bytes": 446 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-4gpu/attempt-5055042/device-evidence/trainer-terminal.json": { + "sha256": "8c6c78b946e0425f3d670d9d9126e88aeb0a68a0921e5bd42a0a669b28744842", + "size_bytes": 162 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-4gpu/attempt-5055042/device-evidence/worker-actor_rollout-0-271826.json": { + "sha256": "97b6066afc768bc47c034f1bb16bfc41afae4bd6d023af0d4b47545679ddbb46", + "size_bytes": 654 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-4gpu/attempt-5055042/device-evidence/worker-actor_rollout-1-271827.json": { + "sha256": "fb23862ab7523861850f25d7c6f34d6a87bfad494ec5336393058dbe7345f850", + "size_bytes": 654 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-4gpu/attempt-5055042/device-evidence/worker-actor_rollout-2-271828.json": { + "sha256": "0b7f6a2a2b9868e4333c0c8732a997411e511982cf4cc6beecbfc69bdfbbc807", + "size_bytes": 654 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-4gpu/attempt-5055042/device-evidence/worker-actor_rollout-3-271829.json": { + "sha256": "e4b17a57e711b930edfc03ac27bdc989ae0ffe1adf91a976cd9ae62ec60635c0", + "size_bytes": 654 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-4gpu/attempt-5055042/dispatch.json": { + "sha256": "508c995eed13d8d3c415b50927ef6f476df03e708eba1d4a6fcb66960914044a", + "size_bytes": 4493 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-4gpu/attempt-5055042/environment-5055042.json": { + "sha256": "2624ff1e23a4bc65cba09c4fbe6f5eadfa493eca1b91d3ec57bc12a989e1442a", + "size_bytes": 4894 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-4gpu/attempt-5055042/gpu-device-map.json": { + "sha256": "015d25cd11446a500b26d8a261d55d2079605e7bfde4bec84c3ab1457b779db1", + "size_bytes": 809 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-4gpu/attempt-5055042/gpu-memory-5055042.csv": { + "sha256": "d29299319947f7bc5ee9b40f999cd4fa131dda811d522c4545fe6ccede367e03", + "size_bytes": 48924 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-4gpu/attempt-5055042/gpu-telemetry-5055042.csv": { + "sha256": "d6ec8fa9354cfaf14edc524e35cb213afd5c891c5ded9c42d74d3f27426b6a42", + "size_bytes": 220846 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-4gpu/attempt-5055042/launcher.log": { + "sha256": "1ee76d3c2b4f3394a26d001d8bf57ccc8f0eae5d7f01c4dcf75ca9e4529e1340", + "size_bytes": 152058 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-4gpu/attempt-5055042/phase-memory-5055042.csv": { + "sha256": "0f27c83bd3e5e93a1072e0ceb35fbc33aa6507a1255f33e074c96b7afaf0490b", + "size_bytes": 67854 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-4gpu/attempt-5055042/training-5055042.log": { + "sha256": "0ff88a9c6cae57835ec98107deafb8ce7f827969ef45bec47cc31409aa39e94d", + "size_bytes": 151895 + }, + "output/prospective_estimation/est-qwen25-7b-math-c3-s153-4gpu/attempt-5055042/trial-5055042.json": { + "sha256": "404ef3f97726c0e1a9ae76a458d3d1e821ce3fe8033281fd0df89d6c865e8abd", + "size_bytes": 1410 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c2-s151/allocation-trace.jsonl": { + "sha256": "f90e4b2f4ef0d88e63ad1c5f8abf286acf8167e60fe15ca5a1c399cea45e4410", + "size_bytes": 276326 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c2-s151/ledger.jsonl": { + "sha256": "677ebb50042b1ef16e7bd82768cee7a2d1b0266b54e3c9794f113c7de86ace1d", + "size_bytes": 12931 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c2-s151/pair-summary.json": { + "sha256": "6d190ac0b100b8ec17b64146822229790c7984ad1969715c22bd111f3dafc63c", + "size_bytes": 799 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c2-s151/pair.json": { + "sha256": "6395afe03e9cd59953f22397461c744de2d05916ffb204a554841f8f90503496", + "size_bytes": 8773 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c2-s151/period-1.json": { + "sha256": "0d322ca95de51e4c9e59b5c821b8a9cece02537ed550068f61dd561f91c553ff", + "size_bytes": 5438 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c2-s151/period-2.json": { + "sha256": "5c6a29f46f9992da8f98a6837a727c8721bade5340c22864324f0097dd6ddfa0", + "size_bytes": 5522 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c2-s152/allocation-trace.jsonl": { + "sha256": "c5ba26588b879eebc3bc93eb9c829a0d57aa23ee752d6f8fa30afe24d37e7e31", + "size_bytes": 347853 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c2-s152/ledger.jsonl": { + "sha256": "b9f49fb1c9f629d7d00cc59e9558b0bcc1ab86462a8ecc48b357cc36086b0564", + "size_bytes": 12935 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c2-s152/pair-summary.json": { + "sha256": "80a9b2673e5e5f2b98ddfc6dbf3891921b7ee4d347a42e4ccaec6f885bac29e9", + "size_bytes": 801 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c2-s152/pair.json": { + "sha256": "e537f9a68f04d79dfc3db97a988dd61656824cab1d1b99587a263c448b225dd6", + "size_bytes": 8780 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c2-s152/period-1.json": { + "sha256": "3947ef95bd853c1bdea4dd89c06984f0a04d5b04d812f9da749820e9e40b8a4f", + "size_bytes": 5439 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c2-s152/period-2.json": { + "sha256": "0ca3ca397b90057e3ad00c1f7c1fa771a1616124be10319087cc6be1afc68c15", + "size_bytes": 5523 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c2-s153/allocation-trace.jsonl": { + "sha256": "33930aca9919d5ba9991881300fda2838e3e4bb7d5e93bedb41fbdd7605a87d3", + "size_bytes": 283662 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c2-s153/ledger.jsonl": { + "sha256": "265adb55fb58ffe85ab0f4e5e2802e00235c2654fd8429eba9595177d83aaa09", + "size_bytes": 12933 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c2-s153/pair-summary.json": { + "sha256": "1b15acf0c4252f63130557e8976faaa8ef738b1846e99aaa40d86a812a98caa1", + "size_bytes": 802 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c2-s153/pair.json": { + "sha256": "14764ab83644f8cd5912cebea75dae9b704ca48ba0b6bc4cb9007fd7751c599b", + "size_bytes": 8776 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c2-s153/period-1.json": { + "sha256": "43b653a327031d00a1b5a8d61efbeb3274ffe8c35f936bc547ba07931017cf5a", + "size_bytes": 5522 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c2-s153/period-2.json": { + "sha256": "fccdc42c532e6dfeb18f2be70812ad34dc16ef7f6186f9292124befb3e7be621", + "size_bytes": 5437 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c3-s151/allocation-trace.jsonl": { + "sha256": "e5b76edeebef279a0e81b67cfee721f4b1091ba1afb03f865f88b77b6cfbe77a", + "size_bytes": 159437 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c3-s151/ledger.jsonl": { + "sha256": "158c05ad2355beccb78137a62d51614f4d958c7f9bb4c9bfc87ab8ecbe770e7d", + "size_bytes": 13340 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c3-s151/pair-summary.json": { + "sha256": "7ca8c968f976131c50df0f099f37c6a73bfa38bf558b1a65c361ef0426bdc6fb", + "size_bytes": 790 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c3-s151/pair.json": { + "sha256": "a4fcb78a9489edfbcfa33200c5d7e9f27b030f982c3e9a28f33ca963b42d44e8", + "size_bytes": 8775 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c3-s151/period-1.json": { + "sha256": "d09bc7a33289c7dfc22f8721d7a66865763d60eb8d29929923c65eebf6dbe5ad", + "size_bytes": 5980 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c3-s151/period-2.json": { + "sha256": "4a80e6150a46f646146126eb4cf6a7117fc4671b3b56d5226a6f684fa0aa88a0", + "size_bytes": 5438 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c3-s152/allocation-trace.jsonl": { + "sha256": "df4bd2e9a681568af8cf7533a86c952d8dd30adcc494e6576f5c00ee6ee9658b", + "size_bytes": 175325 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c3-s152/ledger.jsonl": { + "sha256": "cb53fc6413afbaa786f81fe6a3e57cfe1b07e57d577b6beb5f0904935ebade22", + "size_bytes": 13329 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c3-s152/pair-summary.json": { + "sha256": "3a09f561dbd890c29891b53f84f3c7af8e7d57a3dd2bf16f810f904978222512", + "size_bytes": 792 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c3-s152/pair.json": { + "sha256": "7c27305530e8fe45b8afd153ce30cfe047562310d8ac0ae5ee4583924a7c85c6", + "size_bytes": 8778 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c3-s152/period-1.json": { + "sha256": "a36a00269828bab116cb8db1cc480808b204c2aa403d550c663aadcebc76ed6e", + "size_bytes": 5432 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c3-s152/period-2.json": { + "sha256": "23edc1c9eb457dd8d92f9e38c7abf88fe766e9be9e193fcfa494b44b2760740c", + "size_bytes": 5975 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c3-s153/allocation-trace.jsonl": { + "sha256": "6b3a88de178d4364039781d6973ffc28cd5540bd5141a6aa1cd414af309e0a41", + "size_bytes": 163143 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c3-s153/ledger.jsonl": { + "sha256": "8cb922fc5199e4161a37bbb447ce56f022ff98268ca967d7adefc6d6a4fd5771", + "size_bytes": 13343 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c3-s153/pair-summary.json": { + "sha256": "351ae5b96b9604dbfe4bb5cd55b58fae64baa339b1844587d2e037322b4334c0", + "size_bytes": 793 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c3-s153/pair.json": { + "sha256": "ec8e4a2106cd025faadd7c787be1326212b53334a8aad98ef84dd624dd9eea7b", + "size_bytes": 8775 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c3-s153/period-1.json": { + "sha256": "4d2c84dc5fa51ffae118aedd06e8c09271f2a38ac7805582438d12f7173d512f", + "size_bytes": 5981 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-code_heavy_tail-c3-s153/period-2.json": { + "sha256": "c30480e747475bf45c044b00a07741a652763ea2074865c163403a1f544b49a7", + "size_bytes": 5437 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c2-s151/allocation-trace.jsonl": { + "sha256": "2f57115feefb2f4d89f64488e0e8031eda17696bbcc85d394f714f6f92175909", + "size_bytes": 336928 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c2-s151/ledger.jsonl": { + "sha256": "ec8fab57145124ce0f69f1235a1021166e930084acdfb22ad6d67c48fd712354", + "size_bytes": 12766 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c2-s151/pair-summary.json": { + "sha256": "2ec699836b9f47fa8fbd45e447445ab1bd64680e31b39b10294f9efdce272ce8", + "size_bytes": 791 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c2-s151/pair.json": { + "sha256": "223d0ca5805a06be1ba04808797cec4b35d1a0f4d2fde64bba8e75662da8719c", + "size_bytes": 8445 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c2-s151/period-1.json": { + "sha256": "ef26a4927527100d380325d84c5eae5b15eddec48e6643740a66c329bff31ebf", + "size_bytes": 5329 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c2-s151/period-2.json": { + "sha256": "2fa3961d8c7bc4188db063f2d7af0f656b0ffafd266538e0869ddf3aa1100fe9", + "size_bytes": 5576 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c2-s152/allocation-trace.jsonl": { + "sha256": "d6ebefb119f644d04752a99fdf47f4fd2eaec22a68764d46e890e28f27361cd1", + "size_bytes": 340104 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c2-s152/ledger.jsonl": { + "sha256": "58395abc9d5df303ef2155cf53231c86275c1ee75370b7ac083497b225f997dc", + "size_bytes": 12615 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c2-s152/pair-summary.json": { + "sha256": "3183b9e7cbe7b5a13e408bdbc0bf586fe0320fc7fbcb6e3246348aad60ee0673", + "size_bytes": 792 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c2-s152/pair.json": { + "sha256": "06f240c40980feac67a620ecb986328a9c43f805ee3dbe35e414ea149376a59e", + "size_bytes": 8445 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c2-s152/period-1.json": { + "sha256": "ac026a220f5609a1efb1c1d9ac1ea6f5412bd0c816f726bc3f63d7152016cd62", + "size_bytes": 5325 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c2-s152/period-2.json": { + "sha256": "ea4ad9e24978da3904c5d45bd4cc042d3a62e8def0ec5c1b486b76f10089b5d3", + "size_bytes": 5408 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c2-s153/allocation-trace.jsonl": { + "sha256": "b7304b806774a75e8b111a83e656a325ab8581c3c16e81619f64be8b4a92e667", + "size_bytes": 277990 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c2-s153/ledger.jsonl": { + "sha256": "e3fecbf6c338a0d390915e18757a9b0ac12973f79d191f60925cab0e6a67c367", + "size_bytes": 12625 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c2-s153/pair-summary.json": { + "sha256": "410686e5aa976c61001870b499d85a6cf6233bdccaa032c29bc6bebfecf88191", + "size_bytes": 791 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c2-s153/pair.json": { + "sha256": "ea390f752b0726d01796ec9bd9cdbc7b95ba2e214c76a72933f5e840914630d6", + "size_bytes": 8448 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c2-s153/period-1.json": { + "sha256": "f2636b4abde9a7dcce5573d0bce5dd813f116a3ff4c33e118d33652be4cd5f0f", + "size_bytes": 5412 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c2-s153/period-2.json": { + "sha256": "f4052500d39310d01e7d10f027f879cfae2f32ba8c458896b62e8cbdfbb19a52", + "size_bytes": 5330 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c3-s151/allocation-trace.jsonl": { + "sha256": "83b0445bfdd38479d75b45c14c1d185bd163e1d639b053c02d466faa35b68839", + "size_bytes": 164725 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c3-s151/ledger.jsonl": { + "sha256": "1974529d05d8aa3adf94448d49f6178553bb12da672dd9ffbe8f9d1f283f06b8", + "size_bytes": 13023 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c3-s151/pair-summary.json": { + "sha256": "b87e931ff905db4ae4a55d8fd2f33399972e27bfc46d887d4aa2b6c3ce72e952", + "size_bytes": 782 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c3-s151/pair.json": { + "sha256": "019720c3288d38821c459ca30347a1829315b6cbfc8a267aa2bd35f81fa32b92", + "size_bytes": 8450 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c3-s151/period-1.json": { + "sha256": "b174fc403f481f5b9c69cbb582f89716356970aea5176db7014c344f72bbc731", + "size_bytes": 5865 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c3-s151/period-2.json": { + "sha256": "38a8f2b0f4d7349a09514002b026c8fb4d868dd2da68b27a750eed3863b400bc", + "size_bytes": 5326 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c3-s152/allocation-trace.jsonl": { + "sha256": "aa2e2b2d9fb8720968302dbf5f7f8cd38c9c89db937bd2b041856b2e8f31f018", + "size_bytes": 166986 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c3-s152/ledger.jsonl": { + "sha256": "c92968019ad2d4c7b072941de26e1bc6621c4f980c66e27a304e703d1a3e04a5", + "size_bytes": 13070 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c3-s152/pair-summary.json": { + "sha256": "532050bf4427ddc49a982a8cc0f843ea2f6c120cd882c3208d74a98d9831eca2", + "size_bytes": 781 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c3-s152/pair.json": { + "sha256": "a66599ba72b406d526e0c736ef1daaee8d5d6f844a26eb38f4d3f5df6b57a4dd", + "size_bytes": 8450 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c3-s152/period-1.json": { + "sha256": "44d9a011187dff4f6563e735da526c1abc07c8f577f8dd346ed324fde344ce4f", + "size_bytes": 5872 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c3-s152/period-2.json": { + "sha256": "b1035f53de96d786ca1d945d199702eadce09f4bc1663d1c021d77141238b191", + "size_bytes": 5389 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c3-s153/allocation-trace.jsonl": { + "sha256": "4db45e17bea7b6160f8e051ff73c63b1cb9511d33630c9e3929236080870242b", + "size_bytes": 181831 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c3-s153/ledger.jsonl": { + "sha256": "0ab7e45cad162bfab6e8e6ae2c77bcbdf29c32fa102f3a39d2712159556b48cb", + "size_bytes": 13036 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c3-s153/pair-summary.json": { + "sha256": "5fdef1e489a6973ddb619f4fb981decee82bf654d58246715b80c1c3888214ff", + "size_bytes": 783 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c3-s153/pair.json": { + "sha256": "e2db8b1be2bf8cc717f5d9cecc6493baab989bc84fd92e2465486dfe15bc0b49", + "size_bytes": 8447 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c3-s153/period-1.json": { + "sha256": "7e35e6f767d532f1635e2a51c61268b1b3d3bcbf6d1af838f7b74c130c321197", + "size_bytes": 5329 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-gsm8k-c3-s153/period-2.json": { + "sha256": "f9e1208d6b4fc1bccab85947fb28d793435c3baf049fe9f9cdb4054e65a506a6", + "size_bytes": 5872 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c2-s151/allocation-trace.jsonl": { + "sha256": "01001e86d80590e8df284c54bdd4c11cd92bc96611b050dfa084fb33579abdb7", + "size_bytes": 33018 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c2-s151/ledger.jsonl": { + "sha256": "3dff67c7e3f4c24f35883caef96992ab323630eab11c2bdfcce2d34ada12d403", + "size_bytes": 10655 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c2-s151/pair-summary.json": { + "sha256": "0417a52a7930ee9005b184ad39ac097062319b4a1b431df404a3a99bf765a642", + "size_bytes": 818 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c2-s151/pair.json": { + "sha256": "3352f7e0ba5d4ae3c78325713c59a571f6c1b0a46d936e6975a437fc46565feb", + "size_bytes": 8386 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c2-s151/period-1.json": { + "sha256": "a62701a3a3d2e887a5dcf11dfc741b2610823f44a246a92e5231af32de42c3b1", + "size_bytes": 4535 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c2-s151/period-2.json": { + "sha256": "dc42ec7153404640342efd1b6bf759ca275347ed57e6a7ca228e5ca1e3f31663", + "size_bytes": 3984 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c2-s152/allocation-trace.jsonl": { + "sha256": "c013445b2c8f7c4c9ef002f1566ea315fc8dbce8d4d187b61aadc9fbe39eb90b", + "size_bytes": 290796 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c2-s152/ledger.jsonl": { + "sha256": "1064fb987e1b69fe0bf650314ce057fe2a4c8059614c565045d7155346cfbe06", + "size_bytes": 12581 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c2-s152/pair-summary.json": { + "sha256": "17c0da435b3dbbfea3a5382f501465a9e7909d98f4cb40f2ab599c4ba87f4bdc", + "size_bytes": 791 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c2-s152/pair.json": { + "sha256": "2aeea4bb6c09da6918856caa89918ada979d4d0f64691bc1d5597c23d0ed52fd", + "size_bytes": 8430 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c2-s152/period-1.json": { + "sha256": "6fb0dc67d5195559c72d0cd34b71f38c462ab3a842b61385f1a12269af1ad4a7", + "size_bytes": 5312 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c2-s152/period-2.json": { + "sha256": "b0416039d3dea26af289e2743d09f15297798a19243986540b9837310c8e5b9b", + "size_bytes": 5396 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c2-s153/allocation-trace.jsonl": { + "sha256": "af5e048c8a1b23081d8238d38dcd5d49268eac01c0f8f69f41c321e38cf4146f", + "size_bytes": 50378 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c2-s153/ledger.jsonl": { + "sha256": "8a7b481ea18db0c3bdfae4ab480be6e357c735feaa755a90c0e76031a8de6b7d", + "size_bytes": 11417 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c2-s153/pair-summary.json": { + "sha256": "2899a3997c5b1365fcdf94a4a116df1bfd0aa0c76cef0def4fd3fda9623ce7dd", + "size_bytes": 797 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c2-s153/pair.json": { + "sha256": "991df8f86d667cb619f1114341213752936e091f06df8f609442286416781efc", + "size_bytes": 8390 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c2-s153/period-1.json": { + "sha256": "474661536a016fd6e5239e1b9ef06a1a1993cbfd2fe85aab4676f45fd36c112e", + "size_bytes": 4434 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c2-s153/period-2.json": { + "sha256": "598ee231712394d428000a04280120717d2efa0325f4919c744c3d2284090e94", + "size_bytes": 4914 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c3-s151/allocation-trace.jsonl": { + "sha256": "e09e862f52ca448b1d2ab053f7221f1cf29008c1ff323215a6494de4900f78cb", + "size_bytes": 180126 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c3-s151/ledger.jsonl": { + "sha256": "933314fb69c761afd7901fcca99b7b7ca86a2ff64e93d4cafd39a4c3c02712ac", + "size_bytes": 13000 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c3-s151/pair-summary.json": { + "sha256": "1bdbd50b4bae9ebed6a065dae42bf75590be28ecfefda7c376f85b3329f22bea", + "size_bytes": 781 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c3-s151/pair.json": { + "sha256": "56a9e2bf77364a2c7548a806774042887622e49fb384cc9ae0d540ff95105c2c", + "size_bytes": 8432 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c3-s151/period-1.json": { + "sha256": "3544821ef82aece87f7911e0f4da77ed7b506d9d0ec758b190c0556481454e15", + "size_bytes": 5859 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c3-s151/period-2.json": { + "sha256": "189147a36ee101b178ff7b5d1998c084bef53d5a0941b6a86310b5f6fdebf1e6", + "size_bytes": 5316 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c3-s152/ledger.jsonl": { + "sha256": "85f9da6513e1fe16b8dad1e6fec0d54aabe0af20601fa5883a21c81b4a73a06a", + "size_bytes": 2116 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c3-s152/pair.json": { + "sha256": "f4f5d85031ac044d1db31cb535d93ca010ced3f61d7e8abfae2fe2391dffe85e", + "size_bytes": 4240 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c3-s152/period-1.json": { + "sha256": "a340cb89d26f9c9158cbe131e6bb132d5420f5be48c23cb1199e0441d389eb71", + "size_bytes": 608 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c3-s152/period-2.json": { + "sha256": "ccb8e7920276d16ec0d655cbf4a9d5b4bd15176db54d27216a8f17e25d57ff7b", + "size_bytes": 608 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c3-s153/allocation-trace.jsonl": { + "sha256": "57a29a189b190c01ef7281386dac6c163dc068a47d9f11b541faf7c8ac5869e7", + "size_bytes": 160947 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c3-s153/ledger.jsonl": { + "sha256": "f8bdfa5c80e6aaddc5e45ba364a1fec1f08730ec8ca6149077226468af82fb42", + "size_bytes": 12992 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c3-s153/pair-summary.json": { + "sha256": "f487a2d5cdf6bf23840f6a8fcff6a8d2a4c3a663451c6d41b52bb6a7270f38fa", + "size_bytes": 782 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c3-s153/pair.json": { + "sha256": "4cb4bf4a3c5922f305b91dd0985935fde9f33d62c50b707baeb014d517f9c595", + "size_bytes": 8435 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c3-s153/period-1.json": { + "sha256": "c2c47befba8102cb5b474225c37a483e46fea3b2cd00a76be9fc022846bad405", + "size_bytes": 5314 + }, + "output/prospective_estimation/pairs/est-qwen25-7b-math-c3-s153/period-2.json": { + "sha256": "b18ab401d9e1a5a0de2c281c06e15bab497d3e0aee208b16a4f894ad34bc251b", + "size_bytes": 5854 + } + }, + "prediction_frozen_at_utc": "2026-09-15T14:10:21.535819+00:00", + "prediction_seal_binding": "Effective input/output hashes and pre-claim chronology; runner verifies seal but does not store its hash per period.", + "prediction_seal_sha256": "5ed714caff922a5109346a4242545d19f2824636d0f52bb0bbd2a09b40ed4666", + "protocol_input_evidence": { + "amendment-01.json": { + "sha256": "b02fe0419f865c4cc3d77fb35bb4205a54f3aa9605fdadc7b86a9a11b172d25e", + "size_bytes": 4632 + }, + "fitted_model.json": { + "sha256": "8b84aa1ae57fe56fd478817dbbce505487cbef6b26652ae2c8f45e9fa5ef5df4", + "size_bytes": 11485 + }, + "matrix.csv": { + "sha256": "1ff175e22c76bf1af3857b4ae57cd4f78bafe0d280a295d696e96815ef85e9c8", + "size_bytes": 16755 + }, + "model_metadata.json": { + "sha256": "2f0ce16c13e72900d2a60f9b7ad28d4dc5ed43d7915d8a88925ba9c5ffd684ad", + "size_bytes": 5599 + }, + "prediction_freeze.json": { + "sha256": "5ed714caff922a5109346a4242545d19f2824636d0f52bb0bbd2a09b40ed4666", + "size_bytes": 3004 + }, + "predictions.json": { + "sha256": "96d7f2ae11145f0985c0faf743333d941afbcd3b80966a857b1af0af15306ace", + "size_bytes": 21607 + }, + "protocol.json": { + "sha256": "1694e447e842600612604891e7b3d036ff0791fa84c4bdbeba54ed093feefdeb", + "size_bytes": 11490 + }, + "retrospective.json": { + "sha256": "8ada76da63ab91e0b1ca6276a98f2b096979fe6acd7578c1223469d55d399f3a", + "size_bytes": 228053 + }, + "targets.csv": { + "sha256": "f2481278127fa726f24297d81fc9df46e999315fe543167ed398e671fffa522f", + "size_bytes": 5037 + } + }, + "protocol_sha256": "1694e447e842600612604891e7b3d036ff0791fa84c4bdbeba54ed093feefdeb", + "raw_statuses_used_as_labels": false, + "runner_validation_implementation_sha256": "f340fb7eec5297cb675a48e267618c15d7787a42891d4a2ae4a670ec24a0cc47", + "source_commit_expected": null, + "source_input_evidence": { + "benchmark.py": { + "sha256": "35e36caad46d2814b0ac9f8180961de5564746347cf668d5c3fc1e643fd477e8", + "size_bytes": 10897 + }, + "benchmark/attempts.csv": { + "sha256": "397c8caaf0d3abe06587fd649c181fe71d5b8b5f5e24e0591cc7e56d38e275c1", + "size_bytes": 588698 + }, + "benchmark/configurations.csv": { + "sha256": "3a77c96a67b82e01d23fb6f7a440cba23041bed1dae2749072cb9abb5b5d239d", + "size_bytes": 71349 + }, + "benchmark/estimation/matrix.csv": { + "sha256": "1ff175e22c76bf1af3857b4ae57cd4f78bafe0d280a295d696e96815ef85e9c8", + "size_bytes": 16755 + }, + "benchmark/estimation/model_metadata.json": { + "sha256": "2f0ce16c13e72900d2a60f9b7ad28d4dc5ed43d7915d8a88925ba9c5ffd684ad", + "size_bytes": 5599 + }, + "benchmark/estimation/targets.csv": { + "sha256": "f2481278127fa726f24297d81fc9df46e999315fe543167ed398e671fffa522f", + "size_bytes": 5037 + }, + "benchmark/outcomes.csv": { + "sha256": "af05cd91ed1de843735cc188e5e54dbd1645b3ed91c97f503ac3de67bc5f61c9", + "size_bytes": 21219 + }, + "memory_tuner/estimation_baselines.py": { + "sha256": "8cc20f9697eb38cb6a19a2255e95712ac44469cbd147eaf93e7e7ed183360ad2", + "size_bytes": 26850 + }, + "memory_tuner/model_memory_metadata.py": { + "sha256": "8845a1197b49d9463ebe5dc8c953463dd2106361c54e7b12244057c48aaa2712", + "size_bytes": 6588 + }, + "memory_tuner/plan_estimation_study.py": { + "sha256": "16855f584441528870fcca38c3a19eafd1234c3eacb9b43cc9eebdd4bd6ad862", + "size_bytes": 10472 + } + } +} diff --git a/benchmark/estimation/results/summary.json b/benchmark/estimation/results/summary.json new file mode 100644 index 0000000000000000000000000000000000000000..623ba59b1af8dc6c9a38e72cb6951cfc14e828bf --- /dev/null +++ b/benchmark/estimation/results/summary.json @@ -0,0 +1,13 @@ +{ + "completed_processes": 0, + "eligible_processes": 23, + "known_launched_processes": 34, + "pair_errors": {}, + "planned_configurations": 12, + "planned_pairs": 18, + "planned_processes": 36, + "resolved_configurations": 6, + "schema_version": 1, + "unknown_launch_status_processes": 0, + "unresolved_processes": 13 +} diff --git a/benchmark/estimation/retrospective.json b/benchmark/estimation/retrospective.json new file mode 100644 index 0000000000000000000000000000000000000000..9bb7a9dfb9794e4a30b23709646884c02efb2179 --- /dev/null +++ b/benchmark/estimation/retrospective.json @@ -0,0 +1,5718 @@ +{ + "folds": { + "granite33": { + "model": { + "classifier_classes": [ + "above_margin", + "memory_failure", + "within_margin" + ], + "classifier_coefficients": [ + [ + -0.07648332729335744, + -0.4879335028999397, + -0.27438539693412517, + 0.5708002618087737, + 0.5708002618087737, + 0.06892038045359755, + 0.05603580859652673, + 0.20812798567206875 + ], + [ + 0.5345155673736126, + 0.6422643873754361, + 2.4014179573636096, + -0.20286515475315908, + -0.20286515475315908, + 0.23665426065423395, + 0.24510773091127966, + 0.2731342366004145 + ], + [ + -0.45803224008025506, + -0.1543308844754974, + -2.1270325604294844, + -0.36793510705561366, + -0.36793510705561366, + -0.3055746411078309, + -0.3011435395078059, + -0.48126222227248255 + ] + ], + "classifier_intercepts": [ + -0.2652695334123221, + -1.5267379806608377, + 1.7920075140731606 + ], + "classifier_iterations": [ + 29 + ], + "component_coefficients_gib": [ + 10.7612512470666, + 0.217930674965405, + 0.0, + 18.714031955998458, + 0.0, + 4.583381487620393, + 0.0, + 1.4554318201713192, + 6.018987618003623 + ], + "component_divisors": [ + 7.317022323608398, + 1.0, + 32.0, + 7.317022323608398, + 2.289562225341797, + 9.28125, + 18.31640625, + 7.317022323608398 + ], + "component_training_residual_norm_gib": 10.181990095153003, + "eligible_source_processes": 180, + "feature_names": [ + "actor_shards_gib", + "adapter_optimizer_gib", + "generation_budget_gib", + "generation_replica_gib", + "largest_unit_proxy_gib", + "checkpoint_boundary_volume_gib", + "vocabulary_output_volume_gib", + "resident_actor_shards_gib" + ], + "recorded_source_attempts": 185, + "regression_training_configuration_ids": [ + "cfg-072bd7669d20f8fa", + "cfg-0d9ba9215ab5683e", + "cfg-0f540d3f6fe4661b", + "cfg-0f8a31ef7ac4c551", + "cfg-1432bb8a4d9c3b8c", + "cfg-17e69bcf1b041bd9", + "cfg-18285c3cb50308d4", + "cfg-1c5278f8be90a50a", + "cfg-301e4dae7048c6ec", + "cfg-30f98456fdb322c1", + "cfg-338d9224eb9d2f3a", + "cfg-371ae37408089efe", + "cfg-399f6b038d6f65f9", + "cfg-3f0f48a3c570a0c8", + "cfg-5244d13eac3116ad", + "cfg-5ff9621270e710ed", + "cfg-62f371a80968a0ba", + "cfg-6d54ac65d934f404", + "cfg-714c540f76d9525d", + "cfg-78c440d3891ddbc2", + "cfg-7a9a974958c57abc", + "cfg-89ca48ce88ea694e", + "cfg-8bd7050cd7f49e78", + "cfg-9160dc45a0171a10", + "cfg-96ccf395b84393f7", + "cfg-97107ceba8bf0cbe", + "cfg-9c461415e6bb9702", + "cfg-a026693bae37d30b", + "cfg-a23a9ffc1f670139", + "cfg-a24e9cdd2f6143f2", + "cfg-a3987314cb2f9f12", + "cfg-b5007322fb2e1772", + "cfg-b8dd70ab682c13b2", + "cfg-bfb83ff32b7fd921", + "cfg-c0605c1607caa079", + "cfg-cc2c35944156e5d4", + "cfg-cd3974595f9db882", + "cfg-db87c67536995ab0", + "cfg-e117f15dd5ccfb30", + "cfg-e2a205be8efba055", + "cfg-e91942abb5b4f847", + "cfg-f009e9e3891c06f4", + "cfg-f03e4e98c682240e", + "cfg-f75b68f479767290" + ], + "scaler_mean": [ + 5.979624080657959, + 0.35540771484375, + 25.333333333333332, + 6.644026756286621, + 1.724374771118164, + 4.42425537109375, + 9.153238932291666, + 4.20788361231486 + ], + "scaler_scale": [ + 1.4665293178520666, + 0.09198594203733253, + 7.180219742846007, + 0.6729955673217773, + 0.5651874542236328, + 3.9610090176861723, + 7.5231216467575335, + 2.8823700905353844 + ], + "training_configuration_ids": [ + "cfg-072bd7669d20f8fa", + "cfg-0d9ba9215ab5683e", + "cfg-0f540d3f6fe4661b", + "cfg-0f8a31ef7ac4c551", + "cfg-10644b5d0a82d5c3", + "cfg-1432bb8a4d9c3b8c", + "cfg-17e69bcf1b041bd9", + "cfg-18285c3cb50308d4", + "cfg-1c5278f8be90a50a", + "cfg-301e4dae7048c6ec", + "cfg-30f98456fdb322c1", + "cfg-338d9224eb9d2f3a", + "cfg-352f4dbc55fbcd3c", + "cfg-371ae37408089efe", + "cfg-399f6b038d6f65f9", + "cfg-3c86a11fdb8455cb", + "cfg-3f0f48a3c570a0c8", + "cfg-49d9bccfe873f5c6", + "cfg-5244d13eac3116ad", + "cfg-55811d23c5144bb7", + "cfg-5839ccf95c2c1722", + "cfg-5ff9621270e710ed", + "cfg-62f371a80968a0ba", + "cfg-6d54ac65d934f404", + "cfg-714c540f76d9525d", + "cfg-78c440d3891ddbc2", + "cfg-7a9a974958c57abc", + "cfg-7df2428553b76877", + "cfg-896685010703c0fb", + "cfg-89ca48ce88ea694e", + "cfg-8bd7050cd7f49e78", + "cfg-9160dc45a0171a10", + "cfg-966f5804dc918967", + "cfg-96ccf395b84393f7", + "cfg-97107ceba8bf0cbe", + "cfg-9c461415e6bb9702", + "cfg-a026693bae37d30b", + "cfg-a23a9ffc1f670139", + "cfg-a24e9cdd2f6143f2", + "cfg-a3987314cb2f9f12", + "cfg-b5007322fb2e1772", + "cfg-b8dd70ab682c13b2", + "cfg-bfb83ff32b7fd921", + "cfg-c0605c1607caa079", + "cfg-ca71b56dc3189305", + "cfg-cad8341dbbff10ba", + "cfg-cc2c35944156e5d4", + "cfg-cc900fb7731c3f49", + "cfg-cd3974595f9db882", + "cfg-d138048feeb8fe5d", + "cfg-db87c67536995ab0", + "cfg-e117f15dd5ccfb30", + "cfg-e2a205be8efba055", + "cfg-e91942abb5b4f847", + "cfg-ea5a23a40ad5ca50", + "cfg-ea92cccc91029465", + "cfg-f009e9e3891c06f4", + "cfg-f03e4e98c682240e", + "cfg-f75b68f479767290", + "cfg-fad1b555f8b37235" + ] + }, + "predictions": [ + { + "configuration_id": "cfg-2afbe250af9cd3e1", + "donor_configuration_id": "cfg-714c540f76d9525d", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 34212.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31195.671875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-2afbe250af9cd3e1", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 30829.341238787812, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31195.671875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-2afbe250af9cd3e1", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31195.671875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-32977d5aa42a41cb", + "donor_configuration_id": "cfg-cad8341dbbff10ba", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39863.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-32977d5aa42a41cb", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 36695.64360711028, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39863.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-32977d5aa42a41cb", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39863.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-3dc042a103546eac", + "donor_configuration_id": "cfg-fad1b555f8b37235", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39863.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-3dc042a103546eac", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 37061.8499940472, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39863.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-3dc042a103546eac", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39863.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-61c14de21a34d0e6", + "donor_configuration_id": "cfg-ea5a23a40ad5ca50", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 37815.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-61c14de21a34d0e6", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 35452.169763559265, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 37815.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-61c14de21a34d0e6", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 37815.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-6a4562db98b38d6f", + "donor_configuration_id": "cfg-5839ccf95c2c1722", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 37815.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-6a4562db98b38d6f", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 35620.01435757202, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 37815.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-6a4562db98b38d6f", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 37815.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-73c06764ba28a56b", + "donor_configuration_id": "cfg-7a9a974958c57abc", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 37140.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 33719.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-73c06764ba28a56b", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 33224.61826720422, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 33719.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-73c06764ba28a56b", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 33719.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-7a0e81934d3398ee", + "donor_configuration_id": "cfg-b5007322fb2e1772", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 21562.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-7a0e81934d3398ee", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 19201.17206061656, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-7a0e81934d3398ee", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-7f23cab692546e3f", + "donor_configuration_id": "cfg-ca71b56dc3189305", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39863.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-7f23cab692546e3f", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 36848.229601667335, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39863.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-7f23cab692546e3f", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39863.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-835067f7405e89a2", + "donor_configuration_id": "cfg-f009e9e3891c06f4", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 37478.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 33719.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-835067f7405e89a2", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 32980.480675912935, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 33719.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-835067f7405e89a2", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 33719.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-93d4f41e600c9079", + "donor_configuration_id": "cfg-49d9bccfe873f5c6", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 37815.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-93d4f41e600c9079", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 35558.9799597492, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 37815.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-93d4f41e600c9079", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 37815.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-9758870c0a0b4d69", + "donor_configuration_id": "cfg-f03e4e98c682240e", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 30180.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 27099.671875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-9758870c0a0b4d69", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 28372.910750597188, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 27099.671875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-9758870c0a0b4d69", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 27099.671875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-9f624ec22aa3dde0", + "donor_configuration_id": "cfg-a23a9ffc1f670139", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 38316.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 35291.671875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-9f624ec22aa3dde0", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 33224.737329155614, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 35291.671875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-9f624ec22aa3dde0", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 35291.671875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-a94bf7877633dfda", + "donor_configuration_id": "cfg-6d54ac65d934f404", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 33388.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 29623.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-a94bf7877633dfda", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 30562.196686361578, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 29623.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-a94bf7877633dfda", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 29623.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-aa33c72e0a449b9c", + "donor_configuration_id": "cfg-1c5278f8be90a50a", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 37682.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 33719.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-aa33c72e0a449b9c", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 33163.5838693814, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 33719.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-aa33c72e0a449b9c", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 33719.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-acfa4acda15c3599", + "donor_configuration_id": "cfg-e91942abb5b4f847", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 33436.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 29623.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-acfa4acda15c3599", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 30615.601784456547, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 29623.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-acfa4acda15c3599", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 29623.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-b5352215babfeba6", + "donor_configuration_id": "cfg-30f98456fdb322c1", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 33446.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 29623.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-b5352215babfeba6", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 30524.050187722318, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 29623.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-b5352215babfeba6", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 29623.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-b56fb353c2f2fa3e", + "donor_configuration_id": "cfg-3f0f48a3c570a0c8", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 21860.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-b56fb353c2f2fa3e", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 19185.913461160857, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-b56fb353c2f2fa3e", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-b8e90bd69c9744ea", + "donor_configuration_id": "cfg-399f6b038d6f65f9", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 29840.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 27099.671875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-b8e90bd69c9744ea", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 28494.97954624283, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 27099.671875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-b8e90bd69c9744ea", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 27099.671875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-bae2977da7957f72", + "donor_configuration_id": "cfg-9160dc45a0171a10", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 21560.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-bae2977da7957f72", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 19174.469511569077, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-bae2977da7957f72", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-c2a8e6e8e140e109", + "donor_configuration_id": "cfg-cc2c35944156e5d4", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 33872.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31195.671875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-c2a8e6e8e140e109", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 31073.478830079097, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31195.671875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-c2a8e6e8e140e109", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31195.671875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-c3f0ad157d299d39", + "donor_configuration_id": "cfg-0d9ba9215ab5683e", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 37418.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 33719.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-c3f0ad157d299d39", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 33056.77367319146, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 33719.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-c3f0ad157d299d39", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 33719.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-cc5f4f0c2e1256ba", + "donor_configuration_id": "cfg-db87c67536995ab0", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 21540.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-cc5f4f0c2e1256ba", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 22767.56364712078, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-cc5f4f0c2e1256ba", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-d02503f8c9804a2d", + "donor_configuration_id": "cfg-55811d23c5144bb7", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 37815.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-d02503f8c9804a2d", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 35375.87676628074, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 37815.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-d02503f8c9804a2d", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 37815.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-d4a213515805a9a7", + "donor_configuration_id": "cfg-cd3974595f9db882", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 21580.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-d4a213515805a9a7", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 22809.52479562397, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-d4a213515805a9a7", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-d9397d15f234d541", + "donor_configuration_id": "cfg-bfb83ff32b7fd921", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 21560.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-d9397d15f234d541", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 22794.266196168264, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-d9397d15f234d541", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-e8893b12558f51dd", + "donor_configuration_id": "cfg-352f4dbc55fbcd3c", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39863.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-e8893b12558f51dd", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 37183.91878969284, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39863.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-e8893b12558f51dd", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39863.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-f439d2dd90137a80", + "donor_configuration_id": "cfg-78c440d3891ddbc2", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 37976.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 35291.671875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-f439d2dd90137a80", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 33468.8749204469, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 35291.671875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-f439d2dd90137a80", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 35291.671875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-fbe1c0422f98cb38", + "donor_configuration_id": "cfg-338d9224eb9d2f3a", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 33108.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 29623.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-fbe1c0422f98cb38", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 30646.118983367956, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 29623.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-fbe1c0422f98cb38", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 29623.34375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-fd4d4789145230df", + "donor_configuration_id": "cfg-62f371a80968a0ba", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 21900.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-fd4d4789145230df", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 22779.00759671256, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-fd4d4789145230df", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-fd5bb0db68b0dae3", + "donor_configuration_id": "cfg-5244d13eac3116ad", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 21560.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-fd5bb0db68b0dae3", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 19216.430660072267, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-fd5bb0db68b0dae3", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + } + ], + "scores": { + "component_regression": { + "above_margin_among_approved": 0.13333333333333333, + "approval_rate_on_above_margin_targets": 1.0, + "approval_rate_on_memory_failure_targets": 1.0, + "approved": 30, + "approved_above_margin": 4, + "approved_memory_failure": 4, + "approved_within_margin": 22, + "completed_peak_mae_mib": 2631.143592948213, + "completed_peak_mean_signed_error_mib": -2145.8794031110224, + "completed_peak_prediction_coverage": 26, + "completed_peak_worst_underestimate_mib": 5737.519324087065, + "confusion_matrix_predicted_rows_observed_columns": { + "above_margin": { + "above_margin": 0, + "memory_failure": 0, + "within_margin": 0 + }, + "memory_failure": { + "above_margin": 0, + "memory_failure": 0, + "within_margin": 0 + }, + "within_margin": { + "above_margin": 4, + "memory_failure": 4, + "within_margin": 22 + } + }, + "correct_three_state": 22, + "distinct_target_configurations": 30, + "memory_failure_among_approved": 0.13333333333333333, + "queries": 30, + "rejected_within_margin": 0, + "rejection_rate_on_within_margin_targets": 0.0, + "startup_guard_changed_states": 0, + "target_above_margin": 4, + "target_memory_failure": 4, + "target_within_margin": 22, + "three_state_accuracy": 0.7333333333333333, + "within_margin_approval_precision": 0.7333333333333333, + "within_margin_recall": 1.0 + }, + "donor_copy": { + "above_margin_among_approved": 0.0, + "approval_rate_on_above_margin_targets": 0.0, + "approval_rate_on_memory_failure_targets": 0.0, + "approved": 22, + "approved_above_margin": 0, + "approved_memory_failure": 0, + "approved_within_margin": 22, + "completed_peak_mae_mib": 887.5454545454545, + "completed_peak_mean_signed_error_mib": 774.8181818181819, + "completed_peak_prediction_coverage": 22, + "completed_peak_worst_underestimate_mib": 1240.0, + "confusion_matrix_predicted_rows_observed_columns": { + "above_margin": { + "above_margin": 0, + "memory_failure": 0, + "within_margin": 0 + }, + "memory_failure": { + "above_margin": 4, + "memory_failure": 4, + "within_margin": 0 + }, + "within_margin": { + "above_margin": 0, + "memory_failure": 0, + "within_margin": 22 + } + }, + "correct_three_state": 26, + "distinct_target_configurations": 30, + "memory_failure_among_approved": 0.0, + "queries": 30, + "rejected_within_margin": 0, + "rejection_rate_on_within_margin_targets": 0.0, + "startup_guard_changed_states": 0, + "target_above_margin": 4, + "target_memory_failure": 4, + "target_within_margin": 22, + "three_state_accuracy": 0.8666666666666667, + "within_margin_approval_precision": 1.0, + "within_margin_recall": 1.0 + }, + "logistic": { + "above_margin_among_approved": 0.12, + "approval_rate_on_above_margin_targets": 0.75, + "approval_rate_on_memory_failure_targets": 0.0, + "approved": 25, + "approved_above_margin": 3, + "approved_memory_failure": 0, + "approved_within_margin": 22, + "completed_peak_mae_mib": null, + "completed_peak_mean_signed_error_mib": null, + "completed_peak_prediction_coverage": 0, + "completed_peak_worst_underestimate_mib": null, + "confusion_matrix_predicted_rows_observed_columns": { + "above_margin": { + "above_margin": 0, + "memory_failure": 0, + "within_margin": 0 + }, + "memory_failure": { + "above_margin": 1, + "memory_failure": 4, + "within_margin": 0 + }, + "within_margin": { + "above_margin": 3, + "memory_failure": 0, + "within_margin": 22 + } + }, + "correct_three_state": 26, + "distinct_target_configurations": 30, + "memory_failure_among_approved": 0.0, + "queries": 30, + "rejected_within_margin": 0, + "rejection_rate_on_within_margin_targets": 0.0, + "startup_guard_changed_states": 0, + "target_above_margin": 4, + "target_memory_failure": 4, + "target_within_margin": 22, + "three_state_accuracy": 0.8666666666666667, + "within_margin_approval_precision": 0.88, + "within_margin_recall": 1.0 + } + }, + "source_cost": { + "distinct_selected_donors": { + "configuration_ids": [ + "cfg-0d9ba9215ab5683e", + "cfg-1c5278f8be90a50a", + "cfg-30f98456fdb322c1", + "cfg-338d9224eb9d2f3a", + "cfg-352f4dbc55fbcd3c", + "cfg-399f6b038d6f65f9", + "cfg-3f0f48a3c570a0c8", + "cfg-49d9bccfe873f5c6", + "cfg-5244d13eac3116ad", + "cfg-55811d23c5144bb7", + "cfg-5839ccf95c2c1722", + "cfg-62f371a80968a0ba", + "cfg-6d54ac65d934f404", + "cfg-714c540f76d9525d", + "cfg-78c440d3891ddbc2", + "cfg-7a9a974958c57abc", + "cfg-9160dc45a0171a10", + "cfg-a23a9ffc1f670139", + "cfg-b5007322fb2e1772", + "cfg-bfb83ff32b7fd921", + "cfg-ca71b56dc3189305", + "cfg-cad8341dbbff10ba", + "cfg-cc2c35944156e5d4", + "cfg-cd3974595f9db882", + "cfg-db87c67536995ab0", + "cfg-e91942abb5b4f847", + "cfg-ea5a23a40ad5ca50", + "cfg-f009e9e3891c06f4", + "cfg-f03e4e98c682240e", + "cfg-fad1b555f8b37235" + ], + "configurations": 30, + "eligible_processes": 90, + "recorded_attempts": 95 + }, + "full_permitted_fitting_pool": { + "configuration_ids": [ + "cfg-072bd7669d20f8fa", + "cfg-0d9ba9215ab5683e", + "cfg-0f540d3f6fe4661b", + "cfg-0f8a31ef7ac4c551", + "cfg-10644b5d0a82d5c3", + "cfg-1432bb8a4d9c3b8c", + "cfg-17e69bcf1b041bd9", + "cfg-18285c3cb50308d4", + "cfg-1c5278f8be90a50a", + "cfg-301e4dae7048c6ec", + "cfg-30f98456fdb322c1", + "cfg-338d9224eb9d2f3a", + "cfg-352f4dbc55fbcd3c", + "cfg-371ae37408089efe", + "cfg-399f6b038d6f65f9", + "cfg-3c86a11fdb8455cb", + "cfg-3f0f48a3c570a0c8", + "cfg-49d9bccfe873f5c6", + "cfg-5244d13eac3116ad", + "cfg-55811d23c5144bb7", + "cfg-5839ccf95c2c1722", + "cfg-5ff9621270e710ed", + "cfg-62f371a80968a0ba", + "cfg-6d54ac65d934f404", + "cfg-714c540f76d9525d", + "cfg-78c440d3891ddbc2", + "cfg-7a9a974958c57abc", + "cfg-7df2428553b76877", + "cfg-896685010703c0fb", + "cfg-89ca48ce88ea694e", + "cfg-8bd7050cd7f49e78", + "cfg-9160dc45a0171a10", + "cfg-966f5804dc918967", + "cfg-96ccf395b84393f7", + "cfg-97107ceba8bf0cbe", + "cfg-9c461415e6bb9702", + "cfg-a026693bae37d30b", + "cfg-a23a9ffc1f670139", + "cfg-a24e9cdd2f6143f2", + "cfg-a3987314cb2f9f12", + "cfg-b5007322fb2e1772", + "cfg-b8dd70ab682c13b2", + "cfg-bfb83ff32b7fd921", + "cfg-c0605c1607caa079", + "cfg-ca71b56dc3189305", + "cfg-cad8341dbbff10ba", + "cfg-cc2c35944156e5d4", + "cfg-cc900fb7731c3f49", + "cfg-cd3974595f9db882", + "cfg-d138048feeb8fe5d", + "cfg-db87c67536995ab0", + "cfg-e117f15dd5ccfb30", + "cfg-e2a205be8efba055", + "cfg-e91942abb5b4f847", + "cfg-ea5a23a40ad5ca50", + "cfg-ea92cccc91029465", + "cfg-f009e9e3891c06f4", + "cfg-f03e4e98c682240e", + "cfg-f75b68f479767290", + "cfg-fad1b555f8b37235" + ], + "configurations": 60, + "eligible_processes": 180, + "recorded_attempts": 185 + }, + "interpretation": "Historical evidence acquisition, not incremental GPU execution or a cold-start cost saving." + } + }, + "phi4": { + "model": { + "classifier_classes": [ + "above_margin", + "memory_failure", + "within_margin" + ], + "classifier_coefficients": [ + [ + 0.07422629343906341, + 0.3320326898806991, + 0.5888099791066441, + -0.690023591632588, + -0.690023591632588, + -0.33342164572127014, + -0.4440849706127423, + 0.11798607304841138 + ], + [ + 0.38352928531801883, + 0.2692475278523357, + 1.5424724944056163, + 0.42873181460562526, + 0.42873181460562526, + 0.7480346974433038, + 0.2324403633656105, + 0.19613952820402508 + ], + [ + -0.4577555787570821, + -0.6012802177330344, + -2.1312824735122606, + 0.26129177702696205, + 0.26129177702696205, + -0.41461305172203405, + 0.21164460724713202, + -0.3141256012524362 + ] + ], + "classifier_intercepts": [ + -0.8919563729871965, + -1.486601296471266, + 2.378557669458466 + ], + "classifier_iterations": [ + 29 + ], + "component_coefficients_gib": [ + 8.970654056937079, + 0.17127079066441622, + 0.0, + 24.20385419814186, + 0.0, + 0.012252539712286, + 0.2843587242012595, + 0.0, + 6.088904417365467 + ], + "component_divisors": [ + 5.971031188964844, + 1.0, + 32.0, + 5.971031188964844, + 1.1591873168945312, + 7.6875, + 13.91015625, + 5.971031188964844 + ], + "component_training_residual_norm_gib": 9.23717477585386, + "eligible_source_processes": 180, + "feature_names": [ + "actor_shards_gib", + "adapter_optimizer_gib", + "generation_budget_gib", + "generation_replica_gib", + "largest_unit_proxy_gib", + "checkpoint_boundary_volume_gib", + "vocabulary_output_volume_gib", + "resident_actor_shards_gib" + ], + "recorded_source_attempts": 218, + "regression_training_configuration_ids": [ + "cfg-0d9ba9215ab5683e", + "cfg-1c5278f8be90a50a", + "cfg-2afbe250af9cd3e1", + "cfg-30f98456fdb322c1", + "cfg-338d9224eb9d2f3a", + "cfg-399f6b038d6f65f9", + "cfg-3f0f48a3c570a0c8", + "cfg-5244d13eac3116ad", + "cfg-61c14de21a34d0e6", + "cfg-62f371a80968a0ba", + "cfg-6a4562db98b38d6f", + "cfg-6d54ac65d934f404", + "cfg-714c540f76d9525d", + "cfg-73c06764ba28a56b", + "cfg-78c440d3891ddbc2", + "cfg-7a0e81934d3398ee", + "cfg-7a9a974958c57abc", + "cfg-835067f7405e89a2", + "cfg-9160dc45a0171a10", + "cfg-93d4f41e600c9079", + "cfg-9758870c0a0b4d69", + "cfg-9f624ec22aa3dde0", + "cfg-a23a9ffc1f670139", + "cfg-a94bf7877633dfda", + "cfg-aa33c72e0a449b9c", + "cfg-acfa4acda15c3599", + "cfg-b5007322fb2e1772", + "cfg-b5352215babfeba6", + "cfg-b56fb353c2f2fa3e", + "cfg-b8e90bd69c9744ea", + "cfg-bae2977da7957f72", + "cfg-bfb83ff32b7fd921", + "cfg-c2a8e6e8e140e109", + "cfg-c3f0ad157d299d39", + "cfg-cc2c35944156e5d4", + "cfg-cc5f4f0c2e1256ba", + "cfg-cd3974595f9db882", + "cfg-d02503f8c9804a2d", + "cfg-d4a213515805a9a7", + "cfg-d9397d15f234d541", + "cfg-db87c67536995ab0", + "cfg-e91942abb5b4f847", + "cfg-f009e9e3891c06f4", + "cfg-f03e4e98c682240e", + "cfg-f439d2dd90137a80", + "cfg-fbe1c0422f98cb38", + "cfg-fd4d4789145230df", + "cfg-fd5bb0db68b0dae3" + ], + "scaler_mean": [ + 4.905035018920898, + 0.38968505859375, + 25.333333333333332, + 5.450038909912109, + 0.7671241760253906, + 3.989501953125, + 5.229177792867024, + 3.4516913096110025 + ], + "scaler_scale": [ + 1.1911485954279244, + 0.08742995950034278, + 7.180219742846007, + 0.5209922790527344, + 0.3920631408691406, + 3.49927388050063, + 5.420260600193311, + 2.3602687833978844 + ], + "training_configuration_ids": [ + "cfg-0d9ba9215ab5683e", + "cfg-1c5278f8be90a50a", + "cfg-2afbe250af9cd3e1", + "cfg-30f98456fdb322c1", + "cfg-32977d5aa42a41cb", + "cfg-338d9224eb9d2f3a", + "cfg-352f4dbc55fbcd3c", + "cfg-399f6b038d6f65f9", + "cfg-3dc042a103546eac", + "cfg-3f0f48a3c570a0c8", + "cfg-49d9bccfe873f5c6", + "cfg-5244d13eac3116ad", + "cfg-55811d23c5144bb7", + "cfg-5839ccf95c2c1722", + "cfg-61c14de21a34d0e6", + "cfg-62f371a80968a0ba", + "cfg-6a4562db98b38d6f", + "cfg-6d54ac65d934f404", + "cfg-714c540f76d9525d", + "cfg-73c06764ba28a56b", + "cfg-78c440d3891ddbc2", + "cfg-7a0e81934d3398ee", + "cfg-7a9a974958c57abc", + "cfg-7f23cab692546e3f", + "cfg-835067f7405e89a2", + "cfg-9160dc45a0171a10", + "cfg-93d4f41e600c9079", + "cfg-9758870c0a0b4d69", + "cfg-9f624ec22aa3dde0", + "cfg-a23a9ffc1f670139", + "cfg-a94bf7877633dfda", + "cfg-aa33c72e0a449b9c", + "cfg-acfa4acda15c3599", + "cfg-b5007322fb2e1772", + "cfg-b5352215babfeba6", + "cfg-b56fb353c2f2fa3e", + "cfg-b8e90bd69c9744ea", + "cfg-bae2977da7957f72", + "cfg-bfb83ff32b7fd921", + "cfg-c2a8e6e8e140e109", + "cfg-c3f0ad157d299d39", + "cfg-ca71b56dc3189305", + "cfg-cad8341dbbff10ba", + "cfg-cc2c35944156e5d4", + "cfg-cc5f4f0c2e1256ba", + "cfg-cd3974595f9db882", + "cfg-d02503f8c9804a2d", + "cfg-d4a213515805a9a7", + "cfg-d9397d15f234d541", + "cfg-db87c67536995ab0", + "cfg-e8893b12558f51dd", + "cfg-e91942abb5b4f847", + "cfg-ea5a23a40ad5ca50", + "cfg-f009e9e3891c06f4", + "cfg-f03e4e98c682240e", + "cfg-f439d2dd90137a80", + "cfg-fad1b555f8b37235", + "cfg-fbe1c0422f98cb38", + "cfg-fd4d4789145230df", + "cfg-fd5bb0db68b0dae3" + ] + }, + "predictions": [ + { + "configuration_id": "cfg-072bd7669d20f8fa", + "donor_configuration_id": "cfg-714c540f76d9525d", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 34212.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 32418.3154296875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-072bd7669d20f8fa", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 34942.29676716355, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 32418.3154296875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-072bd7669d20f8fa", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 32418.3154296875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-0f540d3f6fe4661b", + "donor_configuration_id": "cfg-30f98456fdb322c1", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 33446.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 32068.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-0f540d3f6fe4661b", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 35713.340429268916, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 32068.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-0f540d3f6fe4661b", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 32068.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-0f8a31ef7ac4c551", + "donor_configuration_id": "cfg-62f371a80968a0ba", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 21900.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-0f8a31ef7ac4c551", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 23396.36278109936, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-0f8a31ef7ac4c551", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-10644b5d0a82d5c3", + "donor_configuration_id": "cfg-fad1b555f8b37235", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 42308.630859375, + "predicted_startup_budget_failure": true + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-10644b5d0a82d5c3", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "memory_failure", + "predicted_peak_mib": 43985.899687758945, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 42308.630859375, + "predicted_startup_budget_failure": true + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-10644b5d0a82d5c3", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 42308.630859375, + "predicted_startup_budget_failure": true + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-1432bb8a4d9c3b8c", + "donor_configuration_id": "cfg-a23a9ffc1f670139", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 38316.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 36514.3154296875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-1432bb8a4d9c3b8c", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 38040.3901045257, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 36514.3154296875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-1432bb8a4d9c3b8c", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 36514.3154296875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-17e69bcf1b041bd9", + "donor_configuration_id": "cfg-cc2c35944156e5d4", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 33872.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 32418.3154296875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-17e69bcf1b041bd9", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 35176.66384053449, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 32418.3154296875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-17e69bcf1b041bd9", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 32418.3154296875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-18285c3cb50308d4", + "donor_configuration_id": "cfg-bfb83ff32b7fd921", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 21560.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-18285c3cb50308d4", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 23440.306607356415, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-18285c3cb50308d4", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-301e4dae7048c6ec", + "donor_configuration_id": "cfg-399f6b038d6f65f9", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 29840.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 28322.3154296875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-301e4dae7048c6ec", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 31902.795198144133, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 28322.3154296875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-301e4dae7048c6ec", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 28322.3154296875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-371ae37408089efe", + "donor_configuration_id": "cfg-f03e4e98c682240e", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 30180.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 28322.3154296875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-371ae37408089efe", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 31785.611661458654, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 28322.3154296875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-371ae37408089efe", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 28322.3154296875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-3c86a11fdb8455cb", + "donor_configuration_id": "cfg-352f4dbc55fbcd3c", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 42308.630859375, + "predicted_startup_budget_failure": true + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-3c86a11fdb8455cb", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "memory_failure", + "predicted_peak_mib": 44103.083224444425, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 42308.630859375, + "predicted_startup_budget_failure": true + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-3c86a11fdb8455cb", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 42308.630859375, + "predicted_startup_budget_failure": true + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-5ff9621270e710ed", + "donor_configuration_id": "cfg-78c440d3891ddbc2", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 37976.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 36514.3154296875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-5ff9621270e710ed", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 38274.75717789665, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 36514.3154296875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-5ff9621270e710ed", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 36514.3154296875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-7df2428553b76877", + "donor_configuration_id": "cfg-49d9bccfe873f5c6", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 40260.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-7df2428553b76877", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "memory_failure", + "predicted_peak_mib": 42143.89417736418, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 40260.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-7df2428553b76877", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 40260.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-896685010703c0fb", + "donor_configuration_id": "cfg-ca71b56dc3189305", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 42308.630859375, + "predicted_startup_budget_failure": true + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-896685010703c0fb", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "memory_failure", + "predicted_peak_mib": 43780.828498559364, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 42308.630859375, + "predicted_startup_budget_failure": true + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-896685010703c0fb", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 42308.630859375, + "predicted_startup_budget_failure": true + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-89ca48ce88ea694e", + "donor_configuration_id": "cfg-1c5278f8be90a50a", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 37682.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 36164.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-89ca48ce88ea694e", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "above_margin", + "predicted_peak_mib": 39045.80084000202, + "predicted_state": "above_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 36164.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-89ca48ce88ea694e", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 36164.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-8bd7050cd7f49e78", + "donor_configuration_id": "cfg-5244d13eac3116ad", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 21560.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-8bd7050cd7f49e78", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 18741.898804013283, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-8bd7050cd7f49e78", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-966f5804dc918967", + "donor_configuration_id": "cfg-5839ccf95c2c1722", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 40260.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-966f5804dc918967", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "memory_failure", + "predicted_peak_mib": 42202.48594570692, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 40260.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-966f5804dc918967", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 40260.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-96ccf395b84393f7", + "donor_configuration_id": "cfg-338d9224eb9d2f3a", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 33108.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 32068.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-96ccf395b84393f7", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 35830.52396595439, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 32068.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-96ccf395b84393f7", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 32068.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-97107ceba8bf0cbe", + "donor_configuration_id": "cfg-6d54ac65d934f404", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 33388.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 32068.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-97107ceba8bf0cbe", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 35749.960284483124, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 32068.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-97107ceba8bf0cbe", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 32068.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-9c461415e6bb9702", + "donor_configuration_id": "cfg-9160dc45a0171a10", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 21560.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-9c461415e6bb9702", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 18731.828343829373, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-9c461415e6bb9702", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-a026693bae37d30b", + "donor_configuration_id": "cfg-0d9ba9215ab5683e", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 37418.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 36164.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-a026693bae37d30b", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "above_margin", + "predicted_peak_mib": 38943.265245402225, + "predicted_state": "above_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 36164.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-a026693bae37d30b", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 36164.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-a24e9cdd2f6143f2", + "donor_configuration_id": "cfg-b5007322fb2e1772", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 21562.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-a24e9cdd2f6143f2", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 18738.23681849186, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-a24e9cdd2f6143f2", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-a3987314cb2f9f12", + "donor_configuration_id": "cfg-f009e9e3891c06f4", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 37478.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 36164.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-a3987314cb2f9f12", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 38870.02553497381, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 36164.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-a3987314cb2f9f12", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 36164.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-b8dd70ab682c13b2", + "donor_configuration_id": "cfg-cd3974595f9db882", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 21580.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-b8dd70ab682c13b2", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 23454.954549442096, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-b8dd70ab682c13b2", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-c0605c1607caa079", + "donor_configuration_id": "cfg-7a9a974958c57abc", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 37140.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 36164.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-c0605c1607caa079", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "above_margin", + "predicted_peak_mib": 39104.39260834476, + "predicted_state": "above_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 36164.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-c0605c1607caa079", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 36164.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-cc900fb7731c3f49", + "donor_configuration_id": "cfg-cad8341dbbff10ba", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 42308.630859375, + "predicted_startup_budget_failure": true + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-cc900fb7731c3f49", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "memory_failure", + "predicted_peak_mib": 43634.34907770252, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 42308.630859375, + "predicted_startup_budget_failure": true + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-cc900fb7731c3f49", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 42308.630859375, + "predicted_startup_budget_failure": true + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-d138048feeb8fe5d", + "donor_configuration_id": "cfg-55811d23c5144bb7", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 40260.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-d138048feeb8fe5d", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "memory_failure", + "predicted_peak_mib": 41968.11887233597, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 40260.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-d138048feeb8fe5d", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 40260.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-e117f15dd5ccfb30", + "donor_configuration_id": "cfg-e91942abb5b4f847", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 33436.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 32068.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-e117f15dd5ccfb30", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 35801.22808178302, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 32068.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-e117f15dd5ccfb30", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 32068.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-e2a205be8efba055", + "donor_configuration_id": "cfg-3f0f48a3c570a0c8", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 21860.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-e2a205be8efba055", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 18727.250861927598, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-e2a205be8efba055", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-ea92cccc91029465", + "donor_configuration_id": "cfg-ea5a23a40ad5ca50", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 40260.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-ea92cccc91029465", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "memory_failure", + "predicted_peak_mib": 42041.358582764384, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 40260.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-ea92cccc91029465", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 40260.630859375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-f75b68f479767290", + "donor_configuration_id": "cfg-db87c67536995ab0", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 21540.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-f75b68f479767290", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 23414.672708706465, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-f75b68f479767290", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + } + ], + "scores": { + "component_regression": { + "above_margin_among_approved": 0.15789473684210525, + "approval_rate_on_above_margin_targets": 0.5, + "approval_rate_on_memory_failure_targets": 0.0, + "approved": 19, + "approved_above_margin": 3, + "approved_memory_failure": 0, + "approved_within_margin": 16, + "completed_peak_mae_mib": 2331.2761415179616, + "completed_peak_mean_signed_error_mib": -2282.4289447817323, + "completed_peak_prediction_coverage": 22, + "completed_peak_worst_underestimate_mib": 7620.749138072402, + "confusion_matrix_predicted_rows_observed_columns": { + "above_margin": { + "above_margin": 3, + "memory_failure": 0, + "within_margin": 0 + }, + "memory_failure": { + "above_margin": 0, + "memory_failure": 8, + "within_margin": 0 + }, + "within_margin": { + "above_margin": 3, + "memory_failure": 0, + "within_margin": 16 + } + }, + "correct_three_state": 27, + "distinct_target_configurations": 30, + "memory_failure_among_approved": 0.0, + "queries": 30, + "rejected_within_margin": 0, + "rejection_rate_on_within_margin_targets": 0.0, + "startup_guard_changed_states": 0, + "target_above_margin": 6, + "target_memory_failure": 8, + "target_within_margin": 16, + "three_state_accuracy": 0.9, + "within_margin_approval_precision": 0.8421052631578947, + "within_margin_recall": 1.0 + }, + "donor_copy": { + "above_margin_among_approved": 0.2727272727272727, + "approval_rate_on_above_margin_targets": 1.0, + "approval_rate_on_memory_failure_targets": 0.0, + "approved": 22, + "approved_above_margin": 6, + "approved_memory_failure": 0, + "approved_within_margin": 16, + "completed_peak_mae_mib": 3064.818181818182, + "completed_peak_mean_signed_error_mib": -3064.818181818182, + "completed_peak_prediction_coverage": 22, + "completed_peak_worst_underestimate_mib": 4516.0, + "confusion_matrix_predicted_rows_observed_columns": { + "above_margin": { + "above_margin": 0, + "memory_failure": 0, + "within_margin": 0 + }, + "memory_failure": { + "above_margin": 0, + "memory_failure": 8, + "within_margin": 0 + }, + "within_margin": { + "above_margin": 6, + "memory_failure": 0, + "within_margin": 16 + } + }, + "correct_three_state": 24, + "distinct_target_configurations": 30, + "memory_failure_among_approved": 0.0, + "queries": 30, + "rejected_within_margin": 0, + "rejection_rate_on_within_margin_targets": 0.0, + "startup_guard_changed_states": 0, + "target_above_margin": 6, + "target_memory_failure": 8, + "target_within_margin": 16, + "three_state_accuracy": 0.8, + "within_margin_approval_precision": 0.7272727272727273, + "within_margin_recall": 1.0 + }, + "logistic": { + "above_margin_among_approved": 0.1111111111111111, + "approval_rate_on_above_margin_targets": 0.3333333333333333, + "approval_rate_on_memory_failure_targets": 0.0, + "approved": 18, + "approved_above_margin": 2, + "approved_memory_failure": 0, + "approved_within_margin": 16, + "completed_peak_mae_mib": null, + "completed_peak_mean_signed_error_mib": null, + "completed_peak_prediction_coverage": 0, + "completed_peak_worst_underestimate_mib": null, + "confusion_matrix_predicted_rows_observed_columns": { + "above_margin": { + "above_margin": 0, + "memory_failure": 0, + "within_margin": 0 + }, + "memory_failure": { + "above_margin": 4, + "memory_failure": 8, + "within_margin": 0 + }, + "within_margin": { + "above_margin": 2, + "memory_failure": 0, + "within_margin": 16 + } + }, + "correct_three_state": 24, + "distinct_target_configurations": 30, + "memory_failure_among_approved": 0.0, + "queries": 30, + "rejected_within_margin": 0, + "rejection_rate_on_within_margin_targets": 0.0, + "startup_guard_changed_states": 0, + "target_above_margin": 6, + "target_memory_failure": 8, + "target_within_margin": 16, + "three_state_accuracy": 0.8, + "within_margin_approval_precision": 0.8888888888888888, + "within_margin_recall": 1.0 + } + }, + "source_cost": { + "distinct_selected_donors": { + "configuration_ids": [ + "cfg-0d9ba9215ab5683e", + "cfg-1c5278f8be90a50a", + "cfg-30f98456fdb322c1", + "cfg-338d9224eb9d2f3a", + "cfg-352f4dbc55fbcd3c", + "cfg-399f6b038d6f65f9", + "cfg-3f0f48a3c570a0c8", + "cfg-49d9bccfe873f5c6", + "cfg-5244d13eac3116ad", + "cfg-55811d23c5144bb7", + "cfg-5839ccf95c2c1722", + "cfg-62f371a80968a0ba", + "cfg-6d54ac65d934f404", + "cfg-714c540f76d9525d", + "cfg-78c440d3891ddbc2", + "cfg-7a9a974958c57abc", + "cfg-9160dc45a0171a10", + "cfg-a23a9ffc1f670139", + "cfg-b5007322fb2e1772", + "cfg-bfb83ff32b7fd921", + "cfg-ca71b56dc3189305", + "cfg-cad8341dbbff10ba", + "cfg-cc2c35944156e5d4", + "cfg-cd3974595f9db882", + "cfg-db87c67536995ab0", + "cfg-e91942abb5b4f847", + "cfg-ea5a23a40ad5ca50", + "cfg-f009e9e3891c06f4", + "cfg-f03e4e98c682240e", + "cfg-fad1b555f8b37235" + ], + "configurations": 30, + "eligible_processes": 90, + "recorded_attempts": 95 + }, + "full_permitted_fitting_pool": { + "configuration_ids": [ + "cfg-0d9ba9215ab5683e", + "cfg-1c5278f8be90a50a", + "cfg-2afbe250af9cd3e1", + "cfg-30f98456fdb322c1", + "cfg-32977d5aa42a41cb", + "cfg-338d9224eb9d2f3a", + "cfg-352f4dbc55fbcd3c", + "cfg-399f6b038d6f65f9", + "cfg-3dc042a103546eac", + "cfg-3f0f48a3c570a0c8", + "cfg-49d9bccfe873f5c6", + "cfg-5244d13eac3116ad", + "cfg-55811d23c5144bb7", + "cfg-5839ccf95c2c1722", + "cfg-61c14de21a34d0e6", + "cfg-62f371a80968a0ba", + "cfg-6a4562db98b38d6f", + "cfg-6d54ac65d934f404", + "cfg-714c540f76d9525d", + "cfg-73c06764ba28a56b", + "cfg-78c440d3891ddbc2", + "cfg-7a0e81934d3398ee", + "cfg-7a9a974958c57abc", + "cfg-7f23cab692546e3f", + "cfg-835067f7405e89a2", + "cfg-9160dc45a0171a10", + "cfg-93d4f41e600c9079", + "cfg-9758870c0a0b4d69", + "cfg-9f624ec22aa3dde0", + "cfg-a23a9ffc1f670139", + "cfg-a94bf7877633dfda", + "cfg-aa33c72e0a449b9c", + "cfg-acfa4acda15c3599", + "cfg-b5007322fb2e1772", + "cfg-b5352215babfeba6", + "cfg-b56fb353c2f2fa3e", + "cfg-b8e90bd69c9744ea", + "cfg-bae2977da7957f72", + "cfg-bfb83ff32b7fd921", + "cfg-c2a8e6e8e140e109", + "cfg-c3f0ad157d299d39", + "cfg-ca71b56dc3189305", + "cfg-cad8341dbbff10ba", + "cfg-cc2c35944156e5d4", + "cfg-cc5f4f0c2e1256ba", + "cfg-cd3974595f9db882", + "cfg-d02503f8c9804a2d", + "cfg-d4a213515805a9a7", + "cfg-d9397d15f234d541", + "cfg-db87c67536995ab0", + "cfg-e8893b12558f51dd", + "cfg-e91942abb5b4f847", + "cfg-ea5a23a40ad5ca50", + "cfg-f009e9e3891c06f4", + "cfg-f03e4e98c682240e", + "cfg-f439d2dd90137a80", + "cfg-fad1b555f8b37235", + "cfg-fbe1c0422f98cb38", + "cfg-fd4d4789145230df", + "cfg-fd5bb0db68b0dae3" + ], + "configurations": 60, + "eligible_processes": 180, + "recorded_attempts": 218 + }, + "interpretation": "Historical evidence acquisition, not incremental GPU execution or a cold-start cost saving." + } + }, + "qwen25": { + "model": { + "classifier_classes": [ + "above_margin", + "memory_failure", + "within_margin" + ], + "classifier_coefficients": [ + [ + -0.3352920408175429, + 0.0973253209329928, + 0.007414745385724952, + 0.0873948235158526, + 0.0873948235158526, + -0.3586116756936474, + 0.35702395562620093, + 0.020410911748217624 + ], + [ + 0.6587363492704973, + 0.44983172301189883, + 2.387303659832254, + 0.15181352195464984, + 0.15181352195464984, + 0.7220439389603527, + 0.023926936990301115, + 0.3922498249241364 + ], + [ + -0.32344430845295435, + -0.5471570439448921, + -2.3947184052179784, + -0.23920834547050168, + -0.23920834547050168, + -0.3634322632667057, + -0.38095089261650206, + -0.41266073667235414 + ] + ], + "classifier_intercepts": [ + 0.8373066915702163, + -2.510969274908005, + 1.673662583337791 + ], + "classifier_iterations": [ + 31 + ], + "component_coefficients_gib": [ + 5.149785191520251, + 4.1680950782054325, + 0.0, + 27.777266859921, + 0.0, + 2.121369242254246, + 0.0, + 0.0, + 3.751792922025662 + ], + "component_divisors": [ + 7.317022323608398, + 1.0, + 32.0, + 7.317022323608398, + 2.289562225341797, + 9.28125, + 18.31640625, + 7.317022323608398 + ], + "component_training_residual_norm_gib": 11.225777418738016, + "eligible_source_processes": 180, + "feature_names": [ + "actor_shards_gib", + "adapter_optimizer_gib", + "generation_budget_gib", + "generation_replica_gib", + "largest_unit_proxy_gib", + "checkpoint_boundary_volume_gib", + "vocabulary_output_volume_gib", + "resident_actor_shards_gib" + ], + "recorded_source_attempts": 213, + "regression_training_configuration_ids": [ + "cfg-072bd7669d20f8fa", + "cfg-0f540d3f6fe4661b", + "cfg-0f8a31ef7ac4c551", + "cfg-1432bb8a4d9c3b8c", + "cfg-17e69bcf1b041bd9", + "cfg-18285c3cb50308d4", + "cfg-2afbe250af9cd3e1", + "cfg-301e4dae7048c6ec", + "cfg-371ae37408089efe", + "cfg-5ff9621270e710ed", + "cfg-61c14de21a34d0e6", + "cfg-6a4562db98b38d6f", + "cfg-73c06764ba28a56b", + "cfg-7a0e81934d3398ee", + "cfg-835067f7405e89a2", + "cfg-89ca48ce88ea694e", + "cfg-8bd7050cd7f49e78", + "cfg-93d4f41e600c9079", + "cfg-96ccf395b84393f7", + "cfg-97107ceba8bf0cbe", + "cfg-9758870c0a0b4d69", + "cfg-9c461415e6bb9702", + "cfg-9f624ec22aa3dde0", + "cfg-a026693bae37d30b", + "cfg-a24e9cdd2f6143f2", + "cfg-a3987314cb2f9f12", + "cfg-a94bf7877633dfda", + "cfg-aa33c72e0a449b9c", + "cfg-acfa4acda15c3599", + "cfg-b5352215babfeba6", + "cfg-b56fb353c2f2fa3e", + "cfg-b8dd70ab682c13b2", + "cfg-b8e90bd69c9744ea", + "cfg-bae2977da7957f72", + "cfg-c0605c1607caa079", + "cfg-c2a8e6e8e140e109", + "cfg-c3f0ad157d299d39", + "cfg-cc5f4f0c2e1256ba", + "cfg-d02503f8c9804a2d", + "cfg-d4a213515805a9a7", + "cfg-d9397d15f234d541", + "cfg-e117f15dd5ccfb30", + "cfg-e2a205be8efba055", + "cfg-f439d2dd90137a80", + "cfg-f75b68f479767290", + "cfg-fbe1c0422f98cb38", + "cfg-fd4d4789145230df", + "cfg-fd5bb0db68b0dae3" + ], + "scaler_mean": [ + 5.510731029510498, + 0.34365234375, + 25.333333333333332, + 6.123034477233887, + 1.3323116302490234, + 4.62884521484375, + 6.480675188700358, + 3.8779218355814615 + ], + "scaler_scale": [ + 1.64664142395129, + 0.0840529902301279, + 7.180219742846006, + 1.1939878463745117, + 0.9572505950927734, + 4.088784203098391, + 7.2392839819597725, + 2.7682842783970063 + ], + "training_configuration_ids": [ + "cfg-072bd7669d20f8fa", + "cfg-0f540d3f6fe4661b", + "cfg-0f8a31ef7ac4c551", + "cfg-10644b5d0a82d5c3", + "cfg-1432bb8a4d9c3b8c", + "cfg-17e69bcf1b041bd9", + "cfg-18285c3cb50308d4", + "cfg-2afbe250af9cd3e1", + "cfg-301e4dae7048c6ec", + "cfg-32977d5aa42a41cb", + "cfg-371ae37408089efe", + "cfg-3c86a11fdb8455cb", + "cfg-3dc042a103546eac", + "cfg-5ff9621270e710ed", + "cfg-61c14de21a34d0e6", + "cfg-6a4562db98b38d6f", + "cfg-73c06764ba28a56b", + "cfg-7a0e81934d3398ee", + "cfg-7df2428553b76877", + "cfg-7f23cab692546e3f", + "cfg-835067f7405e89a2", + "cfg-896685010703c0fb", + "cfg-89ca48ce88ea694e", + "cfg-8bd7050cd7f49e78", + "cfg-93d4f41e600c9079", + "cfg-966f5804dc918967", + "cfg-96ccf395b84393f7", + "cfg-97107ceba8bf0cbe", + "cfg-9758870c0a0b4d69", + "cfg-9c461415e6bb9702", + "cfg-9f624ec22aa3dde0", + "cfg-a026693bae37d30b", + "cfg-a24e9cdd2f6143f2", + "cfg-a3987314cb2f9f12", + "cfg-a94bf7877633dfda", + "cfg-aa33c72e0a449b9c", + "cfg-acfa4acda15c3599", + "cfg-b5352215babfeba6", + "cfg-b56fb353c2f2fa3e", + "cfg-b8dd70ab682c13b2", + "cfg-b8e90bd69c9744ea", + "cfg-bae2977da7957f72", + "cfg-c0605c1607caa079", + "cfg-c2a8e6e8e140e109", + "cfg-c3f0ad157d299d39", + "cfg-cc5f4f0c2e1256ba", + "cfg-cc900fb7731c3f49", + "cfg-d02503f8c9804a2d", + "cfg-d138048feeb8fe5d", + "cfg-d4a213515805a9a7", + "cfg-d9397d15f234d541", + "cfg-e117f15dd5ccfb30", + "cfg-e2a205be8efba055", + "cfg-e8893b12558f51dd", + "cfg-ea92cccc91029465", + "cfg-f439d2dd90137a80", + "cfg-f75b68f479767290", + "cfg-fbe1c0422f98cb38", + "cfg-fd4d4789145230df", + "cfg-fd5bb0db68b0dae3" + ] + }, + "predictions": [ + { + "configuration_id": "cfg-0d9ba9215ab5683e", + "donor_configuration_id": "cfg-c3f0ad157d299d39", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 36506.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 34786.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-0d9ba9215ab5683e", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 37879.73007685646, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 34786.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-0d9ba9215ab5683e", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "above_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 34786.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-1c5278f8be90a50a", + "donor_configuration_id": "cfg-aa33c72e0a449b9c", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 36758.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 34786.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-1c5278f8be90a50a", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 37879.73007685646, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 34786.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-1c5278f8be90a50a", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "above_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 34786.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-30f98456fdb322c1", + "donor_configuration_id": "cfg-b5352215babfeba6", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 32710.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 30690.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-30f98456fdb322c1", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 34324.239918786574, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 30690.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-30f98456fdb322c1", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 30690.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-338d9224eb9d2f3a", + "donor_configuration_id": "cfg-fbe1c0422f98cb38", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 32302.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 30690.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-338d9224eb9d2f3a", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 34324.239918786574, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 30690.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-338d9224eb9d2f3a", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 30690.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-352f4dbc55fbcd3c", + "donor_configuration_id": "cfg-e8893b12558f51dd", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 40930.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-352f4dbc55fbcd3c", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "memory_failure", + "predicted_peak_mib": 43212.96531396129, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 40930.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-352f4dbc55fbcd3c", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 40930.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-399f6b038d6f65f9", + "donor_configuration_id": "cfg-b8e90bd69c9744ea", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 29660.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 27633.16796875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-399f6b038d6f65f9", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 31015.18510939061, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 27633.16796875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-399f6b038d6f65f9", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 27633.16796875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-3f0f48a3c570a0c8", + "donor_configuration_id": "cfg-b56fb353c2f2fa3e", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 19418.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-3f0f48a3c570a0c8", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 20522.652278242113, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-3f0f48a3c570a0c8", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-49d9bccfe873f5c6", + "donor_configuration_id": "cfg-93d4f41e600c9079", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 40650.0, + "predicted_state": "above_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 38882.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-49d9bccfe873f5c6", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "memory_failure", + "predicted_peak_mib": 41435.22023492635, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 38882.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-49d9bccfe873f5c6", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "above_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 38882.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-5244d13eac3116ad", + "donor_configuration_id": "cfg-fd5bb0db68b0dae3", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 19050.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-5244d13eac3116ad", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 20522.652278242113, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-5244d13eac3116ad", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-55811d23c5144bb7", + "donor_configuration_id": "cfg-d02503f8c9804a2d", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 40750.0, + "predicted_state": "above_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 38882.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-55811d23c5144bb7", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "memory_failure", + "predicted_peak_mib": 41435.22023492635, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 38882.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-55811d23c5144bb7", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "above_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 38882.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-5839ccf95c2c1722", + "donor_configuration_id": "cfg-6a4562db98b38d6f", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 40382.0, + "predicted_state": "above_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 38882.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-5839ccf95c2c1722", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "memory_failure", + "predicted_peak_mib": 41435.22023492635, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 38882.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-5839ccf95c2c1722", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 38882.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-62f371a80968a0ba", + "donor_configuration_id": "cfg-fd4d4789145230df", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 21516.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-62f371a80968a0ba", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 25855.88751534694, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-62f371a80968a0ba", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-6d54ac65d934f404", + "donor_configuration_id": "cfg-a94bf7877633dfda", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 32534.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 30690.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-6d54ac65d934f404", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 34324.239918786574, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 30690.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-6d54ac65d934f404", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 30690.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-714c540f76d9525d", + "donor_configuration_id": "cfg-2afbe250af9cd3e1", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 34068.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31729.16796875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-714c540f76d9525d", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 34570.6752674605, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31729.16796875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-714c540f76d9525d", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31729.16796875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-78c440d3891ddbc2", + "donor_configuration_id": "cfg-f439d2dd90137a80", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 37740.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 35825.16796875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-78c440d3891ddbc2", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 38126.16542553039, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 35825.16796875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-78c440d3891ddbc2", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "above_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 35825.16796875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-7a9a974958c57abc", + "donor_configuration_id": "cfg-73c06764ba28a56b", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 36302.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 34786.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-7a9a974958c57abc", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 37879.73007685646, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 34786.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-7a9a974958c57abc", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "above_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 34786.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-9160dc45a0171a10", + "donor_configuration_id": "cfg-bae2977da7957f72", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 19070.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-9160dc45a0171a10", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 20522.652278242113, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-9160dc45a0171a10", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-a23a9ffc1f670139", + "donor_configuration_id": "cfg-9f624ec22aa3dde0", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 38148.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 35825.16796875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-a23a9ffc1f670139", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 38126.16542553039, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 35825.16796875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-a23a9ffc1f670139", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "above_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 35825.16796875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-b5007322fb2e1772", + "donor_configuration_id": "cfg-7a0e81934d3398ee", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 19070.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-b5007322fb2e1772", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 20522.652278242113, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-b5007322fb2e1772", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-bfb83ff32b7fd921", + "donor_configuration_id": "cfg-d9397d15f234d541", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 21302.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-bfb83ff32b7fd921", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 25855.88751534694, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-bfb83ff32b7fd921", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-ca71b56dc3189305", + "donor_configuration_id": "cfg-7f23cab692546e3f", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 40930.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-ca71b56dc3189305", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "memory_failure", + "predicted_peak_mib": 43212.96531396129, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 40930.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-ca71b56dc3189305", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 40930.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-cad8341dbbff10ba", + "donor_configuration_id": "cfg-32977d5aa42a41cb", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 40930.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-cad8341dbbff10ba", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "memory_failure", + "predicted_peak_mib": 43212.96531396129, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 40930.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-cad8341dbbff10ba", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 40930.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-cc2c35944156e5d4", + "donor_configuration_id": "cfg-c2a8e6e8e140e109", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 33660.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31729.16796875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-cc2c35944156e5d4", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 34570.6752674605, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31729.16796875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-cc2c35944156e5d4", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31729.16796875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-cd3974595f9db882", + "donor_configuration_id": "cfg-d4a213515805a9a7", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 21128.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-cd3974595f9db882", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 25855.88751534694, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-cd3974595f9db882", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-db87c67536995ab0", + "donor_configuration_id": "cfg-cc5f4f0c2e1256ba", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 21298.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-db87c67536995ab0", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 25855.88751534694, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-db87c67536995ab0", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": false, + "budget_plus_actor_mib": null, + "predicted_startup_budget_failure": null, + "reason": "resident-actor assumption does not apply" + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-e91942abb5b4f847", + "donor_configuration_id": "cfg-acfa4acda15c3599", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 32542.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 30690.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-e91942abb5b4f847", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 34324.239918786574, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 30690.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-e91942abb5b4f847", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 30690.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-ea5a23a40ad5ca50", + "donor_configuration_id": "cfg-61c14de21a34d0e6", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 40616.0, + "predicted_state": "above_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 38882.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-ea5a23a40ad5ca50", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "memory_failure", + "predicted_peak_mib": 41435.22023492635, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 38882.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-ea5a23a40ad5ca50", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "above_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 38882.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-f009e9e3891c06f4", + "donor_configuration_id": "cfg-835067f7405e89a2", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 38718.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 34786.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-f009e9e3891c06f4", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 37879.73007685646, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 34786.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-f009e9e3891c06f4", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "above_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 34786.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-f03e4e98c682240e", + "donor_configuration_id": "cfg-9758870c0a0b4d69", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 30068.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 27633.16796875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-f03e4e98c682240e", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 31015.18510939061, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 27633.16796875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-f03e4e98c682240e", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 27633.16796875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-fad1b555f8b37235", + "donor_configuration_id": "cfg-3dc042a103546eac", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 40930.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "cfg-fad1b555f8b37235", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "memory_failure", + "predicted_peak_mib": 43212.96531396129, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 40930.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "cfg-fad1b555f8b37235", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 40930.3359375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + } + ], + "scores": { + "component_regression": { + "above_margin_among_approved": 0.0, + "approval_rate_on_above_margin_targets": null, + "approval_rate_on_memory_failure_targets": 0.0, + "approved": 22, + "approved_above_margin": 0, + "approved_memory_failure": 0, + "approved_within_margin": 22, + "completed_peak_mae_mib": 1382.8428038497127, + "completed_peak_mean_signed_error_mib": 960.9132164405161, + "completed_peak_prediction_coverage": 22, + "completed_peak_worst_underestimate_mib": 1337.3477217578875, + "confusion_matrix_predicted_rows_observed_columns": { + "above_margin": { + "above_margin": 0, + "memory_failure": 0, + "within_margin": 0 + }, + "memory_failure": { + "above_margin": 0, + "memory_failure": 8, + "within_margin": 0 + }, + "within_margin": { + "above_margin": 0, + "memory_failure": 0, + "within_margin": 22 + } + }, + "correct_three_state": 30, + "distinct_target_configurations": 30, + "memory_failure_among_approved": 0.0, + "queries": 30, + "rejected_within_margin": 0, + "rejection_rate_on_within_margin_targets": 0.0, + "startup_guard_changed_states": 0, + "target_above_margin": 0, + "target_memory_failure": 8, + "target_within_margin": 22, + "three_state_accuracy": 1.0, + "within_margin_approval_precision": 1.0, + "within_margin_recall": 1.0 + }, + "donor_copy": { + "above_margin_among_approved": 0.0, + "approval_rate_on_above_margin_targets": null, + "approval_rate_on_memory_failure_targets": 0.0, + "approved": 22, + "approved_above_margin": 0, + "approved_memory_failure": 0, + "approved_within_margin": 22, + "completed_peak_mae_mib": 887.5454545454545, + "completed_peak_mean_signed_error_mib": -774.8181818181819, + "completed_peak_prediction_coverage": 22, + "completed_peak_worst_underestimate_mib": 2510.0, + "confusion_matrix_predicted_rows_observed_columns": { + "above_margin": { + "above_margin": 0, + "memory_failure": 4, + "within_margin": 0 + }, + "memory_failure": { + "above_margin": 0, + "memory_failure": 4, + "within_margin": 0 + }, + "within_margin": { + "above_margin": 0, + "memory_failure": 0, + "within_margin": 22 + } + }, + "correct_three_state": 26, + "distinct_target_configurations": 30, + "memory_failure_among_approved": 0.0, + "queries": 30, + "rejected_within_margin": 0, + "rejection_rate_on_within_margin_targets": 0.0, + "startup_guard_changed_states": 0, + "target_above_margin": 0, + "target_memory_failure": 8, + "target_within_margin": 22, + "three_state_accuracy": 0.8666666666666667, + "within_margin_approval_precision": 1.0, + "within_margin_recall": 1.0 + }, + "logistic": { + "above_margin_among_approved": 0.0, + "approval_rate_on_above_margin_targets": null, + "approval_rate_on_memory_failure_targets": 0.0, + "approved": 16, + "approved_above_margin": 0, + "approved_memory_failure": 0, + "approved_within_margin": 16, + "completed_peak_mae_mib": null, + "completed_peak_mean_signed_error_mib": null, + "completed_peak_prediction_coverage": 0, + "completed_peak_worst_underestimate_mib": null, + "confusion_matrix_predicted_rows_observed_columns": { + "above_margin": { + "above_margin": 0, + "memory_failure": 3, + "within_margin": 6 + }, + "memory_failure": { + "above_margin": 0, + "memory_failure": 5, + "within_margin": 0 + }, + "within_margin": { + "above_margin": 0, + "memory_failure": 0, + "within_margin": 16 + } + }, + "correct_three_state": 21, + "distinct_target_configurations": 30, + "memory_failure_among_approved": 0.0, + "queries": 30, + "rejected_within_margin": 6, + "rejection_rate_on_within_margin_targets": 0.2727272727272727, + "startup_guard_changed_states": 0, + "target_above_margin": 0, + "target_memory_failure": 8, + "target_within_margin": 22, + "three_state_accuracy": 0.7, + "within_margin_approval_precision": 1.0, + "within_margin_recall": 0.7272727272727273 + } + }, + "source_cost": { + "distinct_selected_donors": { + "configuration_ids": [ + "cfg-2afbe250af9cd3e1", + "cfg-32977d5aa42a41cb", + "cfg-3dc042a103546eac", + "cfg-61c14de21a34d0e6", + "cfg-6a4562db98b38d6f", + "cfg-73c06764ba28a56b", + "cfg-7a0e81934d3398ee", + "cfg-7f23cab692546e3f", + "cfg-835067f7405e89a2", + "cfg-93d4f41e600c9079", + "cfg-9758870c0a0b4d69", + "cfg-9f624ec22aa3dde0", + "cfg-a94bf7877633dfda", + "cfg-aa33c72e0a449b9c", + "cfg-acfa4acda15c3599", + "cfg-b5352215babfeba6", + "cfg-b56fb353c2f2fa3e", + "cfg-b8e90bd69c9744ea", + "cfg-bae2977da7957f72", + "cfg-c2a8e6e8e140e109", + "cfg-c3f0ad157d299d39", + "cfg-cc5f4f0c2e1256ba", + "cfg-d02503f8c9804a2d", + "cfg-d4a213515805a9a7", + "cfg-d9397d15f234d541", + "cfg-e8893b12558f51dd", + "cfg-f439d2dd90137a80", + "cfg-fbe1c0422f98cb38", + "cfg-fd4d4789145230df", + "cfg-fd5bb0db68b0dae3" + ], + "configurations": 30, + "eligible_processes": 90, + "recorded_attempts": 123 + }, + "full_permitted_fitting_pool": { + "configuration_ids": [ + "cfg-072bd7669d20f8fa", + "cfg-0f540d3f6fe4661b", + "cfg-0f8a31ef7ac4c551", + "cfg-10644b5d0a82d5c3", + "cfg-1432bb8a4d9c3b8c", + "cfg-17e69bcf1b041bd9", + "cfg-18285c3cb50308d4", + "cfg-2afbe250af9cd3e1", + "cfg-301e4dae7048c6ec", + "cfg-32977d5aa42a41cb", + "cfg-371ae37408089efe", + "cfg-3c86a11fdb8455cb", + "cfg-3dc042a103546eac", + "cfg-5ff9621270e710ed", + "cfg-61c14de21a34d0e6", + "cfg-6a4562db98b38d6f", + "cfg-73c06764ba28a56b", + "cfg-7a0e81934d3398ee", + "cfg-7df2428553b76877", + "cfg-7f23cab692546e3f", + "cfg-835067f7405e89a2", + "cfg-896685010703c0fb", + "cfg-89ca48ce88ea694e", + "cfg-8bd7050cd7f49e78", + "cfg-93d4f41e600c9079", + "cfg-966f5804dc918967", + "cfg-96ccf395b84393f7", + "cfg-97107ceba8bf0cbe", + "cfg-9758870c0a0b4d69", + "cfg-9c461415e6bb9702", + "cfg-9f624ec22aa3dde0", + "cfg-a026693bae37d30b", + "cfg-a24e9cdd2f6143f2", + "cfg-a3987314cb2f9f12", + "cfg-a94bf7877633dfda", + "cfg-aa33c72e0a449b9c", + "cfg-acfa4acda15c3599", + "cfg-b5352215babfeba6", + "cfg-b56fb353c2f2fa3e", + "cfg-b8dd70ab682c13b2", + "cfg-b8e90bd69c9744ea", + "cfg-bae2977da7957f72", + "cfg-c0605c1607caa079", + "cfg-c2a8e6e8e140e109", + "cfg-c3f0ad157d299d39", + "cfg-cc5f4f0c2e1256ba", + "cfg-cc900fb7731c3f49", + "cfg-d02503f8c9804a2d", + "cfg-d138048feeb8fe5d", + "cfg-d4a213515805a9a7", + "cfg-d9397d15f234d541", + "cfg-e117f15dd5ccfb30", + "cfg-e2a205be8efba055", + "cfg-e8893b12558f51dd", + "cfg-ea92cccc91029465", + "cfg-f439d2dd90137a80", + "cfg-f75b68f479767290", + "cfg-fbe1c0422f98cb38", + "cfg-fd4d4789145230df", + "cfg-fd5bb0db68b0dae3" + ], + "configurations": 60, + "eligible_processes": 180, + "recorded_attempts": 213 + }, + "interpretation": "Historical evidence acquisition, not incremental GPU execution or a cold-start cost saving." + } + } + }, + "numerical_environment": { + "fit_thread_limit": 1, + "numpy": "1.24.3", + "platform": "Linux-4.18.0-553.146.1.el8_10.x86_64-x86_64-with-glibc2.28", + "python": "3.11.4", + "scikit_learn": "1.3.0", + "scipy": "1.10.1", + "thread_environment": { + "MKL_NUM_THREADS": null, + "OMP_NUM_THREADS": null, + "OPENBLAS_NUM_THREADS": null + }, + "threadpools": [ + { + "filepath": "/mnt/proj3/open-35-44/verl/rl/.venv-reproduce/lib/python3.11/site-packages/scikit_learn.libs/libgomp-a34b3233.so.1.0.0", + "internal_api": "openmp", + "num_threads": 1, + "prefix": "libgomp", + "user_api": "openmp", + "version": null + }, + { + "architecture": "Zen", + "filepath": "/mnt/proj3/open-35-44/verl/rl/.venv-reproduce/lib/python3.11/site-packages/scipy.libs/libopenblasp-r0-41284840.3.18.so", + "internal_api": "openblas", + "num_threads": 1, + "prefix": "libopenblas", + "threading_layer": "pthreads", + "user_api": "blas", + "version": "0.3.18" + } + ] + } +} diff --git a/benchmark/estimation/targets.csv b/benchmark/estimation/targets.csv new file mode 100644 index 0000000000000000000000000000000000000000..48066c2e51fc2fe3745a88a243ce156b0c6d6409 --- /dev/null +++ b/benchmark/estimation/targets.csv @@ -0,0 +1,13 @@ +configuration_id,case_id,study,model_family,model_name,model_id,workload_id,workload_name,legacy_dataset_id,algorithm,adaptation,configuration_level,condition,gpu_count,rollout_tp_size,actor_micro_batch,rollout_logprob_micro_batch,ref_logprob_micro_batch,vllm_gpu_memory_utilization,parameter_offload,optimizer_offload,free_cache_engine,max_prompt_length,max_response_length,max_model_len,max_num_seqs,rollout_n,train_batch_size,train_max_samples,val_max_samples,total_training_steps,gpu_monitor_interval_ms,allocator_trace_enabled,allocator_trace_sync,save_freq,test_freq,val_before_train,resume_mode,max_actor_ckpt_to_keep,accelerator,device_capacity_mib,margin_limit_mib,peak_measurement +est-qwen25-7b-gsm8k-c2-2gpu,est-qwen25-7b-gsm8k-2gpu,prospective_estimation,qwen25_7b,Qwen2.5-7B-Instruct,Qwen/Qwen2.5-7B-Instruct,gsm8k,GSM8K mathematics,gsm8k,GRPO,LoRA,c2,,2,1,8,4,4,0.6,False,False,True,1024,2048,3072,64,2,32,32,32,1,100,True,1,-1,-1,False,disable,1,NVIDIA A100-SXM4-40GB,40960,38912,maximum externally sampled per-device memory +est-qwen25-7b-gsm8k-c2-4gpu,est-qwen25-7b-gsm8k-4gpu,prospective_estimation,qwen25_7b,Qwen2.5-7B-Instruct,Qwen/Qwen2.5-7B-Instruct,gsm8k,GSM8K mathematics,gsm8k,GRPO,LoRA,c2,,4,1,8,4,4,0.6,False,False,True,1024,2048,3072,64,2,32,32,32,1,100,True,1,-1,-1,False,disable,1,NVIDIA A100-SXM4-40GB,40960,38912,maximum externally sampled per-device memory +est-qwen25-7b-gsm8k-c3-2gpu,est-qwen25-7b-gsm8k-2gpu,prospective_estimation,qwen25_7b,Qwen2.5-7B-Instruct,Qwen/Qwen2.5-7B-Instruct,gsm8k,GSM8K mathematics,gsm8k,GRPO,LoRA,c3,,2,1,16,4,4,0.7,False,False,True,1024,2048,3072,64,2,32,32,32,1,100,True,1,-1,-1,False,disable,1,NVIDIA A100-SXM4-40GB,40960,38912,maximum externally sampled per-device memory +est-qwen25-7b-gsm8k-c3-4gpu,est-qwen25-7b-gsm8k-4gpu,prospective_estimation,qwen25_7b,Qwen2.5-7B-Instruct,Qwen/Qwen2.5-7B-Instruct,gsm8k,GSM8K mathematics,gsm8k,GRPO,LoRA,c3,,4,1,16,4,4,0.7,False,False,True,1024,2048,3072,64,2,32,32,32,1,100,True,1,-1,-1,False,disable,1,NVIDIA A100-SXM4-40GB,40960,38912,maximum externally sampled per-device memory +est-qwen25-7b-math-c2-2gpu,est-qwen25-7b-math-2gpu,prospective_estimation,qwen25_7b,Qwen2.5-7B-Instruct,Qwen/Qwen2.5-7B-Instruct,math,MATH mathematics,math,GRPO,LoRA,c2,,2,1,8,8,8,0.6,False,False,True,512,512,1024,128,2,64,64,64,1,100,True,1,-1,-1,False,disable,1,NVIDIA A100-SXM4-40GB,40960,38912,maximum externally sampled per-device memory +est-qwen25-7b-math-c2-4gpu,est-qwen25-7b-math-4gpu,prospective_estimation,qwen25_7b,Qwen2.5-7B-Instruct,Qwen/Qwen2.5-7B-Instruct,math,MATH mathematics,math,GRPO,LoRA,c2,,4,1,8,8,8,0.6,False,False,True,512,512,1024,128,2,64,64,64,1,100,True,1,-1,-1,False,disable,1,NVIDIA A100-SXM4-40GB,40960,38912,maximum externally sampled per-device memory +est-qwen25-7b-math-c3-2gpu,est-qwen25-7b-math-2gpu,prospective_estimation,qwen25_7b,Qwen2.5-7B-Instruct,Qwen/Qwen2.5-7B-Instruct,math,MATH mathematics,math,GRPO,LoRA,c3,,2,1,16,8,8,0.7,False,False,True,512,512,1024,128,2,64,64,64,1,100,True,1,-1,-1,False,disable,1,NVIDIA A100-SXM4-40GB,40960,38912,maximum externally sampled per-device memory +est-qwen25-7b-math-c3-4gpu,est-qwen25-7b-math-4gpu,prospective_estimation,qwen25_7b,Qwen2.5-7B-Instruct,Qwen/Qwen2.5-7B-Instruct,math,MATH mathematics,math,GRPO,LoRA,c3,,4,1,16,8,8,0.7,False,False,True,512,512,1024,128,2,64,64,64,1,100,True,1,-1,-1,False,disable,1,NVIDIA A100-SXM4-40GB,40960,38912,maximum externally sampled per-device memory +est-qwen25-7b-code_heavy_tail-c2-2gpu,est-qwen25-7b-code_heavy_tail-2gpu,prospective_estimation,qwen25_7b,Qwen2.5-7B-Instruct,Qwen/Qwen2.5-7B-Instruct,longer_prompt_code,Longer-prompt code,code_heavy_tail,GRPO,LoRA,c2,,2,1,8,4,4,0.6,False,False,True,1536,1024,2560,64,2,32,32,32,1,100,True,1,-1,-1,False,disable,1,NVIDIA A100-SXM4-40GB,40960,38912,maximum externally sampled per-device memory +est-qwen25-7b-code_heavy_tail-c2-4gpu,est-qwen25-7b-code_heavy_tail-4gpu,prospective_estimation,qwen25_7b,Qwen2.5-7B-Instruct,Qwen/Qwen2.5-7B-Instruct,longer_prompt_code,Longer-prompt code,code_heavy_tail,GRPO,LoRA,c2,,4,1,8,4,4,0.6,False,False,True,1536,1024,2560,64,2,32,32,32,1,100,True,1,-1,-1,False,disable,1,NVIDIA A100-SXM4-40GB,40960,38912,maximum externally sampled per-device memory +est-qwen25-7b-code_heavy_tail-c3-2gpu,est-qwen25-7b-code_heavy_tail-2gpu,prospective_estimation,qwen25_7b,Qwen2.5-7B-Instruct,Qwen/Qwen2.5-7B-Instruct,longer_prompt_code,Longer-prompt code,code_heavy_tail,GRPO,LoRA,c3,,2,1,16,4,4,0.7,False,False,True,1536,1024,2560,64,2,32,32,32,1,100,True,1,-1,-1,False,disable,1,NVIDIA A100-SXM4-40GB,40960,38912,maximum externally sampled per-device memory +est-qwen25-7b-code_heavy_tail-c3-4gpu,est-qwen25-7b-code_heavy_tail-4gpu,prospective_estimation,qwen25_7b,Qwen2.5-7B-Instruct,Qwen/Qwen2.5-7B-Instruct,longer_prompt_code,Longer-prompt code,code_heavy_tail,GRPO,LoRA,c3,,4,1,16,4,4,0.7,False,False,True,1536,1024,2560,64,2,32,32,32,1,100,True,1,-1,-1,False,disable,1,NVIDIA A100-SXM4-40GB,40960,38912,maximum externally sampled per-device memory diff --git a/benchmark/host_capacity/README.md b/benchmark/host_capacity/README.md new file mode 100644 index 0000000000000000000000000000000000000000..2ec8415dcef7f8f737b61a05c3fbc9bb28d506fd --- /dev/null +++ b/benchmark/host_capacity/README.md @@ -0,0 +1,87 @@ +# Larger-model test with expanded host resources + +This separate follow-up asks whether the original 7B predictions remain +correct under an expanded host-resource condition intended to avoid the +host-memory exhaustion observed in the original test. +It is part of a focused measurement study, not a cross-platform benchmark. + +The original panel could not label any four-GPU c2 configuration: seven +fresh attempts exhausted the 192 GiB host limit, and two earlier attempts +had setup failures. Those original records and unresolved labels remain +unchanged in `benchmark/estimation/results/`. + +## Fixed comparison + +The follow-up specifies all three original workloads at c2, with two or four +task GPUs and three new seeds. Its six configurations use eighteen process +slots in nine matched allocations. Each allocation reserves four A100 GPUs, +64 scheduler CPU units and 384 GiB of host memory. The two task-device counts +run in randomized order within that allocation. + +The model, data, one-step schedule, batching, precision, memory controls and +measurement settings are inherited. This is not a controlled comparison of +host-memory capacity against the original panel: the processor request also +changes. There are no retries, replacement labels or further memory increases. + +## Predictions and scores + +`predictions.json` copies every original c2 prediction field, changing only +the target identifier. `protocol.json` maps each new identifier to its +original prediction. No fit, donor selection or threshold is updated. +The five distinct donors selected for six targets are existing measurements, +not new source acquisitions. + +An approval is a predicted `within_margin` label. Read useful approvals, +memory-failure approvals and above-margin approvals together. Every repeated +label requires three eligible seeds; otherwise the target stays unresolved. +Always report unscored approvals and resolved coverage beside any score. +Neither a host kill nor a partial device peak establishes GPU feasibility. + +## Observed outcomes + +All eighteen invocations launched once. All nine two-GPU invocations have +validated generation-startup memory-admission failures; their three repeated +configuration labels are `memory_failure`. + +All nine four-GPU invocations exit with code zero and record returning +actor-update and weight-synchronization calls, but lack the required final +training-step record. They remain unresolved under the unchanged completion +rule. This is missing completion evidence, not proof that no training occurred +and not a validated GPU-memory outcome. No scheduler host OOM was recorded. + +Only three of six configurations are therefore scored. Donor copy +incorrectly approves all three and has three further unscored approvals. +Regression rejects the resolved failures but also has three unscored +approvals. The classifier approves nothing. An always-failure rule matches +the fitted methods on the failure-only resolved subset. There is no +within-margin target from which to estimate recall and no validated +completed peak for numerical error scoring. + +All nine scheduler allocations are accounted for: 5.79 reserved GPU-hours, +including unresolved work. Task-provisioned invocation time is 4.89 +GPU-hours; these are different, nonadditive scopes. The separate device +preflight used 0.08 GPU-hours. + +Numerical peak errors use only targets with three completed invocations. +The donor-copy peak is the donor's measured peak, whereas regression supplies +a fitted estimate. Failed-run peaks are not imputed as completed outcomes. +Paired numerical differences require both invocations to complete. + +## Files + +- `protocol.json`, `matrix.csv`, `targets.csv`: fixed settings and inventory. +- `predictions.json`, `study_freeze.json`: inherited predictions and the + separate pre-execution design seal. +- `execution_commits.json`, `execution_history.json`: immutable execution + source, preflight and scheduler history. +- `results/configurations_flat.csv`: readable outcomes beside predictions, + with seed coverage and unknown outcomes explicit. +- `results/processes.csv`, `results/pairs.csv`: every planned slot and pair. +- `results/evaluation.json`, `results/costs.json`: separate scores and cost + views, not pooled with the original experiment. + +The result files are populated only by independent raw-evidence +reconstruction. The browsing table contains outcomes and must not be used +as an input to blind prediction. Source acquisition, target invocation time, +whole scheduler allocation time and diagnostic preflight costs remain +separate. diff --git a/benchmark/host_capacity/allocation_accounting.json b/benchmark/host_capacity/allocation_accounting.json new file mode 100644 index 0000000000000000000000000000000000000000..73b12c3b45e957b3823ce36b47a2d1601d4b601a --- /dev/null +++ b/benchmark/host_capacity/allocation_accounting.json @@ -0,0 +1,434 @@ +[ + { + "accounting_schema_version": 1, + "allocated_gpu_count": 4, + "batch_ended_at_epoch": 1789500438.0, + "batch_started_at_epoch": 1789499856.0, + "batch_state": "FAILED", + "ended_at_epoch": 1789500438.0, + "job_id": "5055845", + "pair_id": "host-qwen25-7b-gsm8k-c2-s161", + "pair_ledger_sha256": "0d00f6eb5b648385af3f6ce1714c02b352b0a05dad2d539984f23e9783c471ce", + "pair_manifest_sha256": "cfa0de3b3cd08dfeacfdce5c3ebf3e3e2cef59bc2da204889e86042c4850708f", + "raw_record_provenance": { + "batch": { + "line": "5055845_8.batch|5055845.batch|batch|FAILED|2:0|2026-09-15T19:17:36|2026-09-15T19:27:18|582|cpu=64,gres/gpu=4,mem=384G,node=1|acn11|", + "line_number": 26, + "sha256": "e7f6404e1a7504049b9b02d7345d9fa709fc768ab8f695ea6dc134ac9d2379cd" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5055845_8|5055845|rlvram-host-capacity|FAILED|2:0|2026-09-15T19:17:36|2026-09-15T19:27:18|582|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn11|JobArrayTaskLimit", + "line_number": 25, + "sha256": "b27b72662a890aed168ae0be010cf62255e8950b0dd3c6ba630997d0e2f5f699" + }, + "source_sha256": "5c672842d7ad1ef71e6164ae76ba2e74400c45c2511a66d4139489f34664d2eb", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_exit_code": "2:0", + "scheduler_job_id": "5055845_8", + "scheduler_node_list": "acn11", + "scheduler_scope": "outer_job_record", + "scheduler_state": "FAILED", + "started_at_epoch": 1789499856.0, + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1 + }, + { + "accounting_schema_version": 1, + "allocated_gpu_count": 4, + "batch_ended_at_epoch": 1789499099.0, + "batch_started_at_epoch": 1789498576.0, + "batch_state": "FAILED", + "ended_at_epoch": 1789499099.0, + "job_id": "5055846", + "pair_id": "host-qwen25-7b-code_heavy_tail-c2-s161", + "pair_ledger_sha256": "333c3a9409bf517fd2ca41c6b17298d3c18ffde6d0780bbd7bd4b980953a7a54", + "pair_manifest_sha256": "11f432ad69e4f19a7e964601235286986f30fb97862cb358860a3b4a45b81ed4", + "raw_record_provenance": { + "batch": { + "line": "5055845_0.batch|5055846.batch|batch|FAILED|2:0|2026-09-15T18:56:16|2026-09-15T19:04:59|523|cpu=64,gres/gpu=4,mem=384G,node=1|acn08|", + "line_number": 2, + "sha256": "8d2ec6009fdc342af9c9ecf6533aeb4e2b9545d9e8f7089fd15995840f55e33a" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5055845_0|5055846|rlvram-host-capacity|FAILED|2:0|2026-09-15T18:56:16|2026-09-15T19:04:59|523|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn08|Dependency", + "line_number": 1, + "sha256": "cab718f549fe31355b9d4784794f53a0f579d9cb7b237d07eef3a62982287423" + }, + "source_sha256": "5c672842d7ad1ef71e6164ae76ba2e74400c45c2511a66d4139489f34664d2eb", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_exit_code": "2:0", + "scheduler_job_id": "5055845_0", + "scheduler_node_list": "acn08", + "scheduler_scope": "outer_job_record", + "scheduler_state": "FAILED", + "started_at_epoch": 1789498576.0, + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1 + }, + { + "accounting_schema_version": 1, + "allocated_gpu_count": 4, + "batch_ended_at_epoch": 1789499295.0, + "batch_started_at_epoch": 1789498576.0, + "batch_state": "FAILED", + "ended_at_epoch": 1789499295.0, + "job_id": "5055847", + "pair_id": "host-qwen25-7b-gsm8k-c2-s163", + "pair_ledger_sha256": "cde48950a284805dca7150f75182dab4ab0a034cd0b41efeb9fe2b83512a8dd9", + "pair_manifest_sha256": "10a666b083e54c59a9b1ac6144602a9e67fe987c366ba739ea87679570ff7d22", + "raw_record_provenance": { + "batch": { + "line": "5055845_1.batch|5055847.batch|batch|FAILED|2:0|2026-09-15T18:56:16|2026-09-15T19:08:15|719|cpu=64,gres/gpu=4,mem=384G,node=1|acn11|", + "line_number": 5, + "sha256": "8994c5478bad6a55e4b764cd4bc6332b8574a50a3c0742f2b6ad25d30f9a1e47" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5055845_1|5055847|rlvram-host-capacity|FAILED|2:0|2026-09-15T18:56:16|2026-09-15T19:08:15|719|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn11|Dependency", + "line_number": 4, + "sha256": "675fca75c0ea2b4ea5181294d050795a790f58ff267af91ef1c7874598d151a5" + }, + "source_sha256": "5c672842d7ad1ef71e6164ae76ba2e74400c45c2511a66d4139489f34664d2eb", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_exit_code": "2:0", + "scheduler_job_id": "5055845_1", + "scheduler_node_list": "acn11", + "scheduler_scope": "outer_job_record", + "scheduler_state": "FAILED", + "started_at_epoch": 1789498576.0, + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1 + }, + { + "accounting_schema_version": 1, + "allocated_gpu_count": 4, + "batch_ended_at_epoch": 1789499173.0, + "batch_started_at_epoch": 1789498576.0, + "batch_state": "FAILED", + "ended_at_epoch": 1789499173.0, + "job_id": "5055848", + "pair_id": "host-qwen25-7b-code_heavy_tail-c2-s162", + "pair_ledger_sha256": "1c00aa35420037ed210fc575999f0151051f8b4627b06d67904904e489f0c6dd", + "pair_manifest_sha256": "ba5822d60980ed2368ce46612583322f5385a116a98a85abeb700f1080e19e5a", + "raw_record_provenance": { + "batch": { + "line": "5055845_2.batch|5055848.batch|batch|FAILED|2:0|2026-09-15T18:56:16|2026-09-15T19:06:13|597|cpu=64,gres/gpu=4,mem=384G,node=1|acn14|", + "line_number": 8, + "sha256": "1d79ee14f8d1087a12693fe62fa8ad1bc22c64b67998b68463e8ca5e5f82aafb" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5055845_2|5055848|rlvram-host-capacity|FAILED|2:0|2026-09-15T18:56:16|2026-09-15T19:06:13|597|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn14|Dependency", + "line_number": 7, + "sha256": "1e95b564ea1bfd157b77d0238b8e6ebecc1b8a9ef90fdf381d1b6ea3975fc62c" + }, + "source_sha256": "5c672842d7ad1ef71e6164ae76ba2e74400c45c2511a66d4139489f34664d2eb", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_exit_code": "2:0", + "scheduler_job_id": "5055845_2", + "scheduler_node_list": "acn14", + "scheduler_scope": "outer_job_record", + "scheduler_state": "FAILED", + "started_at_epoch": 1789498576.0, + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1 + }, + { + "accounting_schema_version": 1, + "allocated_gpu_count": 4, + "batch_ended_at_epoch": 1789499600.0, + "batch_started_at_epoch": 1789499099.0, + "batch_state": "FAILED", + "ended_at_epoch": 1789499600.0, + "job_id": "5055858", + "pair_id": "host-qwen25-7b-gsm8k-c2-s162", + "pair_ledger_sha256": "4dcf9a91ae6bbba137a4afab1907896accd528c333c05f9396998563518b7eaa", + "pair_manifest_sha256": "cb05d94260ce1ee8a79015825d072921a10254d9d914a0df3e5c51f3be375fef", + "raw_record_provenance": { + "batch": { + "line": "5055845_3.batch|5055858.batch|batch|FAILED|2:0|2026-09-15T19:04:59|2026-09-15T19:13:20|501|cpu=64,gres/gpu=4,mem=384G,node=1|acn08|", + "line_number": 11, + "sha256": "377935db50f84bc8ca9f7c3106dcd288bdcddf8c9971b0f7b0e9b8128780adb4" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5055845_3|5055858|rlvram-host-capacity|FAILED|2:0|2026-09-15T19:04:59|2026-09-15T19:13:20|501|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn08|JobArrayTaskLimit", + "line_number": 10, + "sha256": "954be8b27acd69a1f217ea3268ca6b29935b717fe64dabd63dd896006f78afd3" + }, + "source_sha256": "5c672842d7ad1ef71e6164ae76ba2e74400c45c2511a66d4139489f34664d2eb", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_exit_code": "2:0", + "scheduler_job_id": "5055845_3", + "scheduler_node_list": "acn08", + "scheduler_scope": "outer_job_record", + "scheduler_state": "FAILED", + "started_at_epoch": 1789499099.0, + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1 + }, + { + "accounting_schema_version": 1, + "allocated_gpu_count": 4, + "batch_ended_at_epoch": 1789499776.0, + "batch_started_at_epoch": 1789499173.0, + "batch_state": "FAILED", + "ended_at_epoch": 1789499776.0, + "job_id": "5055859", + "pair_id": "host-qwen25-7b-math-c2-s162", + "pair_ledger_sha256": "dc8d64e4c1acaab7d56a2609e8325cf9f18b62861342caf38bd2d2458019e1bb", + "pair_manifest_sha256": "29d4201db5096e9951dd1c70a5df4b57c945d95dfe55d30830b1b4db32ab4186", + "raw_record_provenance": { + "batch": { + "line": "5055845_4.batch|5055859.batch|batch|FAILED|2:0|2026-09-15T19:06:13|2026-09-15T19:16:16|603|cpu=64,gres/gpu=4,mem=384G,node=1|acn14|", + "line_number": 14, + "sha256": "1efa723eb021c52a986801f246f23bd4f75fb9aef2d71ae5df80f03faf660153" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5055845_4|5055859|rlvram-host-capacity|FAILED|2:0|2026-09-15T19:06:13|2026-09-15T19:16:16|603|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn14|JobArrayTaskLimit", + "line_number": 13, + "sha256": "4ef776a727e24cd59151915df8147adc8aaf6c94789c3ac82441b007bcd9d5f4" + }, + "source_sha256": "5c672842d7ad1ef71e6164ae76ba2e74400c45c2511a66d4139489f34664d2eb", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_exit_code": "2:0", + "scheduler_job_id": "5055845_4", + "scheduler_node_list": "acn14", + "scheduler_scope": "outer_job_record", + "scheduler_state": "FAILED", + "started_at_epoch": 1789499173.0, + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1 + }, + { + "accounting_schema_version": 1, + "allocated_gpu_count": 4, + "batch_ended_at_epoch": 1789499856.0, + "batch_started_at_epoch": 1789499296.0, + "batch_state": "FAILED", + "ended_at_epoch": 1789499856.0, + "job_id": "5055860", + "pair_id": "host-qwen25-7b-code_heavy_tail-c2-s163", + "pair_ledger_sha256": "9d18e012ffc9f2064dcbad10dafa59b750ce161497eec51e0e6bc176ce46c3c4", + "pair_manifest_sha256": "d9475a95dfbf2600d351db83bfc49b45774c5e13c2ab968d21a2168b43d9573f", + "raw_record_provenance": { + "batch": { + "line": "5055845_5.batch|5055860.batch|batch|FAILED|2:0|2026-09-15T19:08:16|2026-09-15T19:17:36|560|cpu=64,gres/gpu=4,mem=384G,node=1|acn11|", + "line_number": 17, + "sha256": "ed5d5207ac56d624b57b1eeee06eb615b7dd9e9db639df5a8eaedaa62faff098" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5055845_5|5055860|rlvram-host-capacity|FAILED|2:0|2026-09-15T19:08:16|2026-09-15T19:17:36|560|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn11|JobArrayTaskLimit", + "line_number": 16, + "sha256": "ad137608e28ca1b487e56fd36287ffa9b443c14332b3cd5515b473ef7e98c33d" + }, + "source_sha256": "5c672842d7ad1ef71e6164ae76ba2e74400c45c2511a66d4139489f34664d2eb", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_exit_code": "2:0", + "scheduler_job_id": "5055845_5", + "scheduler_node_list": "acn11", + "scheduler_scope": "outer_job_record", + "scheduler_state": "FAILED", + "started_at_epoch": 1789499296.0, + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1 + }, + { + "accounting_schema_version": 1, + "allocated_gpu_count": 4, + "batch_ended_at_epoch": 1789500198.0, + "batch_started_at_epoch": 1789499600.0, + "batch_state": "FAILED", + "ended_at_epoch": 1789500198.0, + "job_id": "5055863", + "pair_id": "host-qwen25-7b-math-c2-s163", + "pair_ledger_sha256": "045f78ee36bcaec336b9f1ab0a0b08096132b5eff8d76089ceaed23da8eeb80a", + "pair_manifest_sha256": "d0fb1b825aaa3e394285a542e31a674611288684d7ef2a2f5eae6a96b4659e95", + "raw_record_provenance": { + "batch": { + "line": "5055845_6.batch|5055863.batch|batch|FAILED|2:0|2026-09-15T19:13:20|2026-09-15T19:23:18|598|cpu=64,gres/gpu=4,mem=384G,node=1|acn08|", + "line_number": 20, + "sha256": "37497680614f9d2c4f625cf5c41f21852e6fe61b3b902a943af6bb45242c370e" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5055845_6|5055863|rlvram-host-capacity|FAILED|2:0|2026-09-15T19:13:20|2026-09-15T19:23:18|598|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn08|JobArrayTaskLimit", + "line_number": 19, + "sha256": "b9800e065af258b257be749953bc183b21b0d6bd2e601d993ac7a57846e651dc" + }, + "source_sha256": "5c672842d7ad1ef71e6164ae76ba2e74400c45c2511a66d4139489f34664d2eb", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_exit_code": "2:0", + "scheduler_job_id": "5055845_6", + "scheduler_node_list": "acn08", + "scheduler_scope": "outer_job_record", + "scheduler_state": "FAILED", + "started_at_epoch": 1789499600.0, + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1 + }, + { + "accounting_schema_version": 1, + "allocated_gpu_count": 4, + "batch_ended_at_epoch": 1789500300.0, + "batch_started_at_epoch": 1789499776.0, + "batch_state": "FAILED", + "ended_at_epoch": 1789500300.0, + "job_id": "5055864", + "pair_id": "host-qwen25-7b-math-c2-s161", + "pair_ledger_sha256": "35af3985e7adf8837ec64dd3380dbac41b910e49a59dcb258c793c7bf4e0b53e", + "pair_manifest_sha256": "da4248ba78237ac4377771fbd810c138222c7d89cc1960cd484ada584e184ab9", + "raw_record_provenance": { + "batch": { + "line": "5055845_7.batch|5055864.batch|batch|FAILED|2:0|2026-09-15T19:16:16|2026-09-15T19:25:00|524|cpu=64,gres/gpu=4,mem=384G,node=1|acn14|", + "line_number": 23, + "sha256": "3c37ebe8e3c09ebc1d6d665c51832daa4b99454c7871c3ecc08f4b032a77b198" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5055845_7|5055864|rlvram-host-capacity|FAILED|2:0|2026-09-15T19:16:16|2026-09-15T19:25:00|524|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn14|JobArrayTaskLimit", + "line_number": 22, + "sha256": "10c525a57304fd6119adae3491797a97fe9892d43e9c05802bd858761d074a53" + }, + "source_sha256": "5c672842d7ad1ef71e6164ae76ba2e74400c45c2511a66d4139489f34664d2eb", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_exit_code": "2:0", + "scheduler_job_id": "5055845_7", + "scheduler_node_list": "acn14", + "scheduler_scope": "outer_job_record", + "scheduler_state": "FAILED", + "started_at_epoch": 1789499776.0, + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1 + } +] diff --git a/benchmark/host_capacity/allocation_accounting_sacct.psv b/benchmark/host_capacity/allocation_accounting_sacct.psv new file mode 100644 index 0000000000000000000000000000000000000000..3d2ae28cfee92ee1c05c33a5e92c2db121fd01e7 --- /dev/null +++ b/benchmark/host_capacity/allocation_accounting_sacct.psv @@ -0,0 +1,27 @@ +5055845_0|5055846|rlvram-host-capacity|FAILED|2:0|2026-09-15T18:56:16|2026-09-15T19:04:59|523|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn08|Dependency +5055845_0.batch|5055846.batch|batch|FAILED|2:0|2026-09-15T18:56:16|2026-09-15T19:04:59|523|cpu=64,gres/gpu=4,mem=384G,node=1|acn08| +5055845_0.extern|5055846.extern|extern|COMPLETED|0:0|2026-09-15T18:56:16|2026-09-15T19:04:59|523|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn08| +5055845_1|5055847|rlvram-host-capacity|FAILED|2:0|2026-09-15T18:56:16|2026-09-15T19:08:15|719|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn11|Dependency +5055845_1.batch|5055847.batch|batch|FAILED|2:0|2026-09-15T18:56:16|2026-09-15T19:08:15|719|cpu=64,gres/gpu=4,mem=384G,node=1|acn11| +5055845_1.extern|5055847.extern|extern|COMPLETED|0:0|2026-09-15T18:56:16|2026-09-15T19:08:15|719|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn11| +5055845_2|5055848|rlvram-host-capacity|FAILED|2:0|2026-09-15T18:56:16|2026-09-15T19:06:13|597|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn14|Dependency +5055845_2.batch|5055848.batch|batch|FAILED|2:0|2026-09-15T18:56:16|2026-09-15T19:06:13|597|cpu=64,gres/gpu=4,mem=384G,node=1|acn14| +5055845_2.extern|5055848.extern|extern|COMPLETED|0:0|2026-09-15T18:56:16|2026-09-15T19:06:13|597|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn14| +5055845_3|5055858|rlvram-host-capacity|FAILED|2:0|2026-09-15T19:04:59|2026-09-15T19:13:20|501|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn08|JobArrayTaskLimit +5055845_3.batch|5055858.batch|batch|FAILED|2:0|2026-09-15T19:04:59|2026-09-15T19:13:20|501|cpu=64,gres/gpu=4,mem=384G,node=1|acn08| +5055845_3.extern|5055858.extern|extern|COMPLETED|0:0|2026-09-15T19:04:59|2026-09-15T19:13:20|501|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn08| +5055845_4|5055859|rlvram-host-capacity|FAILED|2:0|2026-09-15T19:06:13|2026-09-15T19:16:16|603|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn14|JobArrayTaskLimit +5055845_4.batch|5055859.batch|batch|FAILED|2:0|2026-09-15T19:06:13|2026-09-15T19:16:16|603|cpu=64,gres/gpu=4,mem=384G,node=1|acn14| +5055845_4.extern|5055859.extern|extern|COMPLETED|0:0|2026-09-15T19:06:13|2026-09-15T19:16:16|603|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn14| +5055845_5|5055860|rlvram-host-capacity|FAILED|2:0|2026-09-15T19:08:16|2026-09-15T19:17:36|560|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn11|JobArrayTaskLimit +5055845_5.batch|5055860.batch|batch|FAILED|2:0|2026-09-15T19:08:16|2026-09-15T19:17:36|560|cpu=64,gres/gpu=4,mem=384G,node=1|acn11| +5055845_5.extern|5055860.extern|extern|COMPLETED|0:0|2026-09-15T19:08:16|2026-09-15T19:17:36|560|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn11| +5055845_6|5055863|rlvram-host-capacity|FAILED|2:0|2026-09-15T19:13:20|2026-09-15T19:23:18|598|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn08|JobArrayTaskLimit +5055845_6.batch|5055863.batch|batch|FAILED|2:0|2026-09-15T19:13:20|2026-09-15T19:23:18|598|cpu=64,gres/gpu=4,mem=384G,node=1|acn08| +5055845_6.extern|5055863.extern|extern|COMPLETED|0:0|2026-09-15T19:13:20|2026-09-15T19:23:18|598|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn08| +5055845_7|5055864|rlvram-host-capacity|FAILED|2:0|2026-09-15T19:16:16|2026-09-15T19:25:00|524|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn14|JobArrayTaskLimit +5055845_7.batch|5055864.batch|batch|FAILED|2:0|2026-09-15T19:16:16|2026-09-15T19:25:00|524|cpu=64,gres/gpu=4,mem=384G,node=1|acn14| +5055845_7.extern|5055864.extern|extern|COMPLETED|0:0|2026-09-15T19:16:16|2026-09-15T19:25:00|524|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn14| +5055845_8|5055845|rlvram-host-capacity|FAILED|2:0|2026-09-15T19:17:36|2026-09-15T19:27:18|582|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn11|JobArrayTaskLimit +5055845_8.batch|5055845.batch|batch|FAILED|2:0|2026-09-15T19:17:36|2026-09-15T19:27:18|582|cpu=64,gres/gpu=4,mem=384G,node=1|acn11| +5055845_8.extern|5055845.extern|extern|COMPLETED|0:0|2026-09-15T19:17:36|2026-09-15T19:27:18|582|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn11| diff --git a/benchmark/host_capacity/execution_commits.json b/benchmark/host_capacity/execution_commits.json new file mode 100644 index 0000000000000000000000000000000000000000..21b824a18abc05da05a6ed22c18198db8c427ad3 --- /dev/null +++ b/benchmark/host_capacity/execution_commits.json @@ -0,0 +1,11 @@ +{ + "host-qwen25-7b-code_heavy_tail-c2-s161": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-code_heavy_tail-c2-s162": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-code_heavy_tail-c2-s163": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-gsm8k-c2-s161": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-gsm8k-c2-s162": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-gsm8k-c2-s163": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-math-c2-s161": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-math-c2-s162": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-math-c2-s163": "120e29a146294722a6668f55a1e7fed2878eb62a" +} diff --git a/benchmark/host_capacity/execution_diagnostic.md b/benchmark/host_capacity/execution_diagnostic.md new file mode 100644 index 0000000000000000000000000000000000000000..ff90fe0c3329e8f35ff3820ad7acdc54d44864e5 --- /dev/null +++ b/benchmark/host_capacity/execution_diagnostic.md @@ -0,0 +1,139 @@ +# Host-capacity diagnostic and final collection — 2026-09-15 + +All nine array elements of `5055845` ended `FAILED 2:0`, with no scheduler `OUT_OF_MEMORY` state. The last element, index 8 / actual job `5055845`, ended at **19:27:18 UTC**. + +Final reconstruction: **18 launches; 9 eligible 2-GPU rollout-initialization memory failures; 9 unresolved 4-GPU periods; zero eligible completions.** Three of six configurations resolve, all 2-GPU failures. Evaluation is restricted to that resolved subset. No eligibility, predictor, runtime, or raw-evidence changes were made; no experiments or retries were launched. + +## Diagnosis and limits + +The evidence does **not** support “no training was performed.” In every 4-GPU period, all four ranks recorded an `actor_update` call returning without an exception, followed by a separate post-update `weight_sync` call returning without an exception. The recorded phase sequence reaches step-1 actor update, weight synchronization, then idle. These are completed calls, not independent proof that optimizer parameters changed or that the entire required terminal protocol completed. + +All nine 4-GPU dispatches specify `TOTAL_TRAINING_STEPS=1`, `TRAINER_LOGGER=["console"]`, and `RESUME_MODE=disable`. The examined resolved trainer configuration agrees; validation and checkpointing are disabled. Every 4-GPU payload exited `0`, but every period has: + +```text +status: infrastructure_invalid +validation_error: ValueError: Successful exit lacks the exact one-step terminal event +``` + +Neither a `step:` metric line nor `training/global_step:` appears in any of the nine retained training logs. All nine post-period allocation-idle checks passed. + +Pinned source root: + +```text +/mnt/proj3/open-35-44/verl/rl/job_sources/host-capacity-120e29a +commit: 120e29a146294722a6668f55a1e7fed2878eb62a +``` + +Relevant control flow, relative to that root: + +- `sppo_replay/allocator_trace.py`: the decorator records `error_type` in `finally`, catches and rethrows `BaseException`, and does not suppress failures. All 36 actor-update and 36 subsequent synchronization records examined have empty `error_type`. +- `instrumented_python_packages/verl/workers/engine_workers.py:719`: the actor-update decorator encloses `update_actor`, which calls `self.actor.train_mini_batch`. This is the training-call path, not merely a phase label. +- `instrumented_python_packages/verl/trainer/ppo/v1/trainer_sync.py:35`: `on_step_end` performs weight synchronization before marking idle. +- `instrumented_python_packages/verl/trainer/ppo/v1/trainer_base.py:497`: after synchronization, the normal path computes metrics, logs them at line 508, updates progress at line 513, and returns at line 521. The examined post-update path has no normal return bypassing that logging. +- `instrumented_python_packages/verl/utils/logger/aggregate_logger.py:49`: the console logger prints with `flush=True`. A CPU-only check of this exact formatter emitted `step:1 - training/global_step:1`; the frozen terminal regex recognized it. This is not a demonstrated formatting/regex mismatch. +- `instrumented_python_packages/verl/trainer/main_ppo.py:93` waits for the remote trainer with `ray.get`; `memory_tuner/device_contract.py:254` then returns from `runpy` into an immediate explicit `ray.shutdown()` at line 257. + +The pinned container is `/mnt/proj3/open-35-44/verl/images/verl-vllm018-dev1.sif`, SHA-256 `3dc581ca780cd19d565445b8a02e5daa98b350b3d2116d4d64476a64a2614912`. CPU-only source inspection inside it confirmed Ray **2.54.1**: + +- `/usr/local/lib/python3.12/dist-packages/ray/_private/worker.py:2128` waits 0.5 seconds for log delivery only during interpreter-exit shutdown. Explicit shutdown skips this grace. +- `worker.py:2788` stops the logging threads and closes the log subscriber. Flushing already received deduplicated messages does not drain upstream actor logs. +- `_private/log_monitor.py:471` forwards worker log files asynchronously; its idle loop sleeps 0.1 seconds. + +**Immediate explicit shutdown therefore supplies a concrete final-log-loss mechanism and is the strongest explanation, not proof of the missing line's emission.** Original Ray actor stdout was not retained; the launcher removes its scratch runtime. `trainer-terminal.json` records driver teardown, not a completed training step. The final squashfuse background-cleanup messages are not terminal training evidence. All nine 4-GPU periods remain unresolved under the unchanged gate. + +## All nine affected 4-GPU periods + +Times below are the exact latest `finished_ns` across ranks 0–3, rendered as UTC on 2026-09-15. “Sync” means the synchronization **after** the actor update, excluding initialization synchronization. Peaks are diagnostic raw trial peaks, not eligible completed-run peaks. + +| Index | Workload / seed | Actual job / period | Actor update finished | Post-update sync finished | Raw peak MiB | +| --- | --- | --- | --- | --- | ---: | +| 0 | code_heavy_tail / 161 | 5055846 / 1 | 19:01:31.699704267 | 19:02:18.386028932 | 39842 | +| 1 | gsm8k / 163 | 5055847 / 2 | 19:05:57.381348829 | 19:08:02.327763208 | 38560 | +| 2 | code_heavy_tail / 162 | 5055848 / 2 | 19:04:48.459750884 | 19:05:59.536233566 | 39392 | +| 3 | gsm8k / 162 | 5055858 / 1 | 19:09:58.895325061 | 19:10:32.865654877 | 38560 | +| 4 | math / 162 | 5055859 / 2 | 19:14:44.017014711 | 19:15:55.046759828 | 39086 | +| 5 | code_heavy_tail / 163 | 5055860 / 2 | 19:16:12.444318056 | 19:17:22.263802288 | 39648 | +| 6 | math / 163 | 5055863 / 1 | 19:19:05.722544967 | 19:20:47.573877478 | 39086 | +| 7 | math / 161 | 5055864 / 1 | 19:21:44.142735712 | 19:22:20.560774115 | 39086 | +| 8 | gsm8k / 161 | 5055845 / 2 | 19:25:33.619484057 | 19:27:04.665561977 | 38560 | + +Exact evidence locator: + +```text +BASE=/mnt/proj3/open-35-44/verl/rl/output/estimation_host_capacity +PAIR=host-qwen25-7b--c2-s +ATTEMPT=BASE/PAIR-4gpu/attempt- + +ATTEMPT/allocator-traces/allocator-.jsonl +ATTEMPT/device-evidence/worker-actor_rollout--.json +ATTEMPT/training-.log +ATTEMPT/trial-.json +BASE/pairs/PAIR/period-.json +``` + +Worker PIDs in rank order 0, 1, 2, 3: + +```text +5055846: 758712,758713,758714,758715 +5055847: 3563810,3563811,3563812,3563813 +5055848: 3495668,3495669,3495670,3495671 +5055858: 802123,802124,802125,802126 +5055859: 3539673,3539674,3539675,3539676 +5055860: 3613579,3613580,3613582,3613583 +5055863: 843954,843955,843956,843957 +5055864: 3571156,3571157,3571158,3571159 +5055845: 3657525,3657526,3657527,3657528 +``` + +Each allocator file was checked against its period-record hash and size; its PID/rank was matched to worker identity evidence. Each contains exactly one actor-update record and one subsequent synchronization record, with ordered timestamps and empty errors. + +## Accounting and copy paths + +Saved terminal scheduler response: + +```text +/tmp/rlvrambench-host-final-20260915.u6ynp1/sacct-raw.psv +SHA256: 5c672842d7ad1ef71e6164ae76ba2e74400c45c2511a66d4139489f34664d2eb +``` + +Portable, pair-claim-bound accounting input used by the collector: + +```text +/tmp/rlvrambench-host-final-20260915.u6ynp1/allocation_accounting.json +SHA256: 8df8040c1cd1608abd10759dd42448d734aa1b501733c926978247d2177f12ee +``` + +The raw response was captured with `TZ=UTC`, `SLURM_TIME_FORMAT=standard`, and: + +```text +sacct -n -P -j 5055845 --format=JobID,JobIDRaw,JobName,State,ExitCode,Start,End,ElapsedRaw,AllocTRES,NodeList,Reason +``` + +`memory_tuner.capture_allocation_accounting` consumed that saved response with `--root /mnt/proj3/open-35-44/verl/rl --group estimation_host_capacity`. All nine accounting records validate against the final manifests/ledgers. Independently summing outer elapsed seconds × four GPUs gives **20,828 GPU-seconds = 5.785555556 GPU-hours**. Task-provisioned invocation time is 17,605.783111 GPU-seconds; four-GPU invocation-reservation time is 20,199.935106 GPU-seconds. These overlapping scopes must not be added. Preflight cost is outside this study-array collection. + +## Frozen provenance and final outputs + +The mutable collector files matched Volta's integration report. Frozen source/protocol inputs were read from the immutable source root above; raw evidence was read from the canonical project. External execution authority was: + +```text +/home/kobzaond/RLVRAMBench-revision-20260915/benchmark/host_capacity/execution_commits.json +``` + +Its exact nine-pair mapping requires commit `120e29a146294722a6668f55a1e7fed2878eb62a` throughout. Anchors verified: + +```text +Host protocol: f3f49617a7fca63ddac8923bccb83a15eaf52646e5f3db5df4aa716bb25c1ef6 +Host freeze: 80559199c59246d0e4ce52bf268d0375c90fc5168a5246ed900c9266abd951cd +Original prediction seal: +5ed714caff922a5109346a4242545d19f2824636d0f52bb0bbd2a09b40ed4666 +``` + +Final output directory, exactly 13 files: + +```text +/tmp/rlvrambench-host-final-20260915.u6ynp1/host-results-final +output_manifest.json SHA256: +47fe58631433062547800641c4e4a4ed5136b661db1c297d6ef24d15c1012709 +``` + +All 12 manifest-listed output hashes/sizes and 396 raw-evidence, 20 source-input, and five protocol-input hash/size records were independently rechecked. The output includes explicit unresolved configurations and unscored approvals; no all-panel accuracy or eligible 4-GPU peak is asserted. diff --git a/benchmark/host_capacity/execution_history.json b/benchmark/host_capacity/execution_history.json new file mode 100644 index 0000000000000000000000000000000000000000..ce13b3ae806ae1730a086bc1a1bb94a7a0fee76b --- /dev/null +++ b/benchmark/host_capacity/execution_history.json @@ -0,0 +1,63 @@ +{ + "schema_version": 1, + "source_commit": "120e29a146294722a6668f55a1e7fed2878eb62a", + "protocol_sha256": "f3f49617a7fca63ddac8923bccb83a15eaf52646e5f3db5df4aa716bb25c1ef6", + "matrix_sha256": "1d2b2af601361a3cb46b65ff3be263685b3f725f2b0e4787a8ff9966084f5332", + "study_freeze_sha256": "80559199c59246d0e4ce52bf268d0375c90fc5168a5246ed900c9266abd951cd", + "original_prediction_seal_sha256": "5ed714caff922a5109346a4242545d19f2824636d0f52bb0bbd2a09b40ed4666", + "preflight": { + "job_id": "5055844", + "status": "completed", + "started_at_utc": "2026-09-15T18:54:58+00:00", + "ended_at_utc": "2026-09-15T18:56:14+00:00", + "elapsed_seconds": 76, + "allocated_gpus": 4, + "allocated_cpus": 64, + "host_memory_mib": 393216, + "training_launched": false, + "task_gpu_counts_checked": [2, 4], + "pass_record": "output/estimation-device-preflight/5055844/preflight-passed.json", + "cost_reported_separately": true + }, + "study_array": { + "array_job_id": "5055845", + "dependency": "afterok:5055844", + "requested_pair_indices": [0, 1, 2, 3, 4, 5, 6, 7, 8], + "first_started_at_utc": "2026-09-15T18:56:16+00:00", + "maximum_concurrency": 3, + "status": "complete_all_nine_pairs_terminal", + "last_ended_at_utc": "2026-09-15T19:27:18+00:00", + "outer_job_states": "all FAILED with exit 2:0 from incomplete pair outcomes", + "retries": 0, + "replacement_of_original_slots": false, + "source_worktree_clean_before_submission": true + }, + "prelaunch_validation": { + "protocol_and_reviewer_tests_passed": 34, + "combined_protocol_runtime_tests_passed": 183, + "combined_optional_checks_skipped": 1, + "actual_verify_protocol_path_passed": true, + "original_effective_inputs_and_prepared_data_hashes_verified": true + }, + "runtime_change": "Exact scheduler array-element lookup and one-record identity matching replace ambiguous parent-array lookup. The original final allocation's response was not retained, so its detailed mechanism remains a diagnostic inference. Its two nonlaunches were not rerun.", + "maximum_reserved_study_gpu_hours": 33, + "further_host_memory_escalation_permitted": false, + "final_outcomes": { + "planned_slots": 18, + "launched_invocations": 18, + "eligible_two_gpu_memory_admission_failures": 9, + "unresolved_four_gpu_invocations": 9, + "validated_completions": 0, + "resolved_configurations": 3, + "unresolved_configurations": 3, + "scheduler_host_oom_recorded": false, + "four_gpu_missingness": "Successful exit and returning update/synchronization calls, but no exact one-step terminal record. Completion eligibility remains unchanged; partial peaks are not completed peaks.", + "accounted_outer_allocations": 9, + "outer_allocation_gpu_seconds": 20828, + "batch_step_gpu_seconds_separate_not_additive": 20828, + "preflight_gpu_seconds_separate": 304, + "results_directory": "benchmark/host_capacity/results", + "accounting_input": "benchmark/host_capacity/allocation_accounting.json", + "accounting_raw_source": "benchmark/host_capacity/allocation_accounting_sacct.psv" + } +} diff --git a/benchmark/host_capacity/matrix.csv b/benchmark/host_capacity/matrix.csv new file mode 100644 index 0000000000000000000000000000000000000000..30d9b12112e4a7d04c9e9cdeff6f5da21829de17 --- /dev/null +++ b/benchmark/host_capacity/matrix.csv @@ -0,0 +1,19 @@ +experiment_id,run_group,study_stage,configuration_level,model_family,algorithm,dataset,model,data_dir,gpu_count,rollout_tp_size,actor_micro_batch,rollout_logprob_micro_batch,ref_logprob_micro_batch,vllm_gpu_memory_utilization,parameter_offload,optimizer_offload,free_cache_engine,max_prompt_length,max_response_length,max_model_len,max_num_seqs,rollout_n,train_batch_size,train_max_samples,val_max_samples,training_seed,total_training_steps,gpu_monitor_interval_ms,workload_name,custom_reward_function_path,custom_reward_function_name,revision_study,pair_id,period,condition,configuration_id,case_id,gpu_subset_ordinals,allocator_trace_enabled,allocator_trace_sync,save_freq,test_freq,val_before_train,resume_mode,max_actor_ckpt_to_keep,per_process_timeout_seconds +host-qwen25-7b-code_heavy_tail-c2-s161-4gpu,estimation_host_capacity,estimation_host_capacity,c2,qwen25_7b,grpo,code_heavy_tail,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/codecontests/heavy_tail,4,1,8,4,4,0.6,False,False,True,1536,1024,2560,64,2,32,32,32,161,1,100,code_heavy_tail,/mnt/proj3/open-35-44/verl/rl/memory_tuner/code_shape_reward.py,compute_score,estimation_host_capacity,host-qwen25-7b-code_heavy_tail-c2-s161,1,4gpu,host-qwen25-7b-code_heavy_tail-c2-4gpu,host-qwen25-7b-code_heavy_tail-4gpu,0;1;2;3,True,1,-1,-1,False,disable,1,1500 +host-qwen25-7b-code_heavy_tail-c2-s161-2gpu,estimation_host_capacity,estimation_host_capacity,c2,qwen25_7b,grpo,code_heavy_tail,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/codecontests/heavy_tail,2,1,8,4,4,0.6,False,False,True,1536,1024,2560,64,2,32,32,32,161,1,100,code_heavy_tail,/mnt/proj3/open-35-44/verl/rl/memory_tuner/code_shape_reward.py,compute_score,estimation_host_capacity,host-qwen25-7b-code_heavy_tail-c2-s161,2,2gpu,host-qwen25-7b-code_heavy_tail-c2-2gpu,host-qwen25-7b-code_heavy_tail-2gpu,0;2,True,1,-1,-1,False,disable,1,1500 +host-qwen25-7b-gsm8k-c2-s163-2gpu,estimation_host_capacity,estimation_host_capacity,c2,qwen25_7b,grpo,gsm8k,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/gsm8k,2,1,8,4,4,0.6,False,False,True,1024,2048,3072,64,2,32,32,32,163,1,100,,,,estimation_host_capacity,host-qwen25-7b-gsm8k-c2-s163,1,2gpu,host-qwen25-7b-gsm8k-c2-2gpu,host-qwen25-7b-gsm8k-2gpu,1;3,True,1,-1,-1,False,disable,1,1500 +host-qwen25-7b-gsm8k-c2-s163-4gpu,estimation_host_capacity,estimation_host_capacity,c2,qwen25_7b,grpo,gsm8k,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/gsm8k,4,1,8,4,4,0.6,False,False,True,1024,2048,3072,64,2,32,32,32,163,1,100,,,,estimation_host_capacity,host-qwen25-7b-gsm8k-c2-s163,2,4gpu,host-qwen25-7b-gsm8k-c2-4gpu,host-qwen25-7b-gsm8k-4gpu,0;1;2;3,True,1,-1,-1,False,disable,1,1500 +host-qwen25-7b-code_heavy_tail-c2-s162-2gpu,estimation_host_capacity,estimation_host_capacity,c2,qwen25_7b,grpo,code_heavy_tail,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/codecontests/heavy_tail,2,1,8,4,4,0.6,False,False,True,1536,1024,2560,64,2,32,32,32,162,1,100,code_heavy_tail,/mnt/proj3/open-35-44/verl/rl/memory_tuner/code_shape_reward.py,compute_score,estimation_host_capacity,host-qwen25-7b-code_heavy_tail-c2-s162,1,2gpu,host-qwen25-7b-code_heavy_tail-c2-2gpu,host-qwen25-7b-code_heavy_tail-2gpu,2;3,True,1,-1,-1,False,disable,1,1500 +host-qwen25-7b-code_heavy_tail-c2-s162-4gpu,estimation_host_capacity,estimation_host_capacity,c2,qwen25_7b,grpo,code_heavy_tail,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/codecontests/heavy_tail,4,1,8,4,4,0.6,False,False,True,1536,1024,2560,64,2,32,32,32,162,1,100,code_heavy_tail,/mnt/proj3/open-35-44/verl/rl/memory_tuner/code_shape_reward.py,compute_score,estimation_host_capacity,host-qwen25-7b-code_heavy_tail-c2-s162,2,4gpu,host-qwen25-7b-code_heavy_tail-c2-4gpu,host-qwen25-7b-code_heavy_tail-4gpu,0;1;2;3,True,1,-1,-1,False,disable,1,1500 +host-qwen25-7b-gsm8k-c2-s162-4gpu,estimation_host_capacity,estimation_host_capacity,c2,qwen25_7b,grpo,gsm8k,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/gsm8k,4,1,8,4,4,0.6,False,False,True,1024,2048,3072,64,2,32,32,32,162,1,100,,,,estimation_host_capacity,host-qwen25-7b-gsm8k-c2-s162,1,4gpu,host-qwen25-7b-gsm8k-c2-4gpu,host-qwen25-7b-gsm8k-4gpu,0;1;2;3,True,1,-1,-1,False,disable,1,1500 +host-qwen25-7b-gsm8k-c2-s162-2gpu,estimation_host_capacity,estimation_host_capacity,c2,qwen25_7b,grpo,gsm8k,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/gsm8k,2,1,8,4,4,0.6,False,False,True,1024,2048,3072,64,2,32,32,32,162,1,100,,,,estimation_host_capacity,host-qwen25-7b-gsm8k-c2-s162,2,2gpu,host-qwen25-7b-gsm8k-c2-2gpu,host-qwen25-7b-gsm8k-2gpu,0;2,True,1,-1,-1,False,disable,1,1500 +host-qwen25-7b-math-c2-s162-2gpu,estimation_host_capacity,estimation_host_capacity,c2,qwen25_7b,grpo,math,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/math,2,1,8,8,8,0.6,False,False,True,512,512,1024,128,2,64,64,64,162,1,100,,,,estimation_host_capacity,host-qwen25-7b-math-c2-s162,1,2gpu,host-qwen25-7b-math-c2-2gpu,host-qwen25-7b-math-2gpu,0;3,True,1,-1,-1,False,disable,1,1500 +host-qwen25-7b-math-c2-s162-4gpu,estimation_host_capacity,estimation_host_capacity,c2,qwen25_7b,grpo,math,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/math,4,1,8,8,8,0.6,False,False,True,512,512,1024,128,2,64,64,64,162,1,100,,,,estimation_host_capacity,host-qwen25-7b-math-c2-s162,2,4gpu,host-qwen25-7b-math-c2-4gpu,host-qwen25-7b-math-4gpu,0;1;2;3,True,1,-1,-1,False,disable,1,1500 +host-qwen25-7b-code_heavy_tail-c2-s163-2gpu,estimation_host_capacity,estimation_host_capacity,c2,qwen25_7b,grpo,code_heavy_tail,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/codecontests/heavy_tail,2,1,8,4,4,0.6,False,False,True,1536,1024,2560,64,2,32,32,32,163,1,100,code_heavy_tail,/mnt/proj3/open-35-44/verl/rl/memory_tuner/code_shape_reward.py,compute_score,estimation_host_capacity,host-qwen25-7b-code_heavy_tail-c2-s163,1,2gpu,host-qwen25-7b-code_heavy_tail-c2-2gpu,host-qwen25-7b-code_heavy_tail-2gpu,1;2,True,1,-1,-1,False,disable,1,1500 +host-qwen25-7b-code_heavy_tail-c2-s163-4gpu,estimation_host_capacity,estimation_host_capacity,c2,qwen25_7b,grpo,code_heavy_tail,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/codecontests/heavy_tail,4,1,8,4,4,0.6,False,False,True,1536,1024,2560,64,2,32,32,32,163,1,100,code_heavy_tail,/mnt/proj3/open-35-44/verl/rl/memory_tuner/code_shape_reward.py,compute_score,estimation_host_capacity,host-qwen25-7b-code_heavy_tail-c2-s163,2,4gpu,host-qwen25-7b-code_heavy_tail-c2-4gpu,host-qwen25-7b-code_heavy_tail-4gpu,0;1;2;3,True,1,-1,-1,False,disable,1,1500 +host-qwen25-7b-math-c2-s163-4gpu,estimation_host_capacity,estimation_host_capacity,c2,qwen25_7b,grpo,math,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/math,4,1,8,8,8,0.6,False,False,True,512,512,1024,128,2,64,64,64,163,1,100,,,,estimation_host_capacity,host-qwen25-7b-math-c2-s163,1,4gpu,host-qwen25-7b-math-c2-4gpu,host-qwen25-7b-math-4gpu,0;1;2;3,True,1,-1,-1,False,disable,1,1500 +host-qwen25-7b-math-c2-s163-2gpu,estimation_host_capacity,estimation_host_capacity,c2,qwen25_7b,grpo,math,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/math,2,1,8,8,8,0.6,False,False,True,512,512,1024,128,2,64,64,64,163,1,100,,,,estimation_host_capacity,host-qwen25-7b-math-c2-s163,2,2gpu,host-qwen25-7b-math-c2-2gpu,host-qwen25-7b-math-2gpu,0;1,True,1,-1,-1,False,disable,1,1500 +host-qwen25-7b-math-c2-s161-4gpu,estimation_host_capacity,estimation_host_capacity,c2,qwen25_7b,grpo,math,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/math,4,1,8,8,8,0.6,False,False,True,512,512,1024,128,2,64,64,64,161,1,100,,,,estimation_host_capacity,host-qwen25-7b-math-c2-s161,1,4gpu,host-qwen25-7b-math-c2-4gpu,host-qwen25-7b-math-4gpu,0;1;2;3,True,1,-1,-1,False,disable,1,1500 +host-qwen25-7b-math-c2-s161-2gpu,estimation_host_capacity,estimation_host_capacity,c2,qwen25_7b,grpo,math,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/math,2,1,8,8,8,0.6,False,False,True,512,512,1024,128,2,64,64,64,161,1,100,,,,estimation_host_capacity,host-qwen25-7b-math-c2-s161,2,2gpu,host-qwen25-7b-math-c2-2gpu,host-qwen25-7b-math-2gpu,2;3,True,1,-1,-1,False,disable,1,1500 +host-qwen25-7b-gsm8k-c2-s161-2gpu,estimation_host_capacity,estimation_host_capacity,c2,qwen25_7b,grpo,gsm8k,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/gsm8k,2,1,8,4,4,0.6,False,False,True,1024,2048,3072,64,2,32,32,32,161,1,100,,,,estimation_host_capacity,host-qwen25-7b-gsm8k-c2-s161,1,2gpu,host-qwen25-7b-gsm8k-c2-2gpu,host-qwen25-7b-gsm8k-2gpu,0;1,True,1,-1,-1,False,disable,1,1500 +host-qwen25-7b-gsm8k-c2-s161-4gpu,estimation_host_capacity,estimation_host_capacity,c2,qwen25_7b,grpo,gsm8k,Qwen/Qwen2.5-7B-Instruct,/mnt/proj3/open-35-44/verl/rl/data/gsm8k,4,1,8,4,4,0.6,False,False,True,1024,2048,3072,64,2,32,32,32,161,1,100,,,,estimation_host_capacity,host-qwen25-7b-gsm8k-c2-s161,2,4gpu,host-qwen25-7b-gsm8k-c2-4gpu,host-qwen25-7b-gsm8k-4gpu,0;1;2;3,True,1,-1,-1,False,disable,1,1500 diff --git a/benchmark/host_capacity/predictions.json b/benchmark/host_capacity/predictions.json new file mode 100644 index 0000000000000000000000000000000000000000..f80375829b5940294c91bc37123ad1df649a9291 --- /dev/null +++ b/benchmark/host_capacity/predictions.json @@ -0,0 +1,290 @@ +[ + { + "configuration_id": "host-qwen25-7b-gsm8k-c2-2gpu", + "donor_configuration_id": "cfg-338d9224eb9d2f3a", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 33108.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39409.6357421875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "host-qwen25-7b-gsm8k-c2-2gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "memory_failure", + "predicted_peak_mib": 46461.218916058904, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39409.6357421875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "host-qwen25-7b-gsm8k-c2-2gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39409.6357421875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "host-qwen25-7b-gsm8k-c2-4gpu", + "donor_configuration_id": "cfg-399f6b038d6f65f9", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 29840.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31992.81787109375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "host-qwen25-7b-gsm8k-c2-4gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 38751.70456713468, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31992.81787109375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "host-qwen25-7b-gsm8k-c2-4gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "above_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31992.81787109375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "host-qwen25-7b-math-c2-2gpu", + "donor_configuration_id": "cfg-30f98456fdb322c1", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 33446.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39409.6357421875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "host-qwen25-7b-math-c2-2gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "memory_failure", + "predicted_peak_mib": 46391.75739749115, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39409.6357421875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "host-qwen25-7b-math-c2-2gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39409.6357421875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "host-qwen25-7b-math-c2-4gpu", + "donor_configuration_id": "cfg-f03e4e98c682240e", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 30180.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31992.81787109375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "host-qwen25-7b-math-c2-4gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 38682.24304856692, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31992.81787109375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "host-qwen25-7b-math-c2-4gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "above_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31992.81787109375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "host-qwen25-7b-code_heavy_tail-c2-2gpu", + "donor_configuration_id": "cfg-e91942abb5b4f847", + "donor_gpu_count_matches": true, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 33436.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39409.6357421875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "host-qwen25-7b-code_heavy_tail-c2-2gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "memory_failure", + "predicted_peak_mib": 46443.853536416966, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39409.6357421875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "host-qwen25-7b-code_heavy_tail-c2-2gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "memory_failure", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 39409.6357421875, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "host-qwen25-7b-code_heavy_tail-c2-4gpu", + "donor_configuration_id": "cfg-e91942abb5b4f847", + "donor_gpu_count_matches": false, + "method": "donor_copy", + "peak_only_predicted_state": null, + "predicted_peak_mib": 33436.0, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31992.81787109375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + }, + { + "configuration_id": "host-qwen25-7b-code_heavy_tail-c2-4gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "component_regression", + "peak_only_predicted_state": "within_margin", + "predicted_peak_mib": 38734.33918749274, + "predicted_state": "within_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31992.81787109375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": false + }, + { + "configuration_id": "host-qwen25-7b-code_heavy_tail-c2-4gpu", + "donor_configuration_id": null, + "donor_gpu_count_matches": null, + "method": "logistic", + "peak_only_predicted_state": null, + "predicted_peak_mib": null, + "predicted_state": "above_margin", + "startup_budget_diagnostic": { + "applicable": true, + "assumption": "FP32 actor shards resident before vLLM requested-budget free-memory check", + "budget_plus_actor_mib": 31992.81787109375, + "predicted_startup_budget_failure": false + }, + "startup_guard_changed_state": null + } +] diff --git a/benchmark/host_capacity/protocol.json b/benchmark/host_capacity/protocol.json new file mode 100644 index 0000000000000000000000000000000000000000..e5ec1358b1d951b5884f1b97a8589c762edf3166 --- /dev/null +++ b/benchmark/host_capacity/protocol.json @@ -0,0 +1,69 @@ +{ + "acquisition_seed": 2026091513, + "allocation_gpus": 4, + "comparison": "Two versus four task GPUs within the same four-GPU/64-CPU/384-GiB allocation. Matched settings and prompt seed, not necessarily identical generated responses. No causal effect of host capacity is estimated against the original panel.", + "cost_scope": "Both periods reserve four GPUs. Report task-provisioned GPU-time (2*t2+4*t4), reservation during invocations 4*(t2+t4), and whole allocation4*T separately. Include failed/unresolved work; queue wait is separate. None is utilization-weighted time.", + "cpus_per_task": 64, + "data_sha256": { + "data/codecontests/heavy_tail/test.parquet": "e9d077a023de7f80aa013c5541c33b5978a43e203ce916045bc72c4fe5c62a82", + "data/codecontests/heavy_tail/train.parquet": "0bc6ec8fd7ef1618a604417faa4bb13cb9a0117dcffb75685b2c80c1da95f697", + "data/gsm8k/test.parquet": "0dcd50ed32caa9d8f434d4e3ef8031bc3e49b0f823655a53be5c002e727f17dd", + "data/gsm8k/train.parquet": "8fb13a0cd8621e5fcacdee3ffe28b3a369975478ea7ab514781f3f190cba2488", + "data/math/test.parquet": "e0cb48b1e49874c71f4a02dff944eba08844067d98dc5bdc89c004644e6b597d", + "data/math/train.parquet": "4aa303577e968be6f810600730c970b14be659a9ae9072cc2afefd04d67e7cb8" + }, + "device_capacity_mib": 40960, + "estimators": "Exact inherited predictions, donor choices and numerical peaks. No fitting, threshold change, calibration, or update from either panel's outcomes.", + "evaluation_seeds": [ + 161, + 162, + 163 + ], + "failure_policy": "No retries, replacements, requeues or further host-memory escalation. Each pair is claimed once. Continue period two only after verified cleanup and all-device idle checks. Preserve nonlaunches and incomplete pairs.", + "group": "estimation_host_capacity", + "host_memory_mib": 393216, + "implementation_sha256": { + "memory_tuner/host_capacity_protocol.py": "9bf811457f04e02e67e31d3c9289d1155f73ee1364a233b6b77d08d86becf44b", + "memory_tuner/run_host_capacity.py": "23f1eaf462f1825a77e91d2c8d8a4ab073f4debe0fbab134ff1863e096345d37", + "memory_tuner/run_matched_gpu.py": "f340fb7eec5297cb675a48e267618c15d7787a42891d4a2ae4a670ec24a0cc47", + "run_host_capacity.slurm": "59d3daa9499d1fc76d0b856be6c0a813472f488eb49f990e6ffa57ccf819b919" + }, + "memory_margin_limit_mib": 38912, + "model": "Qwen/Qwen2.5-7B-Instruct", + "model_revision": "a09a35458c702b33eeacc393d103063234e8bc28", + "numerical_reporting": "Peak error uses only configurations with three completed invocations, comparing the inherited prediction with their maximum external completed peak. Donor-copy numerical values are inherited source measurements, not refitted target estimates. Host-kill and GPU-failure peaks are not imputed or scored as completed peaks. Paired peak and runtime differences require both invocations to complete.", + "order_seed": 2026091511, + "original_prediction_seal_sha256": "5ed714caff922a5109346a4242545d19f2824636d0f52bb0bbd2a09b40ed4666", + "original_protocol_sha256": "1694e447e842600612604891e7b3d036ff0791fa84c4bdbeba54ed093feefdeb", + "outcome_rule": "All three eligible seeds required for a repeated label. Any validated GPU memory failure implies memory_failure; otherwise any completed peak above 38912 MiB implies above_margin; otherwise within_margin. Other slots are unresolved with partial eligible evidence retained.", + "planned_pairs": 9, + "planned_processes": 18, + "prediction_origins": { + "host-qwen25-7b-code_heavy_tail-c2-2gpu": "est-qwen25-7b-code_heavy_tail-c2-2gpu", + "host-qwen25-7b-code_heavy_tail-c2-4gpu": "est-qwen25-7b-code_heavy_tail-c2-4gpu", + "host-qwen25-7b-gsm8k-c2-2gpu": "est-qwen25-7b-gsm8k-c2-2gpu", + "host-qwen25-7b-gsm8k-c2-4gpu": "est-qwen25-7b-gsm8k-c2-4gpu", + "host-qwen25-7b-math-c2-2gpu": "est-qwen25-7b-math-c2-2gpu", + "host-qwen25-7b-math-c2-4gpu": "est-qwen25-7b-math-c2-4gpu" + }, + "primary_reporting": "Separate six-configuration inventory, all eighteen seed slots, each method's useful/failing/above-margin approvals and unscored approvals. No pooled denominator with the original panel and no population interval.", + "protocol_version": "host-capacity-1.0", + "randomization": "Five two-GPU-first and four four-GPU-first pairs; each workload has both orders. All six two-of-four subsets occur, with three appearing twice; each allocation ordinal appears four or five times.", + "rationale": "Original four-GPU c2 runs were censored by the 192 GiB host limit. This one follow-up tests all three original workloads under a separately declared expanded host-resource condition; it does not replace original slots.", + "resource_cap": { + "gpus_per_allocation": 4, + "idle_timeout_seconds": 180, + "initial_allocations": 9, + "max_concurrent_allocations": 3, + "maximum_reserved_gpu_hours": 33, + "payload_timeout_seconds": 1500, + "requeue": false, + "retries": 0, + "wall_minutes_per_allocation": 55 + }, + "scope": "Known Qwen family and A100 stack. New seeds and host-resource condition; not blind method development, external validation, or cross-hardware transfer.", + "specified_at_utc": "2026-09-15T18:52:11.216739+00:00", + "stopping": "Exactly nine planned allocation claims, each capped at 55 minutes. A systemic safety or infrastructure problem may stop remaining work but cannot authorize replacement slots. Stopping must not depend on prediction correctness or favorability. Diagnosed GPU-memory failures are expected study outcomes, not by themselves infrastructure-abort grounds. Any safety or infrastructure stop is documented, and all unstarted slots remain in the inventory.", + "subset_seed": 2026091512, + "target_configurations": 6 +} diff --git a/benchmark/host_capacity/results/configurations.csv b/benchmark/host_capacity/results/configurations.csv new file mode 100644 index 0000000000000000000000000000000000000000..6d2b49126b2e9e364121686b9a852a0c6f436c2a --- /dev/null +++ b/benchmark/host_capacity/results/configurations.csv @@ -0,0 +1,7 @@ +completed_peak_mib,completed_processes,configuration_id,eligible_processes,eligible_seeds,known_margin_exceedance,known_memory_failure,maximum_known_completed_peak_mib,planned_seeds,settings,state,unresolved_seeds +,0,host-qwen25-7b-gsm8k-c2-2gpu,3,"[161, 162, 163]",False,True,,"[161, 162, 163]","{""accelerator"": ""NVIDIA A100-SXM4-40GB"", ""actor_micro_batch"": ""8"", ""adaptation"": ""LoRA"", ""algorithm"": ""GRPO"", ""allocator_trace_enabled"": ""True"", ""allocator_trace_sync"": ""1"", ""case_id"": ""host-qwen25-7b-gsm8k-2gpu"", ""condition"": """", ""configuration_id"": ""host-qwen25-7b-gsm8k-c2-2gpu"", ""configuration_level"": ""c2"", ""device_capacity_mib"": ""40960"", ""free_cache_engine"": ""True"", ""gpu_count"": ""2"", ""gpu_monitor_interval_ms"": ""100"", ""legacy_dataset_id"": ""gsm8k"", ""margin_limit_mib"": ""38912"", ""max_actor_ckpt_to_keep"": ""1"", ""max_model_len"": ""3072"", ""max_num_seqs"": ""64"", ""max_prompt_length"": ""1024"", ""max_response_length"": ""2048"", ""model_family"": ""qwen25_7b"", ""model_id"": ""Qwen/Qwen2.5-7B-Instruct"", ""model_name"": ""Qwen2.5-7B-Instruct"", ""optimizer_offload"": ""False"", ""parameter_offload"": ""False"", ""peak_measurement"": ""maximum externally sampled per-device memory"", ""ref_logprob_micro_batch"": ""4"", ""resume_mode"": ""disable"", ""rollout_logprob_micro_batch"": ""4"", ""rollout_n"": ""2"", ""rollout_tp_size"": ""1"", ""save_freq"": ""-1"", ""study"": ""estimation_host_capacity"", ""test_freq"": ""-1"", ""total_training_steps"": ""1"", ""train_batch_size"": ""32"", ""train_max_samples"": ""32"", ""val_before_train"": ""False"", ""val_max_samples"": ""32"", ""vllm_gpu_memory_utilization"": ""0.6"", ""workload_id"": ""gsm8k"", ""workload_name"": ""GSM8K mathematics""}",memory_failure,[] +,0,host-qwen25-7b-gsm8k-c2-4gpu,0,[],False,False,,"[161, 162, 163]","{""accelerator"": ""NVIDIA A100-SXM4-40GB"", ""actor_micro_batch"": ""8"", ""adaptation"": ""LoRA"", ""algorithm"": ""GRPO"", ""allocator_trace_enabled"": ""True"", ""allocator_trace_sync"": ""1"", ""case_id"": ""host-qwen25-7b-gsm8k-4gpu"", ""condition"": """", ""configuration_id"": ""host-qwen25-7b-gsm8k-c2-4gpu"", ""configuration_level"": ""c2"", ""device_capacity_mib"": ""40960"", ""free_cache_engine"": ""True"", ""gpu_count"": ""4"", ""gpu_monitor_interval_ms"": ""100"", ""legacy_dataset_id"": ""gsm8k"", ""margin_limit_mib"": ""38912"", ""max_actor_ckpt_to_keep"": ""1"", ""max_model_len"": ""3072"", ""max_num_seqs"": ""64"", ""max_prompt_length"": ""1024"", ""max_response_length"": ""2048"", ""model_family"": ""qwen25_7b"", ""model_id"": ""Qwen/Qwen2.5-7B-Instruct"", ""model_name"": ""Qwen2.5-7B-Instruct"", ""optimizer_offload"": ""False"", ""parameter_offload"": ""False"", ""peak_measurement"": ""maximum externally sampled per-device memory"", ""ref_logprob_micro_batch"": ""4"", ""resume_mode"": ""disable"", ""rollout_logprob_micro_batch"": ""4"", ""rollout_n"": ""2"", ""rollout_tp_size"": ""1"", ""save_freq"": ""-1"", ""study"": ""estimation_host_capacity"", ""test_freq"": ""-1"", ""total_training_steps"": ""1"", ""train_batch_size"": ""32"", ""train_max_samples"": ""32"", ""val_before_train"": ""False"", ""val_max_samples"": ""32"", ""vllm_gpu_memory_utilization"": ""0.6"", ""workload_id"": ""gsm8k"", ""workload_name"": ""GSM8K mathematics""}",unresolved,"[161, 162, 163]" +,0,host-qwen25-7b-math-c2-2gpu,3,"[161, 162, 163]",False,True,,"[161, 162, 163]","{""accelerator"": ""NVIDIA A100-SXM4-40GB"", ""actor_micro_batch"": ""8"", ""adaptation"": ""LoRA"", ""algorithm"": ""GRPO"", ""allocator_trace_enabled"": ""True"", ""allocator_trace_sync"": ""1"", ""case_id"": ""host-qwen25-7b-math-2gpu"", ""condition"": """", ""configuration_id"": ""host-qwen25-7b-math-c2-2gpu"", ""configuration_level"": ""c2"", ""device_capacity_mib"": ""40960"", ""free_cache_engine"": ""True"", ""gpu_count"": ""2"", ""gpu_monitor_interval_ms"": ""100"", ""legacy_dataset_id"": ""math"", ""margin_limit_mib"": ""38912"", ""max_actor_ckpt_to_keep"": ""1"", ""max_model_len"": ""1024"", ""max_num_seqs"": ""128"", ""max_prompt_length"": ""512"", ""max_response_length"": ""512"", ""model_family"": ""qwen25_7b"", ""model_id"": ""Qwen/Qwen2.5-7B-Instruct"", ""model_name"": ""Qwen2.5-7B-Instruct"", ""optimizer_offload"": ""False"", ""parameter_offload"": ""False"", ""peak_measurement"": ""maximum externally sampled per-device memory"", ""ref_logprob_micro_batch"": ""8"", ""resume_mode"": ""disable"", ""rollout_logprob_micro_batch"": ""8"", ""rollout_n"": ""2"", ""rollout_tp_size"": ""1"", ""save_freq"": ""-1"", ""study"": ""estimation_host_capacity"", ""test_freq"": ""-1"", ""total_training_steps"": ""1"", ""train_batch_size"": ""64"", ""train_max_samples"": ""64"", ""val_before_train"": ""False"", ""val_max_samples"": ""64"", ""vllm_gpu_memory_utilization"": ""0.6"", ""workload_id"": ""math"", ""workload_name"": ""MATH mathematics""}",memory_failure,[] +,0,host-qwen25-7b-math-c2-4gpu,0,[],False,False,,"[161, 162, 163]","{""accelerator"": ""NVIDIA A100-SXM4-40GB"", ""actor_micro_batch"": ""8"", ""adaptation"": ""LoRA"", ""algorithm"": ""GRPO"", ""allocator_trace_enabled"": ""True"", ""allocator_trace_sync"": ""1"", ""case_id"": ""host-qwen25-7b-math-4gpu"", ""condition"": """", ""configuration_id"": ""host-qwen25-7b-math-c2-4gpu"", ""configuration_level"": ""c2"", ""device_capacity_mib"": ""40960"", ""free_cache_engine"": ""True"", ""gpu_count"": ""4"", ""gpu_monitor_interval_ms"": ""100"", ""legacy_dataset_id"": ""math"", ""margin_limit_mib"": ""38912"", ""max_actor_ckpt_to_keep"": ""1"", ""max_model_len"": ""1024"", ""max_num_seqs"": ""128"", ""max_prompt_length"": ""512"", ""max_response_length"": ""512"", ""model_family"": ""qwen25_7b"", ""model_id"": ""Qwen/Qwen2.5-7B-Instruct"", ""model_name"": ""Qwen2.5-7B-Instruct"", ""optimizer_offload"": ""False"", ""parameter_offload"": ""False"", ""peak_measurement"": ""maximum externally sampled per-device memory"", ""ref_logprob_micro_batch"": ""8"", ""resume_mode"": ""disable"", ""rollout_logprob_micro_batch"": ""8"", ""rollout_n"": ""2"", ""rollout_tp_size"": ""1"", ""save_freq"": ""-1"", ""study"": ""estimation_host_capacity"", ""test_freq"": ""-1"", ""total_training_steps"": ""1"", ""train_batch_size"": ""64"", ""train_max_samples"": ""64"", ""val_before_train"": ""False"", ""val_max_samples"": ""64"", ""vllm_gpu_memory_utilization"": ""0.6"", ""workload_id"": ""math"", ""workload_name"": ""MATH mathematics""}",unresolved,"[161, 162, 163]" +,0,host-qwen25-7b-code_heavy_tail-c2-2gpu,3,"[161, 162, 163]",False,True,,"[161, 162, 163]","{""accelerator"": ""NVIDIA A100-SXM4-40GB"", ""actor_micro_batch"": ""8"", ""adaptation"": ""LoRA"", ""algorithm"": ""GRPO"", ""allocator_trace_enabled"": ""True"", ""allocator_trace_sync"": ""1"", ""case_id"": ""host-qwen25-7b-code_heavy_tail-2gpu"", ""condition"": """", ""configuration_id"": ""host-qwen25-7b-code_heavy_tail-c2-2gpu"", ""configuration_level"": ""c2"", ""device_capacity_mib"": ""40960"", ""free_cache_engine"": ""True"", ""gpu_count"": ""2"", ""gpu_monitor_interval_ms"": ""100"", ""legacy_dataset_id"": ""code_heavy_tail"", ""margin_limit_mib"": ""38912"", ""max_actor_ckpt_to_keep"": ""1"", ""max_model_len"": ""2560"", ""max_num_seqs"": ""64"", ""max_prompt_length"": ""1536"", ""max_response_length"": ""1024"", ""model_family"": ""qwen25_7b"", ""model_id"": ""Qwen/Qwen2.5-7B-Instruct"", ""model_name"": ""Qwen2.5-7B-Instruct"", ""optimizer_offload"": ""False"", ""parameter_offload"": ""False"", ""peak_measurement"": ""maximum externally sampled per-device memory"", ""ref_logprob_micro_batch"": ""4"", ""resume_mode"": ""disable"", ""rollout_logprob_micro_batch"": ""4"", ""rollout_n"": ""2"", ""rollout_tp_size"": ""1"", ""save_freq"": ""-1"", ""study"": ""estimation_host_capacity"", ""test_freq"": ""-1"", ""total_training_steps"": ""1"", ""train_batch_size"": ""32"", ""train_max_samples"": ""32"", ""val_before_train"": ""False"", ""val_max_samples"": ""32"", ""vllm_gpu_memory_utilization"": ""0.6"", ""workload_id"": ""longer_prompt_code"", ""workload_name"": ""Longer-prompt code""}",memory_failure,[] +,0,host-qwen25-7b-code_heavy_tail-c2-4gpu,0,[],False,False,,"[161, 162, 163]","{""accelerator"": ""NVIDIA A100-SXM4-40GB"", ""actor_micro_batch"": ""8"", ""adaptation"": ""LoRA"", ""algorithm"": ""GRPO"", ""allocator_trace_enabled"": ""True"", ""allocator_trace_sync"": ""1"", ""case_id"": ""host-qwen25-7b-code_heavy_tail-4gpu"", ""condition"": """", ""configuration_id"": ""host-qwen25-7b-code_heavy_tail-c2-4gpu"", ""configuration_level"": ""c2"", ""device_capacity_mib"": ""40960"", ""free_cache_engine"": ""True"", ""gpu_count"": ""4"", ""gpu_monitor_interval_ms"": ""100"", ""legacy_dataset_id"": ""code_heavy_tail"", ""margin_limit_mib"": ""38912"", ""max_actor_ckpt_to_keep"": ""1"", ""max_model_len"": ""2560"", ""max_num_seqs"": ""64"", ""max_prompt_length"": ""1536"", ""max_response_length"": ""1024"", ""model_family"": ""qwen25_7b"", ""model_id"": ""Qwen/Qwen2.5-7B-Instruct"", ""model_name"": ""Qwen2.5-7B-Instruct"", ""optimizer_offload"": ""False"", ""parameter_offload"": ""False"", ""peak_measurement"": ""maximum externally sampled per-device memory"", ""ref_logprob_micro_batch"": ""4"", ""resume_mode"": ""disable"", ""rollout_logprob_micro_batch"": ""4"", ""rollout_n"": ""2"", ""rollout_tp_size"": ""1"", ""save_freq"": ""-1"", ""study"": ""estimation_host_capacity"", ""test_freq"": ""-1"", ""total_training_steps"": ""1"", ""train_batch_size"": ""32"", ""train_max_samples"": ""32"", ""val_before_train"": ""False"", ""val_max_samples"": ""32"", ""vllm_gpu_memory_utilization"": ""0.6"", ""workload_id"": ""longer_prompt_code"", ""workload_name"": ""Longer-prompt code""}",unresolved,"[161, 162, 163]" diff --git a/benchmark/host_capacity/results/configurations.json b/benchmark/host_capacity/results/configurations.json new file mode 100644 index 0000000000000000000000000000000000000000..b49a04bcbe9163a585a25c15feb6995dfcf37edc --- /dev/null +++ b/benchmark/host_capacity/results/configurations.json @@ -0,0 +1,398 @@ +[ + { + "completed_peak_mib": null, + "completed_processes": 0, + "configuration_id": "host-qwen25-7b-gsm8k-c2-2gpu", + "eligible_processes": 3, + "eligible_seeds": [ + 161, + 162, + 163 + ], + "known_margin_exceedance": false, + "known_memory_failure": true, + "maximum_known_completed_peak_mib": null, + "planned_seeds": [ + 161, + 162, + 163 + ], + "settings": { + "accelerator": "NVIDIA A100-SXM4-40GB", + "actor_micro_batch": "8", + "adaptation": "LoRA", + "algorithm": "GRPO", + "allocator_trace_enabled": "True", + "allocator_trace_sync": "1", + "case_id": "host-qwen25-7b-gsm8k-2gpu", + "condition": "", + "configuration_id": "host-qwen25-7b-gsm8k-c2-2gpu", + "configuration_level": "c2", + "device_capacity_mib": "40960", + "free_cache_engine": "True", + "gpu_count": "2", + "gpu_monitor_interval_ms": "100", + "legacy_dataset_id": "gsm8k", + "margin_limit_mib": "38912", + "max_actor_ckpt_to_keep": "1", + "max_model_len": "3072", + "max_num_seqs": "64", + "max_prompt_length": "1024", + "max_response_length": "2048", + "model_family": "qwen25_7b", + "model_id": "Qwen/Qwen2.5-7B-Instruct", + "model_name": "Qwen2.5-7B-Instruct", + "optimizer_offload": "False", + "parameter_offload": "False", + "peak_measurement": "maximum externally sampled per-device memory", + "ref_logprob_micro_batch": "4", + "resume_mode": "disable", + "rollout_logprob_micro_batch": "4", + "rollout_n": "2", + "rollout_tp_size": "1", + "save_freq": "-1", + "study": "estimation_host_capacity", + "test_freq": "-1", + "total_training_steps": "1", + "train_batch_size": "32", + "train_max_samples": "32", + "val_before_train": "False", + "val_max_samples": "32", + "vllm_gpu_memory_utilization": "0.6", + "workload_id": "gsm8k", + "workload_name": "GSM8K mathematics" + }, + "state": "memory_failure", + "unresolved_seeds": [] + }, + { + "completed_peak_mib": null, + "completed_processes": 0, + "configuration_id": "host-qwen25-7b-gsm8k-c2-4gpu", + "eligible_processes": 0, + "eligible_seeds": [], + "known_margin_exceedance": false, + "known_memory_failure": false, + "maximum_known_completed_peak_mib": null, + "planned_seeds": [ + 161, + 162, + 163 + ], + "settings": { + "accelerator": "NVIDIA A100-SXM4-40GB", + "actor_micro_batch": "8", + "adaptation": "LoRA", + "algorithm": "GRPO", + "allocator_trace_enabled": "True", + "allocator_trace_sync": "1", + "case_id": "host-qwen25-7b-gsm8k-4gpu", + "condition": "", + "configuration_id": "host-qwen25-7b-gsm8k-c2-4gpu", + "configuration_level": "c2", + "device_capacity_mib": "40960", + "free_cache_engine": "True", + "gpu_count": "4", + "gpu_monitor_interval_ms": "100", + "legacy_dataset_id": "gsm8k", + "margin_limit_mib": "38912", + "max_actor_ckpt_to_keep": "1", + "max_model_len": "3072", + "max_num_seqs": "64", + "max_prompt_length": "1024", + "max_response_length": "2048", + "model_family": "qwen25_7b", + "model_id": "Qwen/Qwen2.5-7B-Instruct", + "model_name": "Qwen2.5-7B-Instruct", + "optimizer_offload": "False", + "parameter_offload": "False", + "peak_measurement": "maximum externally sampled per-device memory", + "ref_logprob_micro_batch": "4", + "resume_mode": "disable", + "rollout_logprob_micro_batch": "4", + "rollout_n": "2", + "rollout_tp_size": "1", + "save_freq": "-1", + "study": "estimation_host_capacity", + "test_freq": "-1", + "total_training_steps": "1", + "train_batch_size": "32", + "train_max_samples": "32", + "val_before_train": "False", + "val_max_samples": "32", + "vllm_gpu_memory_utilization": "0.6", + "workload_id": "gsm8k", + "workload_name": "GSM8K mathematics" + }, + "state": "unresolved", + "unresolved_seeds": [ + 161, + 162, + 163 + ] + }, + { + "completed_peak_mib": null, + "completed_processes": 0, + "configuration_id": "host-qwen25-7b-math-c2-2gpu", + "eligible_processes": 3, + "eligible_seeds": [ + 161, + 162, + 163 + ], + "known_margin_exceedance": false, + "known_memory_failure": true, + "maximum_known_completed_peak_mib": null, + "planned_seeds": [ + 161, + 162, + 163 + ], + "settings": { + "accelerator": "NVIDIA A100-SXM4-40GB", + "actor_micro_batch": "8", + "adaptation": "LoRA", + "algorithm": "GRPO", + "allocator_trace_enabled": "True", + "allocator_trace_sync": "1", + "case_id": "host-qwen25-7b-math-2gpu", + "condition": "", + "configuration_id": "host-qwen25-7b-math-c2-2gpu", + "configuration_level": "c2", + "device_capacity_mib": "40960", + "free_cache_engine": "True", + "gpu_count": "2", + "gpu_monitor_interval_ms": "100", + "legacy_dataset_id": "math", + "margin_limit_mib": "38912", + "max_actor_ckpt_to_keep": "1", + "max_model_len": "1024", + "max_num_seqs": "128", + "max_prompt_length": "512", + "max_response_length": "512", + "model_family": "qwen25_7b", + "model_id": "Qwen/Qwen2.5-7B-Instruct", + "model_name": "Qwen2.5-7B-Instruct", + "optimizer_offload": "False", + "parameter_offload": "False", + "peak_measurement": "maximum externally sampled per-device memory", + "ref_logprob_micro_batch": "8", + "resume_mode": "disable", + "rollout_logprob_micro_batch": "8", + "rollout_n": "2", + "rollout_tp_size": "1", + "save_freq": "-1", + "study": "estimation_host_capacity", + "test_freq": "-1", + "total_training_steps": "1", + "train_batch_size": "64", + "train_max_samples": "64", + "val_before_train": "False", + "val_max_samples": "64", + "vllm_gpu_memory_utilization": "0.6", + "workload_id": "math", + "workload_name": "MATH mathematics" + }, + "state": "memory_failure", + "unresolved_seeds": [] + }, + { + "completed_peak_mib": null, + "completed_processes": 0, + "configuration_id": "host-qwen25-7b-math-c2-4gpu", + "eligible_processes": 0, + "eligible_seeds": [], + "known_margin_exceedance": false, + "known_memory_failure": false, + "maximum_known_completed_peak_mib": null, + "planned_seeds": [ + 161, + 162, + 163 + ], + "settings": { + "accelerator": "NVIDIA A100-SXM4-40GB", + "actor_micro_batch": "8", + "adaptation": "LoRA", + "algorithm": "GRPO", + "allocator_trace_enabled": "True", + "allocator_trace_sync": "1", + "case_id": "host-qwen25-7b-math-4gpu", + "condition": "", + "configuration_id": "host-qwen25-7b-math-c2-4gpu", + "configuration_level": "c2", + "device_capacity_mib": "40960", + "free_cache_engine": "True", + "gpu_count": "4", + "gpu_monitor_interval_ms": "100", + "legacy_dataset_id": "math", + "margin_limit_mib": "38912", + "max_actor_ckpt_to_keep": "1", + "max_model_len": "1024", + "max_num_seqs": "128", + "max_prompt_length": "512", + "max_response_length": "512", + "model_family": "qwen25_7b", + "model_id": "Qwen/Qwen2.5-7B-Instruct", + "model_name": "Qwen2.5-7B-Instruct", + "optimizer_offload": "False", + "parameter_offload": "False", + "peak_measurement": "maximum externally sampled per-device memory", + "ref_logprob_micro_batch": "8", + "resume_mode": "disable", + "rollout_logprob_micro_batch": "8", + "rollout_n": "2", + "rollout_tp_size": "1", + "save_freq": "-1", + "study": "estimation_host_capacity", + "test_freq": "-1", + "total_training_steps": "1", + "train_batch_size": "64", + "train_max_samples": "64", + "val_before_train": "False", + "val_max_samples": "64", + "vllm_gpu_memory_utilization": "0.6", + "workload_id": "math", + "workload_name": "MATH mathematics" + }, + "state": "unresolved", + "unresolved_seeds": [ + 161, + 162, + 163 + ] + }, + { + "completed_peak_mib": null, + "completed_processes": 0, + "configuration_id": "host-qwen25-7b-code_heavy_tail-c2-2gpu", + "eligible_processes": 3, + "eligible_seeds": [ + 161, + 162, + 163 + ], + "known_margin_exceedance": false, + "known_memory_failure": true, + "maximum_known_completed_peak_mib": null, + "planned_seeds": [ + 161, + 162, + 163 + ], + "settings": { + "accelerator": "NVIDIA A100-SXM4-40GB", + "actor_micro_batch": "8", + "adaptation": "LoRA", + "algorithm": "GRPO", + "allocator_trace_enabled": "True", + "allocator_trace_sync": "1", + "case_id": "host-qwen25-7b-code_heavy_tail-2gpu", + "condition": "", + "configuration_id": "host-qwen25-7b-code_heavy_tail-c2-2gpu", + "configuration_level": "c2", + "device_capacity_mib": "40960", + "free_cache_engine": "True", + "gpu_count": "2", + "gpu_monitor_interval_ms": "100", + "legacy_dataset_id": "code_heavy_tail", + "margin_limit_mib": "38912", + "max_actor_ckpt_to_keep": "1", + "max_model_len": "2560", + "max_num_seqs": "64", + "max_prompt_length": "1536", + "max_response_length": "1024", + "model_family": "qwen25_7b", + "model_id": "Qwen/Qwen2.5-7B-Instruct", + "model_name": "Qwen2.5-7B-Instruct", + "optimizer_offload": "False", + "parameter_offload": "False", + "peak_measurement": "maximum externally sampled per-device memory", + "ref_logprob_micro_batch": "4", + "resume_mode": "disable", + "rollout_logprob_micro_batch": "4", + "rollout_n": "2", + "rollout_tp_size": "1", + "save_freq": "-1", + "study": "estimation_host_capacity", + "test_freq": "-1", + "total_training_steps": "1", + "train_batch_size": "32", + "train_max_samples": "32", + "val_before_train": "False", + "val_max_samples": "32", + "vllm_gpu_memory_utilization": "0.6", + "workload_id": "longer_prompt_code", + "workload_name": "Longer-prompt code" + }, + "state": "memory_failure", + "unresolved_seeds": [] + }, + { + "completed_peak_mib": null, + "completed_processes": 0, + "configuration_id": "host-qwen25-7b-code_heavy_tail-c2-4gpu", + "eligible_processes": 0, + "eligible_seeds": [], + "known_margin_exceedance": false, + "known_memory_failure": false, + "maximum_known_completed_peak_mib": null, + "planned_seeds": [ + 161, + 162, + 163 + ], + "settings": { + "accelerator": "NVIDIA A100-SXM4-40GB", + "actor_micro_batch": "8", + "adaptation": "LoRA", + "algorithm": "GRPO", + "allocator_trace_enabled": "True", + "allocator_trace_sync": "1", + "case_id": "host-qwen25-7b-code_heavy_tail-4gpu", + "condition": "", + "configuration_id": "host-qwen25-7b-code_heavy_tail-c2-4gpu", + "configuration_level": "c2", + "device_capacity_mib": "40960", + "free_cache_engine": "True", + "gpu_count": "4", + "gpu_monitor_interval_ms": "100", + "legacy_dataset_id": "code_heavy_tail", + "margin_limit_mib": "38912", + "max_actor_ckpt_to_keep": "1", + "max_model_len": "2560", + "max_num_seqs": "64", + "max_prompt_length": "1536", + "max_response_length": "1024", + "model_family": "qwen25_7b", + "model_id": "Qwen/Qwen2.5-7B-Instruct", + "model_name": "Qwen2.5-7B-Instruct", + "optimizer_offload": "False", + "parameter_offload": "False", + "peak_measurement": "maximum externally sampled per-device memory", + "ref_logprob_micro_batch": "4", + "resume_mode": "disable", + "rollout_logprob_micro_batch": "4", + "rollout_n": "2", + "rollout_tp_size": "1", + "save_freq": "-1", + "study": "estimation_host_capacity", + "test_freq": "-1", + "total_training_steps": "1", + "train_batch_size": "32", + "train_max_samples": "32", + "val_before_train": "False", + "val_max_samples": "32", + "vllm_gpu_memory_utilization": "0.6", + "workload_id": "longer_prompt_code", + "workload_name": "Longer-prompt code" + }, + "state": "unresolved", + "unresolved_seeds": [ + 161, + 162, + 163 + ] + } +] diff --git a/benchmark/host_capacity/results/configurations_flat.csv b/benchmark/host_capacity/results/configurations_flat.csv new file mode 100644 index 0000000000000000000000000000000000000000..99d0847fbd86ec8bf6d635804fe92fa88bc33ad0 --- /dev/null +++ b/benchmark/host_capacity/results/configurations_flat.csv @@ -0,0 +1,7 @@ +all_three_seeds_eligible,completed_peak_mib,completed_processes,component_regression_predicted_peak_mib,component_regression_predicted_state,component_regression_unscored_approval,configuration_id,configuration_level,donor_copy_predicted_peak_mib,donor_copy_predicted_state,donor_copy_unscored_approval,eligible_seed_count,eligible_seeds,gpu_count,known_margin_exceedance,known_memory_failure,logistic_predicted_peak_mib,logistic_predicted_state,logistic_unscored_approval,maximum_known_completed_peak_mib,model,observed_state,required_seeds,unresolved,unresolved_seeds,workload +True,,0,46461.218916058904,memory_failure,False,host-qwen25-7b-gsm8k-c2-2gpu,c2,33108.0,within_margin,False,3,161;162;163,2,False,True,,memory_failure,False,,Qwen/Qwen2.5-7B-Instruct,memory_failure,3,False,,gsm8k +False,,0,38751.70456713468,within_margin,True,host-qwen25-7b-gsm8k-c2-4gpu,c2,29840.0,within_margin,True,0,,4,False,False,,above_margin,False,,Qwen/Qwen2.5-7B-Instruct,unresolved,3,True,161;162;163,gsm8k +True,,0,46391.75739749115,memory_failure,False,host-qwen25-7b-math-c2-2gpu,c2,33446.0,within_margin,False,3,161;162;163,2,False,True,,memory_failure,False,,Qwen/Qwen2.5-7B-Instruct,memory_failure,3,False,,math +False,,0,38682.24304856692,within_margin,True,host-qwen25-7b-math-c2-4gpu,c2,30180.0,within_margin,True,0,,4,False,False,,above_margin,False,,Qwen/Qwen2.5-7B-Instruct,unresolved,3,True,161;162;163,math +True,,0,46443.853536416966,memory_failure,False,host-qwen25-7b-code_heavy_tail-c2-2gpu,c2,33436.0,within_margin,False,3,161;162;163,2,False,True,,memory_failure,False,,Qwen/Qwen2.5-7B-Instruct,memory_failure,3,False,,code_heavy_tail +False,,0,38734.33918749274,within_margin,True,host-qwen25-7b-code_heavy_tail-c2-4gpu,c2,33436.0,within_margin,True,0,,4,False,False,,above_margin,False,,Qwen/Qwen2.5-7B-Instruct,unresolved,3,True,161;162;163,code_heavy_tail diff --git a/benchmark/host_capacity/results/configurations_flat.json b/benchmark/host_capacity/results/configurations_flat.json new file mode 100644 index 0000000000000000000000000000000000000000..51d7f44612cabb8fd259bfc4eec5f8561244d5a3 --- /dev/null +++ b/benchmark/host_capacity/results/configurations_flat.json @@ -0,0 +1,170 @@ +[ + { + "all_three_seeds_eligible": true, + "completed_peak_mib": null, + "completed_processes": 0, + "component_regression_predicted_peak_mib": 46461.218916058904, + "component_regression_predicted_state": "memory_failure", + "component_regression_unscored_approval": false, + "configuration_id": "host-qwen25-7b-gsm8k-c2-2gpu", + "configuration_level": "c2", + "donor_copy_predicted_peak_mib": 33108.0, + "donor_copy_predicted_state": "within_margin", + "donor_copy_unscored_approval": false, + "eligible_seed_count": 3, + "eligible_seeds": "161;162;163", + "gpu_count": 2, + "known_margin_exceedance": false, + "known_memory_failure": true, + "logistic_predicted_peak_mib": null, + "logistic_predicted_state": "memory_failure", + "logistic_unscored_approval": false, + "maximum_known_completed_peak_mib": null, + "model": "Qwen/Qwen2.5-7B-Instruct", + "observed_state": "memory_failure", + "required_seeds": 3, + "unresolved": false, + "unresolved_seeds": "", + "workload": "gsm8k" + }, + { + "all_three_seeds_eligible": false, + "completed_peak_mib": null, + "completed_processes": 0, + "component_regression_predicted_peak_mib": 38751.70456713468, + "component_regression_predicted_state": "within_margin", + "component_regression_unscored_approval": true, + "configuration_id": "host-qwen25-7b-gsm8k-c2-4gpu", + "configuration_level": "c2", + "donor_copy_predicted_peak_mib": 29840.0, + "donor_copy_predicted_state": "within_margin", + "donor_copy_unscored_approval": true, + "eligible_seed_count": 0, + "eligible_seeds": "", + "gpu_count": 4, + "known_margin_exceedance": false, + "known_memory_failure": false, + "logistic_predicted_peak_mib": null, + "logistic_predicted_state": "above_margin", + "logistic_unscored_approval": false, + "maximum_known_completed_peak_mib": null, + "model": "Qwen/Qwen2.5-7B-Instruct", + "observed_state": "unresolved", + "required_seeds": 3, + "unresolved": true, + "unresolved_seeds": "161;162;163", + "workload": "gsm8k" + }, + { + "all_three_seeds_eligible": true, + "completed_peak_mib": null, + "completed_processes": 0, + "component_regression_predicted_peak_mib": 46391.75739749115, + "component_regression_predicted_state": "memory_failure", + "component_regression_unscored_approval": false, + "configuration_id": "host-qwen25-7b-math-c2-2gpu", + "configuration_level": "c2", + "donor_copy_predicted_peak_mib": 33446.0, + "donor_copy_predicted_state": "within_margin", + "donor_copy_unscored_approval": false, + "eligible_seed_count": 3, + "eligible_seeds": "161;162;163", + "gpu_count": 2, + "known_margin_exceedance": false, + "known_memory_failure": true, + "logistic_predicted_peak_mib": null, + "logistic_predicted_state": "memory_failure", + "logistic_unscored_approval": false, + "maximum_known_completed_peak_mib": null, + "model": "Qwen/Qwen2.5-7B-Instruct", + "observed_state": "memory_failure", + "required_seeds": 3, + "unresolved": false, + "unresolved_seeds": "", + "workload": "math" + }, + { + "all_three_seeds_eligible": false, + "completed_peak_mib": null, + "completed_processes": 0, + "component_regression_predicted_peak_mib": 38682.24304856692, + "component_regression_predicted_state": "within_margin", + "component_regression_unscored_approval": true, + "configuration_id": "host-qwen25-7b-math-c2-4gpu", + "configuration_level": "c2", + "donor_copy_predicted_peak_mib": 30180.0, + "donor_copy_predicted_state": "within_margin", + "donor_copy_unscored_approval": true, + "eligible_seed_count": 0, + "eligible_seeds": "", + "gpu_count": 4, + "known_margin_exceedance": false, + "known_memory_failure": false, + "logistic_predicted_peak_mib": null, + "logistic_predicted_state": "above_margin", + "logistic_unscored_approval": false, + "maximum_known_completed_peak_mib": null, + "model": "Qwen/Qwen2.5-7B-Instruct", + "observed_state": "unresolved", + "required_seeds": 3, + "unresolved": true, + "unresolved_seeds": "161;162;163", + "workload": "math" + }, + { + "all_three_seeds_eligible": true, + "completed_peak_mib": null, + "completed_processes": 0, + "component_regression_predicted_peak_mib": 46443.853536416966, + "component_regression_predicted_state": "memory_failure", + "component_regression_unscored_approval": false, + "configuration_id": "host-qwen25-7b-code_heavy_tail-c2-2gpu", + "configuration_level": "c2", + "donor_copy_predicted_peak_mib": 33436.0, + "donor_copy_predicted_state": "within_margin", + "donor_copy_unscored_approval": false, + "eligible_seed_count": 3, + "eligible_seeds": "161;162;163", + "gpu_count": 2, + "known_margin_exceedance": false, + "known_memory_failure": true, + "logistic_predicted_peak_mib": null, + "logistic_predicted_state": "memory_failure", + "logistic_unscored_approval": false, + "maximum_known_completed_peak_mib": null, + "model": "Qwen/Qwen2.5-7B-Instruct", + "observed_state": "memory_failure", + "required_seeds": 3, + "unresolved": false, + "unresolved_seeds": "", + "workload": "code_heavy_tail" + }, + { + "all_three_seeds_eligible": false, + "completed_peak_mib": null, + "completed_processes": 0, + "component_regression_predicted_peak_mib": 38734.33918749274, + "component_regression_predicted_state": "within_margin", + "component_regression_unscored_approval": true, + "configuration_id": "host-qwen25-7b-code_heavy_tail-c2-4gpu", + "configuration_level": "c2", + "donor_copy_predicted_peak_mib": 33436.0, + "donor_copy_predicted_state": "within_margin", + "donor_copy_unscored_approval": true, + "eligible_seed_count": 0, + "eligible_seeds": "", + "gpu_count": 4, + "known_margin_exceedance": false, + "known_memory_failure": false, + "logistic_predicted_peak_mib": null, + "logistic_predicted_state": "above_margin", + "logistic_unscored_approval": false, + "maximum_known_completed_peak_mib": null, + "model": "Qwen/Qwen2.5-7B-Instruct", + "observed_state": "unresolved", + "required_seeds": 3, + "unresolved": true, + "unresolved_seeds": "161;162;163", + "workload": "code_heavy_tail" + } +] diff --git a/benchmark/host_capacity/results/costs.json b/benchmark/host_capacity/results/costs.json new file mode 100644 index 0000000000000000000000000000000000000000..9d1fdef7db043aefdf9caa7a521ddad0f7463061 --- /dev/null +++ b/benchmark/host_capacity/results/costs.json @@ -0,0 +1,504 @@ +{ + "allocations": [ + { + "accounting_valid": true, + "allocation_gpu_count": 4, + "batch_ended_at_epoch": 1789499099.0, + "batch_minus_outer_end_seconds": 0.0, + "batch_started_at_epoch": 1789498576.0, + "batch_state": "FAILED", + "batch_step_gpu_seconds": 2092.0, + "errors": [], + "job_id": "5055846", + "ledger_ended_at_epoch": 1789499098.9258184, + "ledger_minus_outer_end_seconds": -0.07418155670166016, + "observed_allocation_gpu_seconds": 2091.6942930221558, + "pair_id": "host-qwen25-7b-code_heavy_tail-c2-s161", + "pair_ledger_sha256": "333c3a9409bf517fd2ca41c6b17298d3c18ffde6d0780bbd7bd4b980953a7a54", + "pair_manifest_sha256": "11f432ad69e4f19a7e964601235286986f30fb97862cb358860a3b4a45b81ed4", + "raw_record_provenance": { + "batch": { + "line": "5055845_0.batch|5055846.batch|batch|FAILED|2:0|2026-09-15T18:56:16|2026-09-15T19:04:59|523|cpu=64,gres/gpu=4,mem=384G,node=1|acn08|", + "line_number": 2, + "sha256": "8d2ec6009fdc342af9c9ecf6533aeb4e2b9545d9e8f7089fd15995840f55e33a" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5055845_0|5055846|rlvram-host-capacity|FAILED|2:0|2026-09-15T18:56:16|2026-09-15T19:04:59|523|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn08|Dependency", + "line_number": 1, + "sha256": "cab718f549fe31355b9d4784794f53a0f579d9cb7b237d07eef3a62982287423" + }, + "source_sha256": "5c672842d7ad1ef71e6164ae76ba2e74400c45c2511a66d4139489f34664d2eb", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_ended_at_epoch": 1789499099.0, + "scheduler_scope": "outer_job_record", + "scheduler_started_at_epoch": 1789498576.0, + "scheduler_state": "FAILED", + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1, + "whole_allocation_gpu_seconds": 2092.0 + }, + { + "accounting_valid": true, + "allocation_gpu_count": 4, + "batch_ended_at_epoch": 1789499295.0, + "batch_minus_outer_end_seconds": 0.0, + "batch_started_at_epoch": 1789498576.0, + "batch_state": "FAILED", + "batch_step_gpu_seconds": 2876.0, + "errors": [], + "job_id": "5055847", + "ledger_ended_at_epoch": 1789499295.7663517, + "ledger_minus_outer_end_seconds": 0.7663516998291016, + "observed_allocation_gpu_seconds": 2879.0556058883667, + "pair_id": "host-qwen25-7b-gsm8k-c2-s163", + "pair_ledger_sha256": "cde48950a284805dca7150f75182dab4ab0a034cd0b41efeb9fe2b83512a8dd9", + "pair_manifest_sha256": "10a666b083e54c59a9b1ac6144602a9e67fe987c366ba739ea87679570ff7d22", + "raw_record_provenance": { + "batch": { + "line": "5055845_1.batch|5055847.batch|batch|FAILED|2:0|2026-09-15T18:56:16|2026-09-15T19:08:15|719|cpu=64,gres/gpu=4,mem=384G,node=1|acn11|", + "line_number": 5, + "sha256": "8994c5478bad6a55e4b764cd4bc6332b8574a50a3c0742f2b6ad25d30f9a1e47" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5055845_1|5055847|rlvram-host-capacity|FAILED|2:0|2026-09-15T18:56:16|2026-09-15T19:08:15|719|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn11|Dependency", + "line_number": 4, + "sha256": "675fca75c0ea2b4ea5181294d050795a790f58ff267af91ef1c7874598d151a5" + }, + "source_sha256": "5c672842d7ad1ef71e6164ae76ba2e74400c45c2511a66d4139489f34664d2eb", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_ended_at_epoch": 1789499295.0, + "scheduler_scope": "outer_job_record", + "scheduler_started_at_epoch": 1789498576.0, + "scheduler_state": "FAILED", + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1, + "whole_allocation_gpu_seconds": 2876.0 + }, + { + "accounting_valid": true, + "allocation_gpu_count": 4, + "batch_ended_at_epoch": 1789499173.0, + "batch_minus_outer_end_seconds": 0.0, + "batch_started_at_epoch": 1789498576.0, + "batch_state": "FAILED", + "batch_step_gpu_seconds": 2388.0, + "errors": [], + "job_id": "5055848", + "ledger_ended_at_epoch": 1789499173.486668, + "ledger_minus_outer_end_seconds": 0.48666810989379883, + "observed_allocation_gpu_seconds": 2389.9369688034058, + "pair_id": "host-qwen25-7b-code_heavy_tail-c2-s162", + "pair_ledger_sha256": "1c00aa35420037ed210fc575999f0151051f8b4627b06d67904904e489f0c6dd", + "pair_manifest_sha256": "ba5822d60980ed2368ce46612583322f5385a116a98a85abeb700f1080e19e5a", + "raw_record_provenance": { + "batch": { + "line": "5055845_2.batch|5055848.batch|batch|FAILED|2:0|2026-09-15T18:56:16|2026-09-15T19:06:13|597|cpu=64,gres/gpu=4,mem=384G,node=1|acn14|", + "line_number": 8, + "sha256": "1d79ee14f8d1087a12693fe62fa8ad1bc22c64b67998b68463e8ca5e5f82aafb" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5055845_2|5055848|rlvram-host-capacity|FAILED|2:0|2026-09-15T18:56:16|2026-09-15T19:06:13|597|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn14|Dependency", + "line_number": 7, + "sha256": "1e95b564ea1bfd157b77d0238b8e6ebecc1b8a9ef90fdf381d1b6ea3975fc62c" + }, + "source_sha256": "5c672842d7ad1ef71e6164ae76ba2e74400c45c2511a66d4139489f34664d2eb", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_ended_at_epoch": 1789499173.0, + "scheduler_scope": "outer_job_record", + "scheduler_started_at_epoch": 1789498576.0, + "scheduler_state": "FAILED", + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1, + "whole_allocation_gpu_seconds": 2388.0 + }, + { + "accounting_valid": true, + "allocation_gpu_count": 4, + "batch_ended_at_epoch": 1789499600.0, + "batch_minus_outer_end_seconds": 0.0, + "batch_started_at_epoch": 1789499099.0, + "batch_state": "FAILED", + "batch_step_gpu_seconds": 2004.0, + "errors": [], + "job_id": "5055858", + "ledger_ended_at_epoch": 1789499600.3516219, + "ledger_minus_outer_end_seconds": 0.3516218662261963, + "observed_allocation_gpu_seconds": 2005.3962230682373, + "pair_id": "host-qwen25-7b-gsm8k-c2-s162", + "pair_ledger_sha256": "4dcf9a91ae6bbba137a4afab1907896accd528c333c05f9396998563518b7eaa", + "pair_manifest_sha256": "cb05d94260ce1ee8a79015825d072921a10254d9d914a0df3e5c51f3be375fef", + "raw_record_provenance": { + "batch": { + "line": "5055845_3.batch|5055858.batch|batch|FAILED|2:0|2026-09-15T19:04:59|2026-09-15T19:13:20|501|cpu=64,gres/gpu=4,mem=384G,node=1|acn08|", + "line_number": 11, + "sha256": "377935db50f84bc8ca9f7c3106dcd288bdcddf8c9971b0f7b0e9b8128780adb4" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5055845_3|5055858|rlvram-host-capacity|FAILED|2:0|2026-09-15T19:04:59|2026-09-15T19:13:20|501|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn08|JobArrayTaskLimit", + "line_number": 10, + "sha256": "954be8b27acd69a1f217ea3268ca6b29935b717fe64dabd63dd896006f78afd3" + }, + "source_sha256": "5c672842d7ad1ef71e6164ae76ba2e74400c45c2511a66d4139489f34664d2eb", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_ended_at_epoch": 1789499600.0, + "scheduler_scope": "outer_job_record", + "scheduler_started_at_epoch": 1789499099.0, + "scheduler_state": "FAILED", + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1, + "whole_allocation_gpu_seconds": 2004.0 + }, + { + "accounting_valid": true, + "allocation_gpu_count": 4, + "batch_ended_at_epoch": 1789499776.0, + "batch_minus_outer_end_seconds": 0.0, + "batch_started_at_epoch": 1789499173.0, + "batch_state": "FAILED", + "batch_step_gpu_seconds": 2412.0, + "errors": [], + "job_id": "5055859", + "ledger_ended_at_epoch": 1789499776.3182597, + "ledger_minus_outer_end_seconds": 0.31825971603393555, + "observed_allocation_gpu_seconds": 2413.2634019851685, + "pair_id": "host-qwen25-7b-math-c2-s162", + "pair_ledger_sha256": "dc8d64e4c1acaab7d56a2609e8325cf9f18b62861342caf38bd2d2458019e1bb", + "pair_manifest_sha256": "29d4201db5096e9951dd1c70a5df4b57c945d95dfe55d30830b1b4db32ab4186", + "raw_record_provenance": { + "batch": { + "line": "5055845_4.batch|5055859.batch|batch|FAILED|2:0|2026-09-15T19:06:13|2026-09-15T19:16:16|603|cpu=64,gres/gpu=4,mem=384G,node=1|acn14|", + "line_number": 14, + "sha256": "1efa723eb021c52a986801f246f23bd4f75fb9aef2d71ae5df80f03faf660153" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5055845_4|5055859|rlvram-host-capacity|FAILED|2:0|2026-09-15T19:06:13|2026-09-15T19:16:16|603|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn14|JobArrayTaskLimit", + "line_number": 13, + "sha256": "4ef776a727e24cd59151915df8147adc8aaf6c94789c3ac82441b007bcd9d5f4" + }, + "source_sha256": "5c672842d7ad1ef71e6164ae76ba2e74400c45c2511a66d4139489f34664d2eb", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_ended_at_epoch": 1789499776.0, + "scheduler_scope": "outer_job_record", + "scheduler_started_at_epoch": 1789499173.0, + "scheduler_state": "FAILED", + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1, + "whole_allocation_gpu_seconds": 2412.0 + }, + { + "accounting_valid": true, + "allocation_gpu_count": 4, + "batch_ended_at_epoch": 1789499856.0, + "batch_minus_outer_end_seconds": 0.0, + "batch_started_at_epoch": 1789499296.0, + "batch_state": "FAILED", + "batch_step_gpu_seconds": 2240.0, + "errors": [], + "job_id": "5055860", + "ledger_ended_at_epoch": 1789499856.0857725, + "ledger_minus_outer_end_seconds": 0.08577251434326172, + "observed_allocation_gpu_seconds": 2240.333164215088, + "pair_id": "host-qwen25-7b-code_heavy_tail-c2-s163", + "pair_ledger_sha256": "9d18e012ffc9f2064dcbad10dafa59b750ce161497eec51e0e6bc176ce46c3c4", + "pair_manifest_sha256": "d9475a95dfbf2600d351db83bfc49b45774c5e13c2ab968d21a2168b43d9573f", + "raw_record_provenance": { + "batch": { + "line": "5055845_5.batch|5055860.batch|batch|FAILED|2:0|2026-09-15T19:08:16|2026-09-15T19:17:36|560|cpu=64,gres/gpu=4,mem=384G,node=1|acn11|", + "line_number": 17, + "sha256": "ed5d5207ac56d624b57b1eeee06eb615b7dd9e9db639df5a8eaedaa62faff098" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5055845_5|5055860|rlvram-host-capacity|FAILED|2:0|2026-09-15T19:08:16|2026-09-15T19:17:36|560|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn11|JobArrayTaskLimit", + "line_number": 16, + "sha256": "ad137608e28ca1b487e56fd36287ffa9b443c14332b3cd5515b473ef7e98c33d" + }, + "source_sha256": "5c672842d7ad1ef71e6164ae76ba2e74400c45c2511a66d4139489f34664d2eb", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_ended_at_epoch": 1789499856.0, + "scheduler_scope": "outer_job_record", + "scheduler_started_at_epoch": 1789499296.0, + "scheduler_state": "FAILED", + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1, + "whole_allocation_gpu_seconds": 2240.0 + }, + { + "accounting_valid": true, + "allocation_gpu_count": 4, + "batch_ended_at_epoch": 1789500198.0, + "batch_minus_outer_end_seconds": 0.0, + "batch_started_at_epoch": 1789499600.0, + "batch_state": "FAILED", + "batch_step_gpu_seconds": 2392.0, + "errors": [], + "job_id": "5055863", + "ledger_ended_at_epoch": 1789500197.9997396, + "ledger_minus_outer_end_seconds": -0.00026035308837890625, + "observed_allocation_gpu_seconds": 2391.9889993667603, + "pair_id": "host-qwen25-7b-math-c2-s163", + "pair_ledger_sha256": "045f78ee36bcaec336b9f1ab0a0b08096132b5eff8d76089ceaed23da8eeb80a", + "pair_manifest_sha256": "d0fb1b825aaa3e394285a542e31a674611288684d7ef2a2f5eae6a96b4659e95", + "raw_record_provenance": { + "batch": { + "line": "5055845_6.batch|5055863.batch|batch|FAILED|2:0|2026-09-15T19:13:20|2026-09-15T19:23:18|598|cpu=64,gres/gpu=4,mem=384G,node=1|acn08|", + "line_number": 20, + "sha256": "37497680614f9d2c4f625cf5c41f21852e6fe61b3b902a943af6bb45242c370e" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5055845_6|5055863|rlvram-host-capacity|FAILED|2:0|2026-09-15T19:13:20|2026-09-15T19:23:18|598|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn08|JobArrayTaskLimit", + "line_number": 19, + "sha256": "b9800e065af258b257be749953bc183b21b0d6bd2e601d993ac7a57846e651dc" + }, + "source_sha256": "5c672842d7ad1ef71e6164ae76ba2e74400c45c2511a66d4139489f34664d2eb", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_ended_at_epoch": 1789500198.0, + "scheduler_scope": "outer_job_record", + "scheduler_started_at_epoch": 1789499600.0, + "scheduler_state": "FAILED", + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1, + "whole_allocation_gpu_seconds": 2392.0 + }, + { + "accounting_valid": true, + "allocation_gpu_count": 4, + "batch_ended_at_epoch": 1789500300.0, + "batch_minus_outer_end_seconds": 0.0, + "batch_started_at_epoch": 1789499776.0, + "batch_state": "FAILED", + "batch_step_gpu_seconds": 2096.0, + "errors": [], + "job_id": "5055864", + "ledger_ended_at_epoch": 1789500300.5732012, + "ledger_minus_outer_end_seconds": 0.5732011795043945, + "observed_allocation_gpu_seconds": 2098.28288936615, + "pair_id": "host-qwen25-7b-math-c2-s161", + "pair_ledger_sha256": "35af3985e7adf8837ec64dd3380dbac41b910e49a59dcb258c793c7bf4e0b53e", + "pair_manifest_sha256": "da4248ba78237ac4377771fbd810c138222c7d89cc1960cd484ada584e184ab9", + "raw_record_provenance": { + "batch": { + "line": "5055845_7.batch|5055864.batch|batch|FAILED|2:0|2026-09-15T19:16:16|2026-09-15T19:25:00|524|cpu=64,gres/gpu=4,mem=384G,node=1|acn14|", + "line_number": 23, + "sha256": "3c37ebe8e3c09ebc1d6d665c51832daa4b99454c7871c3ecc08f4b032a77b198" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5055845_7|5055864|rlvram-host-capacity|FAILED|2:0|2026-09-15T19:16:16|2026-09-15T19:25:00|524|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn14|JobArrayTaskLimit", + "line_number": 22, + "sha256": "10c525a57304fd6119adae3491797a97fe9892d43e9c05802bd858761d074a53" + }, + "source_sha256": "5c672842d7ad1ef71e6164ae76ba2e74400c45c2511a66d4139489f34664d2eb", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_ended_at_epoch": 1789500300.0, + "scheduler_scope": "outer_job_record", + "scheduler_started_at_epoch": 1789499776.0, + "scheduler_state": "FAILED", + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1, + "whole_allocation_gpu_seconds": 2096.0 + }, + { + "accounting_valid": true, + "allocation_gpu_count": 4, + "batch_ended_at_epoch": 1789500438.0, + "batch_minus_outer_end_seconds": 0.0, + "batch_started_at_epoch": 1789499856.0, + "batch_state": "FAILED", + "batch_step_gpu_seconds": 2328.0, + "errors": [], + "job_id": "5055845", + "ledger_ended_at_epoch": 1789500438.0780666, + "ledger_minus_outer_end_seconds": 0.07806658744812012, + "observed_allocation_gpu_seconds": 2328.303740501404, + "pair_id": "host-qwen25-7b-gsm8k-c2-s161", + "pair_ledger_sha256": "0d00f6eb5b648385af3f6ce1714c02b352b0a05dad2d539984f23e9783c471ce", + "pair_manifest_sha256": "cfa0de3b3cd08dfeacfdce5c3ebf3e3e2cef59bc2da204889e86042c4850708f", + "raw_record_provenance": { + "batch": { + "line": "5055845_8.batch|5055845.batch|batch|FAILED|2:0|2026-09-15T19:17:36|2026-09-15T19:27:18|582|cpu=64,gres/gpu=4,mem=384G,node=1|acn11|", + "line_number": 26, + "sha256": "e7f6404e1a7504049b9b02d7345d9fa709fc768ab8f695ea6dc134ac9d2379cd" + }, + "format_fields": [ + "JobID", + "JobIDRaw", + "JobName", + "State", + "ExitCode", + "Start", + "End", + "ElapsedRaw", + "AllocTRES", + "NodeList", + "Reason" + ], + "outer": { + "line": "5055845_8|5055845|rlvram-host-capacity|FAILED|2:0|2026-09-15T19:17:36|2026-09-15T19:27:18|582|billing=4,cpu=64,gres/gpu=4,mem=384G,node=1|acn11|JobArrayTaskLimit", + "line_number": 25, + "sha256": "b27b72662a890aed168ae0be010cf62255e8950b0dd3c6ba630997d0e2f5f699" + }, + "source_sha256": "5c672842d7ad1ef71e6164ae76ba2e74400c45c2511a66d4139489f34664d2eb", + "time_format": "standard", + "timezone": "UTC" + }, + "scheduler_ended_at_epoch": 1789500438.0, + "scheduler_scope": "outer_job_record", + "scheduler_started_at_epoch": 1789499856.0, + "scheduler_state": "FAILED", + "timestamp_interval": "[reported_epoch, reported_epoch + precision_seconds)", + "timestamp_precision_seconds": 1, + "whole_allocation_gpu_seconds": 2328.0 + } + ], + "batch_step_time": { + "complete_total_gpu_seconds": 20828.0, + "known_gpu_seconds": 20828.0, + "known_records": 9, + "unknown_records": 0 + }, + "invocation_reservation_time": { + "complete_total_gpu_seconds": 20199.935106290504, + "known_gpu_seconds": 20199.935106290504, + "known_records": 18, + "unknown_records": 0 + }, + "observed_allocation_time": { + "complete_total_gpu_seconds": 20838.255286216736, + "known_gpu_seconds": 20838.255286216736, + "known_records": 9, + "unknown_records": 0 + }, + "scheduler_scope_note": "Whole-allocation totals use reported outer Start/End only. Batch-step totals are separate, overlapping spans and must never be added to outer totals. Timestamp precision validates ledger containment but adds no billed seconds.", + "scope": "Task/invocation time includes failed and unresolved work when timing is validated. Observed allocation time ends at runner finalization; whole-job time requires scheduler accounting. The 55-minute limit is not actual elapsed time. Missing records are unknown, not zero cost. No utilization weighting.", + "task_provisioned_time": { + "complete_total_gpu_seconds": 17605.78311114479, + "known_gpu_seconds": 17605.78311114479, + "known_records": 18, + "unknown_records": 0 + }, + "whole_allocation_time": { + "complete_total_gpu_seconds": 20828.0, + "known_gpu_seconds": 20828.0, + "known_records": 9, + "unknown_records": 0 + } +} diff --git a/benchmark/host_capacity/results/evaluation.json b/benchmark/host_capacity/results/evaluation.json new file mode 100644 index 0000000000000000000000000000000000000000..b1875c7c6eb144777d89640860db3ab929ad0c06 --- /dev/null +++ b/benchmark/host_capacity/results/evaluation.json @@ -0,0 +1,197 @@ +{ + "approvals": { + "component_regression": { + "full_panel_approved_configurations": 3, + "full_panel_approved_ids": [ + "host-qwen25-7b-code_heavy_tail-c2-4gpu", + "host-qwen25-7b-gsm8k-c2-4gpu", + "host-qwen25-7b-math-c2-4gpu" + ], + "scored_approved_configurations": 0, + "unscored_approvals_with_known_failure": [], + "unscored_approvals_with_known_margin_exceedance": [], + "unscored_approved_configurations": 3, + "unscored_approved_ids": [ + "host-qwen25-7b-code_heavy_tail-c2-4gpu", + "host-qwen25-7b-gsm8k-c2-4gpu", + "host-qwen25-7b-math-c2-4gpu" + ] + }, + "donor_copy": { + "full_panel_approved_configurations": 6, + "full_panel_approved_ids": [ + "host-qwen25-7b-code_heavy_tail-c2-2gpu", + "host-qwen25-7b-code_heavy_tail-c2-4gpu", + "host-qwen25-7b-gsm8k-c2-2gpu", + "host-qwen25-7b-gsm8k-c2-4gpu", + "host-qwen25-7b-math-c2-2gpu", + "host-qwen25-7b-math-c2-4gpu" + ], + "scored_approved_configurations": 3, + "unscored_approvals_with_known_failure": [], + "unscored_approvals_with_known_margin_exceedance": [], + "unscored_approved_configurations": 3, + "unscored_approved_ids": [ + "host-qwen25-7b-code_heavy_tail-c2-4gpu", + "host-qwen25-7b-gsm8k-c2-4gpu", + "host-qwen25-7b-math-c2-4gpu" + ] + }, + "logistic": { + "full_panel_approved_configurations": 0, + "full_panel_approved_ids": [], + "scored_approved_configurations": 0, + "unscored_approvals_with_known_failure": [], + "unscored_approvals_with_known_margin_exceedance": [], + "unscored_approved_configurations": 0, + "unscored_approved_ids": [] + } + }, + "full_panel_coverage": 0.5, + "interpretation": "Unresolved configurations are unscored, not correct predictions or safe labels.", + "metric_scope": "resolved_subset_only", + "planned_configurations": 6, + "resolved_configurations": 3, + "resolved_ids": [ + "host-qwen25-7b-code_heavy_tail-c2-2gpu", + "host-qwen25-7b-gsm8k-c2-2gpu", + "host-qwen25-7b-math-c2-2gpu" + ], + "resolved_subset_metrics": { + "component_regression": { + "above_margin_among_approved": null, + "approval_rate_on_above_margin_targets": null, + "approval_rate_on_memory_failure_targets": 0.0, + "approved": 0, + "approved_above_margin": 0, + "approved_memory_failure": 0, + "approved_within_margin": 0, + "completed_peak_mae_mib": null, + "completed_peak_mean_signed_error_mib": null, + "completed_peak_prediction_coverage": 0, + "completed_peak_worst_underestimate_mib": null, + "confusion_matrix_predicted_rows_observed_columns": { + "above_margin": { + "above_margin": 0, + "memory_failure": 0, + "within_margin": 0 + }, + "memory_failure": { + "above_margin": 0, + "memory_failure": 3, + "within_margin": 0 + }, + "within_margin": { + "above_margin": 0, + "memory_failure": 0, + "within_margin": 0 + } + }, + "correct_three_state": 3, + "distinct_target_configurations": 3, + "memory_failure_among_approved": null, + "queries": 3, + "rejected_within_margin": 0, + "rejection_rate_on_within_margin_targets": null, + "startup_guard_changed_states": 0, + "target_above_margin": 0, + "target_memory_failure": 3, + "target_within_margin": 0, + "three_state_accuracy": 1.0, + "within_margin_approval_precision": null, + "within_margin_recall": null + }, + "donor_copy": { + "above_margin_among_approved": 0.0, + "approval_rate_on_above_margin_targets": null, + "approval_rate_on_memory_failure_targets": 1.0, + "approved": 3, + "approved_above_margin": 0, + "approved_memory_failure": 3, + "approved_within_margin": 0, + "completed_peak_mae_mib": null, + "completed_peak_mean_signed_error_mib": null, + "completed_peak_prediction_coverage": 0, + "completed_peak_worst_underestimate_mib": null, + "confusion_matrix_predicted_rows_observed_columns": { + "above_margin": { + "above_margin": 0, + "memory_failure": 0, + "within_margin": 0 + }, + "memory_failure": { + "above_margin": 0, + "memory_failure": 0, + "within_margin": 0 + }, + "within_margin": { + "above_margin": 0, + "memory_failure": 3, + "within_margin": 0 + } + }, + "correct_three_state": 0, + "distinct_target_configurations": 3, + "memory_failure_among_approved": 1.0, + "queries": 3, + "rejected_within_margin": 0, + "rejection_rate_on_within_margin_targets": null, + "startup_guard_changed_states": 0, + "target_above_margin": 0, + "target_memory_failure": 3, + "target_within_margin": 0, + "three_state_accuracy": 0.0, + "within_margin_approval_precision": 0.0, + "within_margin_recall": null + }, + "logistic": { + "above_margin_among_approved": null, + "approval_rate_on_above_margin_targets": null, + "approval_rate_on_memory_failure_targets": 0.0, + "approved": 0, + "approved_above_margin": 0, + "approved_memory_failure": 0, + "approved_within_margin": 0, + "completed_peak_mae_mib": null, + "completed_peak_mean_signed_error_mib": null, + "completed_peak_prediction_coverage": 0, + "completed_peak_worst_underestimate_mib": null, + "confusion_matrix_predicted_rows_observed_columns": { + "above_margin": { + "above_margin": 0, + "memory_failure": 0, + "within_margin": 0 + }, + "memory_failure": { + "above_margin": 0, + "memory_failure": 3, + "within_margin": 0 + }, + "within_margin": { + "above_margin": 0, + "memory_failure": 0, + "within_margin": 0 + } + }, + "correct_three_state": 3, + "distinct_target_configurations": 3, + "memory_failure_among_approved": null, + "queries": 3, + "rejected_within_margin": 0, + "rejection_rate_on_within_margin_targets": null, + "startup_guard_changed_states": 0, + "target_above_margin": 0, + "target_memory_failure": 3, + "target_within_margin": 0, + "three_state_accuracy": 1.0, + "within_margin_approval_precision": null, + "within_margin_recall": null + } + }, + "unresolved_configurations": 3, + "unresolved_ids": [ + "host-qwen25-7b-code_heavy_tail-c2-4gpu", + "host-qwen25-7b-gsm8k-c2-4gpu", + "host-qwen25-7b-math-c2-4gpu" + ] +} diff --git a/benchmark/host_capacity/results/output_manifest.json b/benchmark/host_capacity/results/output_manifest.json new file mode 100644 index 0000000000000000000000000000000000000000..1b6f594c52cab12756812d7504e37b72a6834c9b --- /dev/null +++ b/benchmark/host_capacity/results/output_manifest.json @@ -0,0 +1,50 @@ +{ + "configurations.csv": { + "sha256": "7158a2d61b35f03b457ec9987bdc5fe1e2adec93ecbc93ada04e0bcffd3a456f", + "size_bytes": 9761 + }, + "configurations.json": { + "sha256": "df070b3e7f853925ae85c07d65e5a72f61c96a18ff548c337567bfe4282e1c47", + "size_bytes": 12334 + }, + "configurations_flat.csv": { + "sha256": "7c2eee86ed408429abe158a82290b82b6afa2b770ad3b5bf387c7d924f9a1c69", + "size_bytes": 1865 + }, + "configurations_flat.json": { + "sha256": "d7f475771231662c54d243734c5116765679454d4310d6a19207adeaec816532", + "size_bytes": 6251 + }, + "costs.json": { + "sha256": "00a63241afae76e18bcd21b81b91f3a9abe6987710ee3e2eaf3c59dfc9657369", + "size_bytes": 21101 + }, + "evaluation.json": { + "sha256": "ea7bbc7407055ab483797c937524d10748c4b41076dd1bd9bd1b847cee94c327", + "size_bytes": 6625 + }, + "pairs.csv": { + "sha256": "f1c6ca2025a681b3f5185b7f133ec4f2028a79bfa1e0e3b6a8a0954f4478e5ce", + "size_bytes": 1006 + }, + "pairs.json": { + "sha256": "860375b5407614a936838f51511396a1df08948a728de641345a40ade5d5fdf5", + "size_bytes": 3603 + }, + "processes.csv": { + "sha256": "87a1603625747e96f2aa21a338b9ea86b63ade888f249489546387f646ed8dae", + "size_bytes": 9665 + }, + "processes.json": { + "sha256": "87cf0097ef312be33ca1903052904e412ef24485aea7b348354e01e12bca0ad7", + "size_bytes": 24186 + }, + "provenance.json": { + "sha256": "72f176399add01f5275d5fc59b5e8da0e1d34c59a56433fe65cb33e1dc82733a", + "size_bytes": 101792 + }, + "summary.json": { + "sha256": "c8dc099edc466d15e08fae6c9fd708c7a48c1a56fd535fc96c9f8d5424b61ccc", + "size_bytes": 317 + } +} diff --git a/benchmark/host_capacity/results/pairs.csv b/benchmark/host_capacity/results/pairs.csv new file mode 100644 index 0000000000000000000000000000000000000000..334a335766d69697bc4bf61b7ee41eba79b9036a --- /dev/null +++ b/benchmark/host_capacity/results/pairs.csv @@ -0,0 +1,10 @@ +both_completed,both_eligible,configuration_level,dataset,first_task_gpu_count,four_gpu_state,pair_id,peak_difference_4_minus_2_mib,runtime_difference_4_minus_2_seconds,training_seed,two_gpu_state +False,False,c2,code_heavy_tail,4,unresolved,host-qwen25-7b-code_heavy_tail-c2-s161,,,161,memory_failure +False,False,c2,gsm8k,2,unresolved,host-qwen25-7b-gsm8k-c2-s163,,,163,memory_failure +False,False,c2,code_heavy_tail,2,unresolved,host-qwen25-7b-code_heavy_tail-c2-s162,,,162,memory_failure +False,False,c2,gsm8k,4,unresolved,host-qwen25-7b-gsm8k-c2-s162,,,162,memory_failure +False,False,c2,math,2,unresolved,host-qwen25-7b-math-c2-s162,,,162,memory_failure +False,False,c2,code_heavy_tail,2,unresolved,host-qwen25-7b-code_heavy_tail-c2-s163,,,163,memory_failure +False,False,c2,math,4,unresolved,host-qwen25-7b-math-c2-s163,,,163,memory_failure +False,False,c2,math,4,unresolved,host-qwen25-7b-math-c2-s161,,,161,memory_failure +False,False,c2,gsm8k,2,unresolved,host-qwen25-7b-gsm8k-c2-s161,,,161,memory_failure diff --git a/benchmark/host_capacity/results/pairs.json b/benchmark/host_capacity/results/pairs.json new file mode 100644 index 0000000000000000000000000000000000000000..aed932fbc03b30b0cb123b0766cce5480b9336bf --- /dev/null +++ b/benchmark/host_capacity/results/pairs.json @@ -0,0 +1,119 @@ +[ + { + "both_completed": false, + "both_eligible": false, + "configuration_level": "c2", + "dataset": "code_heavy_tail", + "first_task_gpu_count": 4, + "four_gpu_state": "unresolved", + "pair_id": "host-qwen25-7b-code_heavy_tail-c2-s161", + "peak_difference_4_minus_2_mib": null, + "runtime_difference_4_minus_2_seconds": null, + "training_seed": 161, + "two_gpu_state": "memory_failure" + }, + { + "both_completed": false, + "both_eligible": false, + "configuration_level": "c2", + "dataset": "gsm8k", + "first_task_gpu_count": 2, + "four_gpu_state": "unresolved", + "pair_id": "host-qwen25-7b-gsm8k-c2-s163", + "peak_difference_4_minus_2_mib": null, + "runtime_difference_4_minus_2_seconds": null, + "training_seed": 163, + "two_gpu_state": "memory_failure" + }, + { + "both_completed": false, + "both_eligible": false, + "configuration_level": "c2", + "dataset": "code_heavy_tail", + "first_task_gpu_count": 2, + "four_gpu_state": "unresolved", + "pair_id": "host-qwen25-7b-code_heavy_tail-c2-s162", + "peak_difference_4_minus_2_mib": null, + "runtime_difference_4_minus_2_seconds": null, + "training_seed": 162, + "two_gpu_state": "memory_failure" + }, + { + "both_completed": false, + "both_eligible": false, + "configuration_level": "c2", + "dataset": "gsm8k", + "first_task_gpu_count": 4, + "four_gpu_state": "unresolved", + "pair_id": "host-qwen25-7b-gsm8k-c2-s162", + "peak_difference_4_minus_2_mib": null, + "runtime_difference_4_minus_2_seconds": null, + "training_seed": 162, + "two_gpu_state": "memory_failure" + }, + { + "both_completed": false, + "both_eligible": false, + "configuration_level": "c2", + "dataset": "math", + "first_task_gpu_count": 2, + "four_gpu_state": "unresolved", + "pair_id": "host-qwen25-7b-math-c2-s162", + "peak_difference_4_minus_2_mib": null, + "runtime_difference_4_minus_2_seconds": null, + "training_seed": 162, + "two_gpu_state": "memory_failure" + }, + { + "both_completed": false, + "both_eligible": false, + "configuration_level": "c2", + "dataset": "code_heavy_tail", + "first_task_gpu_count": 2, + "four_gpu_state": "unresolved", + "pair_id": "host-qwen25-7b-code_heavy_tail-c2-s163", + "peak_difference_4_minus_2_mib": null, + "runtime_difference_4_minus_2_seconds": null, + "training_seed": 163, + "two_gpu_state": "memory_failure" + }, + { + "both_completed": false, + "both_eligible": false, + "configuration_level": "c2", + "dataset": "math", + "first_task_gpu_count": 4, + "four_gpu_state": "unresolved", + "pair_id": "host-qwen25-7b-math-c2-s163", + "peak_difference_4_minus_2_mib": null, + "runtime_difference_4_minus_2_seconds": null, + "training_seed": 163, + "two_gpu_state": "memory_failure" + }, + { + "both_completed": false, + "both_eligible": false, + "configuration_level": "c2", + "dataset": "math", + "first_task_gpu_count": 4, + "four_gpu_state": "unresolved", + "pair_id": "host-qwen25-7b-math-c2-s161", + "peak_difference_4_minus_2_mib": null, + "runtime_difference_4_minus_2_seconds": null, + "training_seed": 161, + "two_gpu_state": "memory_failure" + }, + { + "both_completed": false, + "both_eligible": false, + "configuration_level": "c2", + "dataset": "gsm8k", + "first_task_gpu_count": 2, + "four_gpu_state": "unresolved", + "pair_id": "host-qwen25-7b-gsm8k-c2-s161", + "peak_difference_4_minus_2_mib": null, + "runtime_difference_4_minus_2_seconds": null, + "training_seed": 161, + "two_gpu_state": "memory_failure" + } +] diff --git a/benchmark/host_capacity/results/processes.csv b/benchmark/host_capacity/results/processes.csv new file mode 100644 index 0000000000000000000000000000000000000000..bc98f1db02de552c81e15b97b31fdc6f62a6f539 --- /dev/null +++ b/benchmark/host_capacity/results/processes.csv @@ -0,0 +1,19 @@ +allocation_gpu_count,artifact_path,completed,completed_peak_mib,configuration_id,configuration_level,dataset,dispatched,durable_terminal_record_present,eligible,expected_execution_commit,experiment_id,failure_kind,failure_stage,invocation_elapsed_seconds,invocation_id,invocation_reservation_gpu_seconds,job_id,known_margin_exceedance,known_memory_failure,launched,observed_task_peak_mib,online_status,pair_id,period,state,task_gpu_count,task_gpu_seconds,task_runtime_seconds,training_seed,validation_errors,worker_initialization +4,output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-4gpu/attempt-5055846,False,,host-qwen25-7b-code_heavy_tail-c2-4gpu,c2,code_heavy_tail,True,True,False,120e29a146294722a6668f55a1e7fed2878eb62a,host-qwen25-7b-code_heavy_tail-c2-s161-4gpu,,training,376.48426548810676,host-qwen25-7b-code_heavy_tail-c2-s161-4gpu-j5055846,1505.937061952427,5055846,False,False,True,,infrastructure_invalid,host-qwen25-7b-code_heavy_tail-c2-s161,1,unresolved,4,1505.937061952427,,161,"[""ValueError: ValueError: Successful exit lacks the exact one-step terminal event""]", +4,output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-2gpu/attempt-5055846,False,,host-qwen25-7b-code_heavy_tail-c2-2gpu,c2,code_heavy_tail,True,True,True,120e29a146294722a6668f55a1e7fed2878eb62a,host-qwen25-7b-code_heavy_tail-c2-s161-2gpu,rollout_init_memory,training,130.87535370094702,host-qwen25-7b-code_heavy_tail-c2-s161-2gpu-j5055846,523.5014148037881,5055846,False,True,True,28882.0,memory_failure,host-qwen25-7b-code_heavy_tail-c2-s161,2,memory_failure,2,261.75070740189403,123.0,161,[],complete +4,output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-2gpu/attempt-5055847,False,,host-qwen25-7b-gsm8k-c2-2gpu,c2,gsm8k,True,True,True,120e29a146294722a6668f55a1e7fed2878eb62a,host-qwen25-7b-gsm8k-c2-s163-2gpu,rollout_init_memory,training,169.41491882689297,host-qwen25-7b-gsm8k-c2-s163-2gpu-j5055847,677.6596753075719,5055847,False,True,True,28882.0,memory_failure,host-qwen25-7b-gsm8k-c2-s163,1,memory_failure,2,338.82983765378594,152.0,163,[],complete +4,output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-4gpu/attempt-5055847,False,,host-qwen25-7b-gsm8k-c2-4gpu,c2,gsm8k,True,True,False,120e29a146294722a6668f55a1e7fed2878eb62a,host-qwen25-7b-gsm8k-c2-s163-4gpu,,training,532.626896739006,host-qwen25-7b-gsm8k-c2-s163-4gpu-j5055847,2130.507586956024,5055847,False,False,True,,infrastructure_invalid,host-qwen25-7b-gsm8k-c2-s163,2,unresolved,4,2130.507586956024,,163,"[""ValueError: ValueError: Successful exit lacks the exact one-step terminal event""]", +4,output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-2gpu/attempt-5055848,False,,host-qwen25-7b-code_heavy_tail-c2-2gpu,c2,code_heavy_tail,True,True,True,120e29a146294722a6668f55a1e7fed2878eb62a,host-qwen25-7b-code_heavy_tail-c2-s162-2gpu,rollout_init_memory,training,169.27469882834703,host-qwen25-7b-code_heavy_tail-c2-s162-2gpu-j5055848,677.0987953133881,5055848,False,True,True,28882.0,memory_failure,host-qwen25-7b-code_heavy_tail-c2-s162,1,memory_failure,2,338.54939765669405,163.0,162,[],complete +4,output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-4gpu/attempt-5055848,False,,host-qwen25-7b-code_heavy_tail-c2-4gpu,c2,code_heavy_tail,True,True,False,120e29a146294722a6668f55a1e7fed2878eb62a,host-qwen25-7b-code_heavy_tail-c2-s162-4gpu,,training,409.4206085870974,host-qwen25-7b-code_heavy_tail-c2-s162-4gpu-j5055848,1637.6824343483895,5055848,False,False,True,,infrastructure_invalid,host-qwen25-7b-code_heavy_tail-c2-s162,2,unresolved,4,1637.6824343483895,,162,"[""ValueError: ValueError: Successful exit lacks the exact one-step terminal event""]", +4,output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-4gpu/attempt-5055858,False,,host-qwen25-7b-gsm8k-c2-4gpu,c2,gsm8k,True,True,False,120e29a146294722a6668f55a1e7fed2878eb62a,host-qwen25-7b-gsm8k-c2-s162-4gpu,,training,335.5115462481044,host-qwen25-7b-gsm8k-c2-s162-4gpu-j5055858,1342.0461849924177,5055858,False,False,True,,infrastructure_invalid,host-qwen25-7b-gsm8k-c2-s162,1,unresolved,4,1342.0461849924177,,162,"[""ValueError: ValueError: Successful exit lacks the exact one-step terminal event""]", +4,output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-2gpu/attempt-5055858,False,,host-qwen25-7b-gsm8k-c2-2gpu,c2,gsm8k,True,True,True,120e29a146294722a6668f55a1e7fed2878eb62a,host-qwen25-7b-gsm8k-c2-s162-2gpu,rollout_init_memory,training,149.68433167319745,host-qwen25-7b-gsm8k-c2-s162-2gpu-j5055858,598.7373266927898,5055858,False,True,True,28882.0,memory_failure,host-qwen25-7b-gsm8k-c2-s162,2,memory_failure,2,299.3686633463949,143.0,162,[],complete +4,output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-2gpu/attempt-5055859,False,,host-qwen25-7b-math-c2-2gpu,c2,math,True,True,True,120e29a146294722a6668f55a1e7fed2878eb62a,host-qwen25-7b-math-c2-s162-2gpu,rollout_init_memory,training,134.65871287789196,host-qwen25-7b-math-c2-s162-2gpu-j5055859,538.6348515115678,5055859,False,True,True,28882.0,memory_failure,host-qwen25-7b-math-c2-s162,1,memory_failure,2,269.3174257557839,128.0,162,[],complete +4,output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-4gpu/attempt-5055859,False,,host-qwen25-7b-math-c2-4gpu,c2,math,True,True,False,120e29a146294722a6668f55a1e7fed2878eb62a,host-qwen25-7b-math-c2-s162-4gpu,,training,450.53688256815076,host-qwen25-7b-math-c2-s162-4gpu-j5055859,1802.147530272603,5055859,False,False,True,,infrastructure_invalid,host-qwen25-7b-math-c2-s162,2,unresolved,4,1802.147530272603,,162,"[""ValueError: ValueError: Successful exit lacks the exact one-step terminal event""]", +4,output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-2gpu/attempt-5055860,False,,host-qwen25-7b-code_heavy_tail-c2-2gpu,c2,code_heavy_tail,True,True,True,120e29a146294722a6668f55a1e7fed2878eb62a,host-qwen25-7b-code_heavy_tail-c2-s163-2gpu,rollout_init_memory,training,143.88826199201867,host-qwen25-7b-code_heavy_tail-c2-s163-2gpu-j5055860,575.5530479680747,5055860,False,True,True,28882.0,memory_failure,host-qwen25-7b-code_heavy_tail-c2-s163,1,memory_failure,2,287.77652398403734,137.0,163,[],complete +4,output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-4gpu/attempt-5055860,False,,host-qwen25-7b-code_heavy_tail-c2-4gpu,c2,code_heavy_tail,True,True,False,120e29a146294722a6668f55a1e7fed2878eb62a,host-qwen25-7b-code_heavy_tail-c2-s163-4gpu,,training,397.81703199539334,host-qwen25-7b-code_heavy_tail-c2-s163-4gpu-j5055860,1591.2681279815733,5055860,False,False,True,,infrastructure_invalid,host-qwen25-7b-code_heavy_tail-c2-s163,2,unresolved,4,1591.2681279815733,,163,"[""ValueError: ValueError: Successful exit lacks the exact one-step terminal event""]", +4,output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-4gpu/attempt-5055863,False,,host-qwen25-7b-math-c2-4gpu,c2,math,True,True,False,120e29a146294722a6668f55a1e7fed2878eb62a,host-qwen25-7b-math-c2-s163-4gpu,,training,448.6107034590095,host-qwen25-7b-math-c2-s163-4gpu-j5055863,1794.442813836038,5055863,False,False,True,,infrastructure_invalid,host-qwen25-7b-math-c2-s163,1,unresolved,4,1794.442813836038,,163,"[""ValueError: ValueError: Successful exit lacks the exact one-step terminal event""]", +4,output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-2gpu/attempt-5055863,False,,host-qwen25-7b-math-c2-2gpu,c2,math,True,True,True,120e29a146294722a6668f55a1e7fed2878eb62a,host-qwen25-7b-math-c2-s163-2gpu,rollout_init_memory,training,131.28932780772448,host-qwen25-7b-math-c2-s163-2gpu-j5055863,525.1573112308979,5055863,False,True,True,28882.0,memory_failure,host-qwen25-7b-math-c2-s163,2,memory_failure,2,262.57865561544895,125.0,163,[],complete +4,output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-4gpu/attempt-5055864,False,,host-qwen25-7b-math-c2-4gpu,c2,math,True,True,False,120e29a146294722a6668f55a1e7fed2878eb62a,host-qwen25-7b-math-c2-s161-4gpu,,training,365.2147745988332,host-qwen25-7b-math-c2-s161-4gpu-j5055864,1460.8590983953327,5055864,False,False,True,,infrastructure_invalid,host-qwen25-7b-math-c2-s161,1,unresolved,4,1460.8590983953327,,161,"[""ValueError: ValueError: Successful exit lacks the exact one-step terminal event""]", +4,output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-2gpu/attempt-5055864,False,,host-qwen25-7b-math-c2-2gpu,c2,math,True,True,True,120e29a146294722a6668f55a1e7fed2878eb62a,host-qwen25-7b-math-c2-s161-2gpu,rollout_init_memory,training,140.9088330278173,host-qwen25-7b-math-c2-s161-2gpu-j5055864,563.6353321112692,5055864,False,True,True,28882.0,memory_failure,host-qwen25-7b-math-c2-s161,2,memory_failure,2,281.8176660556346,134.0,161,[],complete +4,output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-2gpu/attempt-5055845,False,,host-qwen25-7b-gsm8k-c2-2gpu,c2,gsm8k,True,True,True,120e29a146294722a6668f55a1e7fed2878eb62a,host-qwen25-7b-gsm8k-c2-s161-2gpu,rollout_init_memory,training,127.08155883802101,host-qwen25-7b-gsm8k-c2-s161-2gpu-j5055845,508.32623535208404,5055845,False,True,True,28882.0,memory_failure,host-qwen25-7b-gsm8k-c2-s161,1,memory_failure,2,254.16311767604202,120.0,161,[],complete +4,output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-4gpu/attempt-5055845,False,,host-qwen25-7b-gsm8k-c2-4gpu,c2,gsm8k,True,True,False,120e29a146294722a6668f55a1e7fed2878eb62a,host-qwen25-7b-gsm8k-c2-s161-4gpu,,training,436.6850693160668,host-qwen25-7b-gsm8k-c2-s161-4gpu-j5055845,1746.7402772642672,5055845,False,False,True,,infrastructure_invalid,host-qwen25-7b-gsm8k-c2-s161,2,unresolved,4,1746.7402772642672,,161,"[""ValueError: ValueError: Successful exit lacks the exact one-step terminal event""]", diff --git a/benchmark/host_capacity/results/processes.json b/benchmark/host_capacity/results/processes.json new file mode 100644 index 0000000000000000000000000000000000000000..34178d6753c48d453a66897f5b3016c06d479335 --- /dev/null +++ b/benchmark/host_capacity/results/processes.json @@ -0,0 +1,632 @@ +[ + { + "allocation_gpu_count": 4, + "artifact_path": "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-4gpu/attempt-5055846", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "host-qwen25-7b-code_heavy_tail-c2-4gpu", + "configuration_level": "c2", + "dataset": "code_heavy_tail", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": false, + "expected_execution_commit": "120e29a146294722a6668f55a1e7fed2878eb62a", + "experiment_id": "host-qwen25-7b-code_heavy_tail-c2-s161-4gpu", + "failure_kind": null, + "failure_stage": "training", + "invocation_elapsed_seconds": 376.48426548810676, + "invocation_id": "host-qwen25-7b-code_heavy_tail-c2-s161-4gpu-j5055846", + "invocation_reservation_gpu_seconds": 1505.937061952427, + "job_id": "5055846", + "known_margin_exceedance": false, + "known_memory_failure": false, + "launched": true, + "observed_task_peak_mib": null, + "online_status": "infrastructure_invalid", + "pair_id": "host-qwen25-7b-code_heavy_tail-c2-s161", + "period": 1, + "state": "unresolved", + "task_gpu_count": 4, + "task_gpu_seconds": 1505.937061952427, + "task_runtime_seconds": null, + "training_seed": 161, + "validation_errors": [ + "ValueError: ValueError: Successful exit lacks the exact one-step terminal event" + ], + "worker_initialization": null + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-2gpu/attempt-5055846", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "host-qwen25-7b-code_heavy_tail-c2-2gpu", + "configuration_level": "c2", + "dataset": "code_heavy_tail", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "120e29a146294722a6668f55a1e7fed2878eb62a", + "experiment_id": "host-qwen25-7b-code_heavy_tail-c2-s161-2gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 130.87535370094702, + "invocation_id": "host-qwen25-7b-code_heavy_tail-c2-s161-2gpu-j5055846", + "invocation_reservation_gpu_seconds": 523.5014148037881, + "job_id": "5055846", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 28882.0, + "online_status": "memory_failure", + "pair_id": "host-qwen25-7b-code_heavy_tail-c2-s161", + "period": 2, + "state": "memory_failure", + "task_gpu_count": 2, + "task_gpu_seconds": 261.75070740189403, + "task_runtime_seconds": 123.0, + "training_seed": 161, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-2gpu/attempt-5055847", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "host-qwen25-7b-gsm8k-c2-2gpu", + "configuration_level": "c2", + "dataset": "gsm8k", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "120e29a146294722a6668f55a1e7fed2878eb62a", + "experiment_id": "host-qwen25-7b-gsm8k-c2-s163-2gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 169.41491882689297, + "invocation_id": "host-qwen25-7b-gsm8k-c2-s163-2gpu-j5055847", + "invocation_reservation_gpu_seconds": 677.6596753075719, + "job_id": "5055847", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 28882.0, + "online_status": "memory_failure", + "pair_id": "host-qwen25-7b-gsm8k-c2-s163", + "period": 1, + "state": "memory_failure", + "task_gpu_count": 2, + "task_gpu_seconds": 338.82983765378594, + "task_runtime_seconds": 152.0, + "training_seed": 163, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-4gpu/attempt-5055847", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "host-qwen25-7b-gsm8k-c2-4gpu", + "configuration_level": "c2", + "dataset": "gsm8k", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": false, + "expected_execution_commit": "120e29a146294722a6668f55a1e7fed2878eb62a", + "experiment_id": "host-qwen25-7b-gsm8k-c2-s163-4gpu", + "failure_kind": null, + "failure_stage": "training", + "invocation_elapsed_seconds": 532.626896739006, + "invocation_id": "host-qwen25-7b-gsm8k-c2-s163-4gpu-j5055847", + "invocation_reservation_gpu_seconds": 2130.507586956024, + "job_id": "5055847", + "known_margin_exceedance": false, + "known_memory_failure": false, + "launched": true, + "observed_task_peak_mib": null, + "online_status": "infrastructure_invalid", + "pair_id": "host-qwen25-7b-gsm8k-c2-s163", + "period": 2, + "state": "unresolved", + "task_gpu_count": 4, + "task_gpu_seconds": 2130.507586956024, + "task_runtime_seconds": null, + "training_seed": 163, + "validation_errors": [ + "ValueError: ValueError: Successful exit lacks the exact one-step terminal event" + ], + "worker_initialization": null + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-2gpu/attempt-5055848", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "host-qwen25-7b-code_heavy_tail-c2-2gpu", + "configuration_level": "c2", + "dataset": "code_heavy_tail", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "120e29a146294722a6668f55a1e7fed2878eb62a", + "experiment_id": "host-qwen25-7b-code_heavy_tail-c2-s162-2gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 169.27469882834703, + "invocation_id": "host-qwen25-7b-code_heavy_tail-c2-s162-2gpu-j5055848", + "invocation_reservation_gpu_seconds": 677.0987953133881, + "job_id": "5055848", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 28882.0, + "online_status": "memory_failure", + "pair_id": "host-qwen25-7b-code_heavy_tail-c2-s162", + "period": 1, + "state": "memory_failure", + "task_gpu_count": 2, + "task_gpu_seconds": 338.54939765669405, + "task_runtime_seconds": 163.0, + "training_seed": 162, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-4gpu/attempt-5055848", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "host-qwen25-7b-code_heavy_tail-c2-4gpu", + "configuration_level": "c2", + "dataset": "code_heavy_tail", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": false, + "expected_execution_commit": "120e29a146294722a6668f55a1e7fed2878eb62a", + "experiment_id": "host-qwen25-7b-code_heavy_tail-c2-s162-4gpu", + "failure_kind": null, + "failure_stage": "training", + "invocation_elapsed_seconds": 409.4206085870974, + "invocation_id": "host-qwen25-7b-code_heavy_tail-c2-s162-4gpu-j5055848", + "invocation_reservation_gpu_seconds": 1637.6824343483895, + "job_id": "5055848", + "known_margin_exceedance": false, + "known_memory_failure": false, + "launched": true, + "observed_task_peak_mib": null, + "online_status": "infrastructure_invalid", + "pair_id": "host-qwen25-7b-code_heavy_tail-c2-s162", + "period": 2, + "state": "unresolved", + "task_gpu_count": 4, + "task_gpu_seconds": 1637.6824343483895, + "task_runtime_seconds": null, + "training_seed": 162, + "validation_errors": [ + "ValueError: ValueError: Successful exit lacks the exact one-step terminal event" + ], + "worker_initialization": null + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-4gpu/attempt-5055858", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "host-qwen25-7b-gsm8k-c2-4gpu", + "configuration_level": "c2", + "dataset": "gsm8k", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": false, + "expected_execution_commit": "120e29a146294722a6668f55a1e7fed2878eb62a", + "experiment_id": "host-qwen25-7b-gsm8k-c2-s162-4gpu", + "failure_kind": null, + "failure_stage": "training", + "invocation_elapsed_seconds": 335.5115462481044, + "invocation_id": "host-qwen25-7b-gsm8k-c2-s162-4gpu-j5055858", + "invocation_reservation_gpu_seconds": 1342.0461849924177, + "job_id": "5055858", + "known_margin_exceedance": false, + "known_memory_failure": false, + "launched": true, + "observed_task_peak_mib": null, + "online_status": "infrastructure_invalid", + "pair_id": "host-qwen25-7b-gsm8k-c2-s162", + "period": 1, + "state": "unresolved", + "task_gpu_count": 4, + "task_gpu_seconds": 1342.0461849924177, + "task_runtime_seconds": null, + "training_seed": 162, + "validation_errors": [ + "ValueError: ValueError: Successful exit lacks the exact one-step terminal event" + ], + "worker_initialization": null + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-2gpu/attempt-5055858", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "host-qwen25-7b-gsm8k-c2-2gpu", + "configuration_level": "c2", + "dataset": "gsm8k", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "120e29a146294722a6668f55a1e7fed2878eb62a", + "experiment_id": "host-qwen25-7b-gsm8k-c2-s162-2gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 149.68433167319745, + "invocation_id": "host-qwen25-7b-gsm8k-c2-s162-2gpu-j5055858", + "invocation_reservation_gpu_seconds": 598.7373266927898, + "job_id": "5055858", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 28882.0, + "online_status": "memory_failure", + "pair_id": "host-qwen25-7b-gsm8k-c2-s162", + "period": 2, + "state": "memory_failure", + "task_gpu_count": 2, + "task_gpu_seconds": 299.3686633463949, + "task_runtime_seconds": 143.0, + "training_seed": 162, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-2gpu/attempt-5055859", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "host-qwen25-7b-math-c2-2gpu", + "configuration_level": "c2", + "dataset": "math", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "120e29a146294722a6668f55a1e7fed2878eb62a", + "experiment_id": "host-qwen25-7b-math-c2-s162-2gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 134.65871287789196, + "invocation_id": "host-qwen25-7b-math-c2-s162-2gpu-j5055859", + "invocation_reservation_gpu_seconds": 538.6348515115678, + "job_id": "5055859", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 28882.0, + "online_status": "memory_failure", + "pair_id": "host-qwen25-7b-math-c2-s162", + "period": 1, + "state": "memory_failure", + "task_gpu_count": 2, + "task_gpu_seconds": 269.3174257557839, + "task_runtime_seconds": 128.0, + "training_seed": 162, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-4gpu/attempt-5055859", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "host-qwen25-7b-math-c2-4gpu", + "configuration_level": "c2", + "dataset": "math", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": false, + "expected_execution_commit": "120e29a146294722a6668f55a1e7fed2878eb62a", + "experiment_id": "host-qwen25-7b-math-c2-s162-4gpu", + "failure_kind": null, + "failure_stage": "training", + "invocation_elapsed_seconds": 450.53688256815076, + "invocation_id": "host-qwen25-7b-math-c2-s162-4gpu-j5055859", + "invocation_reservation_gpu_seconds": 1802.147530272603, + "job_id": "5055859", + "known_margin_exceedance": false, + "known_memory_failure": false, + "launched": true, + "observed_task_peak_mib": null, + "online_status": "infrastructure_invalid", + "pair_id": "host-qwen25-7b-math-c2-s162", + "period": 2, + "state": "unresolved", + "task_gpu_count": 4, + "task_gpu_seconds": 1802.147530272603, + "task_runtime_seconds": null, + "training_seed": 162, + "validation_errors": [ + "ValueError: ValueError: Successful exit lacks the exact one-step terminal event" + ], + "worker_initialization": null + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-2gpu/attempt-5055860", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "host-qwen25-7b-code_heavy_tail-c2-2gpu", + "configuration_level": "c2", + "dataset": "code_heavy_tail", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "120e29a146294722a6668f55a1e7fed2878eb62a", + "experiment_id": "host-qwen25-7b-code_heavy_tail-c2-s163-2gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 143.88826199201867, + "invocation_id": "host-qwen25-7b-code_heavy_tail-c2-s163-2gpu-j5055860", + "invocation_reservation_gpu_seconds": 575.5530479680747, + "job_id": "5055860", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 28882.0, + "online_status": "memory_failure", + "pair_id": "host-qwen25-7b-code_heavy_tail-c2-s163", + "period": 1, + "state": "memory_failure", + "task_gpu_count": 2, + "task_gpu_seconds": 287.77652398403734, + "task_runtime_seconds": 137.0, + "training_seed": 163, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-4gpu/attempt-5055860", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "host-qwen25-7b-code_heavy_tail-c2-4gpu", + "configuration_level": "c2", + "dataset": "code_heavy_tail", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": false, + "expected_execution_commit": "120e29a146294722a6668f55a1e7fed2878eb62a", + "experiment_id": "host-qwen25-7b-code_heavy_tail-c2-s163-4gpu", + "failure_kind": null, + "failure_stage": "training", + "invocation_elapsed_seconds": 397.81703199539334, + "invocation_id": "host-qwen25-7b-code_heavy_tail-c2-s163-4gpu-j5055860", + "invocation_reservation_gpu_seconds": 1591.2681279815733, + "job_id": "5055860", + "known_margin_exceedance": false, + "known_memory_failure": false, + "launched": true, + "observed_task_peak_mib": null, + "online_status": "infrastructure_invalid", + "pair_id": "host-qwen25-7b-code_heavy_tail-c2-s163", + "period": 2, + "state": "unresolved", + "task_gpu_count": 4, + "task_gpu_seconds": 1591.2681279815733, + "task_runtime_seconds": null, + "training_seed": 163, + "validation_errors": [ + "ValueError: ValueError: Successful exit lacks the exact one-step terminal event" + ], + "worker_initialization": null + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-4gpu/attempt-5055863", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "host-qwen25-7b-math-c2-4gpu", + "configuration_level": "c2", + "dataset": "math", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": false, + "expected_execution_commit": "120e29a146294722a6668f55a1e7fed2878eb62a", + "experiment_id": "host-qwen25-7b-math-c2-s163-4gpu", + "failure_kind": null, + "failure_stage": "training", + "invocation_elapsed_seconds": 448.6107034590095, + "invocation_id": "host-qwen25-7b-math-c2-s163-4gpu-j5055863", + "invocation_reservation_gpu_seconds": 1794.442813836038, + "job_id": "5055863", + "known_margin_exceedance": false, + "known_memory_failure": false, + "launched": true, + "observed_task_peak_mib": null, + "online_status": "infrastructure_invalid", + "pair_id": "host-qwen25-7b-math-c2-s163", + "period": 1, + "state": "unresolved", + "task_gpu_count": 4, + "task_gpu_seconds": 1794.442813836038, + "task_runtime_seconds": null, + "training_seed": 163, + "validation_errors": [ + "ValueError: ValueError: Successful exit lacks the exact one-step terminal event" + ], + "worker_initialization": null + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-2gpu/attempt-5055863", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "host-qwen25-7b-math-c2-2gpu", + "configuration_level": "c2", + "dataset": "math", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "120e29a146294722a6668f55a1e7fed2878eb62a", + "experiment_id": "host-qwen25-7b-math-c2-s163-2gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 131.28932780772448, + "invocation_id": "host-qwen25-7b-math-c2-s163-2gpu-j5055863", + "invocation_reservation_gpu_seconds": 525.1573112308979, + "job_id": "5055863", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 28882.0, + "online_status": "memory_failure", + "pair_id": "host-qwen25-7b-math-c2-s163", + "period": 2, + "state": "memory_failure", + "task_gpu_count": 2, + "task_gpu_seconds": 262.57865561544895, + "task_runtime_seconds": 125.0, + "training_seed": 163, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-4gpu/attempt-5055864", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "host-qwen25-7b-math-c2-4gpu", + "configuration_level": "c2", + "dataset": "math", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": false, + "expected_execution_commit": "120e29a146294722a6668f55a1e7fed2878eb62a", + "experiment_id": "host-qwen25-7b-math-c2-s161-4gpu", + "failure_kind": null, + "failure_stage": "training", + "invocation_elapsed_seconds": 365.2147745988332, + "invocation_id": "host-qwen25-7b-math-c2-s161-4gpu-j5055864", + "invocation_reservation_gpu_seconds": 1460.8590983953327, + "job_id": "5055864", + "known_margin_exceedance": false, + "known_memory_failure": false, + "launched": true, + "observed_task_peak_mib": null, + "online_status": "infrastructure_invalid", + "pair_id": "host-qwen25-7b-math-c2-s161", + "period": 1, + "state": "unresolved", + "task_gpu_count": 4, + "task_gpu_seconds": 1460.8590983953327, + "task_runtime_seconds": null, + "training_seed": 161, + "validation_errors": [ + "ValueError: ValueError: Successful exit lacks the exact one-step terminal event" + ], + "worker_initialization": null + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-2gpu/attempt-5055864", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "host-qwen25-7b-math-c2-2gpu", + "configuration_level": "c2", + "dataset": "math", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "120e29a146294722a6668f55a1e7fed2878eb62a", + "experiment_id": "host-qwen25-7b-math-c2-s161-2gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 140.9088330278173, + "invocation_id": "host-qwen25-7b-math-c2-s161-2gpu-j5055864", + "invocation_reservation_gpu_seconds": 563.6353321112692, + "job_id": "5055864", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 28882.0, + "online_status": "memory_failure", + "pair_id": "host-qwen25-7b-math-c2-s161", + "period": 2, + "state": "memory_failure", + "task_gpu_count": 2, + "task_gpu_seconds": 281.8176660556346, + "task_runtime_seconds": 134.0, + "training_seed": 161, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-2gpu/attempt-5055845", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "host-qwen25-7b-gsm8k-c2-2gpu", + "configuration_level": "c2", + "dataset": "gsm8k", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": true, + "expected_execution_commit": "120e29a146294722a6668f55a1e7fed2878eb62a", + "experiment_id": "host-qwen25-7b-gsm8k-c2-s161-2gpu", + "failure_kind": "rollout_init_memory", + "failure_stage": "training", + "invocation_elapsed_seconds": 127.08155883802101, + "invocation_id": "host-qwen25-7b-gsm8k-c2-s161-2gpu-j5055845", + "invocation_reservation_gpu_seconds": 508.32623535208404, + "job_id": "5055845", + "known_margin_exceedance": false, + "known_memory_failure": true, + "launched": true, + "observed_task_peak_mib": 28882.0, + "online_status": "memory_failure", + "pair_id": "host-qwen25-7b-gsm8k-c2-s161", + "period": 1, + "state": "memory_failure", + "task_gpu_count": 2, + "task_gpu_seconds": 254.16311767604202, + "task_runtime_seconds": 120.0, + "training_seed": 161, + "validation_errors": [], + "worker_initialization": "complete" + }, + { + "allocation_gpu_count": 4, + "artifact_path": "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-4gpu/attempt-5055845", + "completed": false, + "completed_peak_mib": null, + "configuration_id": "host-qwen25-7b-gsm8k-c2-4gpu", + "configuration_level": "c2", + "dataset": "gsm8k", + "dispatched": true, + "durable_terminal_record_present": true, + "eligible": false, + "expected_execution_commit": "120e29a146294722a6668f55a1e7fed2878eb62a", + "experiment_id": "host-qwen25-7b-gsm8k-c2-s161-4gpu", + "failure_kind": null, + "failure_stage": "training", + "invocation_elapsed_seconds": 436.6850693160668, + "invocation_id": "host-qwen25-7b-gsm8k-c2-s161-4gpu-j5055845", + "invocation_reservation_gpu_seconds": 1746.7402772642672, + "job_id": "5055845", + "known_margin_exceedance": false, + "known_memory_failure": false, + "launched": true, + "observed_task_peak_mib": null, + "online_status": "infrastructure_invalid", + "pair_id": "host-qwen25-7b-gsm8k-c2-s161", + "period": 2, + "state": "unresolved", + "task_gpu_count": 4, + "task_gpu_seconds": 1746.7402772642672, + "task_runtime_seconds": null, + "training_seed": 161, + "validation_errors": [ + "ValueError: ValueError: Successful exit lacks the exact one-step terminal event" + ], + "worker_initialization": null + } +] diff --git a/benchmark/host_capacity/results/provenance.json b/benchmark/host_capacity/results/provenance.json new file mode 100644 index 0000000000000000000000000000000000000000..0fe81214511cbc09491d567b77cdb71ea97f9b30 --- /dev/null +++ b/benchmark/host_capacity/results/provenance.json @@ -0,0 +1,1741 @@ +{ + "accounting_validation_implementation_sha256": "9b00de319d8ac76d32897322bfbdfde47248737436854111f1e1808f156d4d08", + "allocation_accounting_sha256": "8df8040c1cd1608abd10759dd42448d734aa1b501733c926978247d2177f12ee", + "amendments_sha256": {}, + "analysis_implementation_sha256": "8326859efb69b0c0f7003932d1b491efb80618671437e87d77edbb3ec5f484ba", + "contamination_scope": "All-four allocation samples, excluded-device process membership, and recorded live-child ownership annotations. This is not continuous monitoring or an independent offline reconstruction of PID ancestry.", + "data_verification_scope": "Environment-captured dataset file hashes compared with frozen protocol; no external dataset/cache paths are opened.", + "evaluation_seeds": [ + 161, + 162, + 163 + ], + "execution_commits_expected": { + "host-qwen25-7b-code_heavy_tail-c2-s161": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-code_heavy_tail-c2-s162": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-code_heavy_tail-c2-s163": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-gsm8k-c2-s161": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-gsm8k-c2-s162": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-gsm8k-c2-s163": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-math-c2-s161": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-math-c2-s162": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-math-c2-s163": "120e29a146294722a6668f55a1e7fed2878eb62a" + }, + "execution_commits_file_sha256": "117b2449b2365db03f5af05aeae2c2943421ab739e5fd3c0ae16e3ce67ac5fcc", + "followup_execution_implementation_sha256": { + "memory_tuner/host_capacity_protocol.py": "9bf811457f04e02e67e31d3c9289d1155f73ee1364a233b6b77d08d86becf44b", + "memory_tuner/run_host_capacity.py": "23f1eaf462f1825a77e91d2c8d8a4ab073f4debe0fbab134ff1863e096345d37", + "memory_tuner/run_matched_gpu.py": "f340fb7eec5297cb675a48e267618c15d7787a42891d4a2ae4a670ec24a0cc47", + "run_host_capacity.slurm": "59d3daa9499d1fc76d0b856be6c0a813472f488eb49f990e6ffa57ccf819b919" + }, + "input_evidence": { + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-2gpu/attempt-5055846/device-evidence/probe-exit-idle.json": { + "sha256": "317a8229f1b7a033ae6cea9ff8b03ca210281693e19f65878528da02f52f356b", + "size_bytes": 3698 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-2gpu/attempt-5055846/device-evidence/ray-resources.json": { + "sha256": "610908fab2f887705d78ce4cfed5bb2ecb3038871ff5bfde8f12bb315939ce99", + "size_bytes": 1562 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-2gpu/attempt-5055846/device-evidence/trainer-cuda.json": { + "sha256": "8e2e176c981d952af70e0964bd3eb8ac8a48a5a75b98a93e1cf3d3f9d1636079", + "size_bytes": 583 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-2gpu/attempt-5055846/device-evidence/trainer-exec.json": { + "sha256": "f9290c8a8267e4ccd99dcc3338e9fdca20de37469a768ac6c0aa4c362f7a4c8f", + "size_bytes": 376 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-2gpu/attempt-5055846/device-evidence/trainer-terminal.json": { + "sha256": "67e1f2dd748e16079c5e284434b3ffbdc1ea75bf2c0f047c6a6cd8c629d5e7f0", + "size_bytes": 174 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-2gpu/attempt-5055846/device-evidence/worker-actor_rollout-0-785045.json": { + "sha256": "48198fb0e30f9fc06f6cb14f97dccc5ccdcb5470d1149e2e836b6cc970ff6e9a", + "size_bytes": 666 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-2gpu/attempt-5055846/device-evidence/worker-actor_rollout-1-785046.json": { + "sha256": "c48dea5650a9eda288922dffb3f31a0ee01649668835ce43bc48e948b1aee86f", + "size_bytes": 666 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-2gpu/attempt-5055846/dispatch.json": { + "sha256": "0b5219038e55e417a00cfbfc8a84e45472b382d994c2eb0d2474b3202749859f", + "size_bytes": 4630 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-2gpu/attempt-5055846/environment-5055846.json": { + "sha256": "ad77dda0d199c07c339f028e57459e8bc8c4079923fcaa0bd120d01a37945df4", + "size_bytes": 4480 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-2gpu/attempt-5055846/gpu-device-map.json": { + "sha256": "48fc347cb1f7442021896590ef2acf5c73cece9137f96512ae24160e52162e25", + "size_bytes": 481 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-2gpu/attempt-5055846/gpu-memory-5055846.csv": { + "sha256": "00fc8fc99defa900cc3b0488493cba4ee717574cb93aa1dcb35a1867d976395f", + "size_bytes": 36538 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-2gpu/attempt-5055846/gpu-telemetry-5055846.csv": { + "sha256": "56227a8a9acaf2cc1da6c1ecc40d55276b2c6ccf892f1eb4a217d0a0553b1fd2", + "size_bytes": 108682 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-2gpu/attempt-5055846/launcher.log": { + "sha256": "6e5443e8139eee4e8446fa0cd7be23b99102dc86d565b9712a7f9bb030a2b4f7", + "size_bytes": 150634 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-2gpu/attempt-5055846/phase-memory-5055846.csv": { + "sha256": "04069b47f97de2910f17ff8144e18d3a116fb7e582e60b69645bb7594fccc92e", + "size_bytes": 55153 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-2gpu/attempt-5055846/training-5055846.log": { + "sha256": "b5d9d111a6048ed6842d1fcb68ef5b659e91a000757cc5ddda980044a9e36c48", + "size_bytes": 150457 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-2gpu/attempt-5055846/trial-5055846.json": { + "sha256": "d0d275cde371f33712b8417412596f2473a6f2ff7764dfbc470ab23b5922028b", + "size_bytes": 1499 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-4gpu/attempt-5055846/allocator-traces/allocator-758712.jsonl": { + "sha256": "237f3db3f4ee0502092e87504b3e3ab952c06c68ebf075635123e24b623e7962", + "size_bytes": 1393 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-4gpu/attempt-5055846/allocator-traces/allocator-758713.jsonl": { + "sha256": "6cc10bc0d3dc002353835670fbd48b71c88e4aaf2267ff986c8f3238d63a4bdf", + "size_bytes": 1407 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-4gpu/attempt-5055846/allocator-traces/allocator-758714.jsonl": { + "sha256": "8fc219edfad6eca5ef1694f266a3f19ca3ee0a9ee8207c2774b3db1222ba0bab", + "size_bytes": 1407 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-4gpu/attempt-5055846/allocator-traces/allocator-758715.jsonl": { + "sha256": "0508ff371ee5318190ca65dcf287d1f94d327c08c1c5e01c6981131e3f100738", + "size_bytes": 1406 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-4gpu/attempt-5055846/device-evidence/probe-exit-idle.json": { + "sha256": "6a6e35af2ea340a856e18be694168e8587f9d3d3575bd64400ab88b471d5fc0b", + "size_bytes": 3698 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-4gpu/attempt-5055846/device-evidence/ray-resources.json": { + "sha256": "89b0eef9f11bf2410e16adf138881e9279b8180c8e088e408da7be4cad6cf1ff", + "size_bytes": 1562 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-4gpu/attempt-5055846/device-evidence/trainer-cuda.json": { + "sha256": "5820a2e0f3ba51e642931c885f3f43502ef56b351f7670290646bccad896ad64", + "size_bytes": 865 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-4gpu/attempt-5055846/device-evidence/trainer-exec.json": { + "sha256": "3a1ad878499715670aa9ece0872393e108b3393bc0166b43aaedb4bce24bc005", + "size_bytes": 458 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-4gpu/attempt-5055846/device-evidence/trainer-terminal.json": { + "sha256": "0942a52ac8957c497268d618d7bf358d8b99b6b9c02da9e9e9b1d0892fa4491c", + "size_bytes": 174 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-4gpu/attempt-5055846/device-evidence/worker-actor_rollout-0-758712.json": { + "sha256": "cc22deb10c6cc59eb6172d5f2d697f2d3de0bcd0fbf920a38daf332badc2867a", + "size_bytes": 666 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-4gpu/attempt-5055846/device-evidence/worker-actor_rollout-1-758713.json": { + "sha256": "c01d7f8a2dc010a749c3540d00d98bf1ccf23d2bbc7c36f01f8609d92e382cf8", + "size_bytes": 666 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-4gpu/attempt-5055846/device-evidence/worker-actor_rollout-2-758714.json": { + "sha256": "6248256328c9a8a8d1b17776164bb09e98166448d0020db151013315df63253a", + "size_bytes": 666 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-4gpu/attempt-5055846/device-evidence/worker-actor_rollout-3-758715.json": { + "sha256": "aceeffd2a6ee21ab2fa57b9cd44f1835f7fe94bdc32e87dd8822ecfac355216b", + "size_bytes": 666 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-4gpu/attempt-5055846/dispatch.json": { + "sha256": "2c8d944f9c5385b576c03f6f1f550a313abac8472239f8112e31d0afb33da359", + "size_bytes": 4812 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-4gpu/attempt-5055846/environment-5055846.json": { + "sha256": "d2fd4cb853a77d21cd58983c2063cf4462cd38f1c5ac3c350f31d9ca5d03cbf6", + "size_bytes": 4962 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-4gpu/attempt-5055846/gpu-device-map.json": { + "sha256": "bf42e34a2974029d72c2b7863dae610e3ab7d616434773fd5b16797fea263dd0", + "size_bytes": 821 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-4gpu/attempt-5055846/gpu-memory-5055846.csv": { + "sha256": "b947238f0b063509a2d3848b9c15fbee8ccc080a02e28a1594c5ac1c0f943dfb", + "size_bytes": 149624 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-4gpu/attempt-5055846/gpu-telemetry-5055846.csv": { + "sha256": "e14a9d5bd790a955a984a61fd0f8e818f0040c9bed52327b26eb97287fd1e410", + "size_bytes": 627123 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-4gpu/attempt-5055846/launcher.log": { + "sha256": "99e909eabae1325c5bca0c58a2b35ee1910aed49ef393e80aba79ec187ce63e6", + "size_bytes": 139859 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-4gpu/attempt-5055846/phase-memory-5055846.csv": { + "sha256": "184c366e1cbb8b477d4186ac608cc3b686d934c99f9760083fdebe7f50ae9bba", + "size_bytes": 199718 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-4gpu/attempt-5055846/training-5055846.log": { + "sha256": "c0a06ee1b70eb401ba453451faff7ab6910a9edc73f2db1f8b65191c9f0663e7", + "size_bytes": 139682 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s161-4gpu/attempt-5055846/trial-5055846.json": { + "sha256": "15b9753bd1175cf3f27daf5d24eee202065c409aa189dde3d3c3bb0633fc572a", + "size_bytes": 1499 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-2gpu/attempt-5055848/device-evidence/probe-exit-idle.json": { + "sha256": "ccf86c42e2729d8819e2f489db8c46c4e8dc0e576287c72d240eaa7dd85e9335", + "size_bytes": 3700 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-2gpu/attempt-5055848/device-evidence/ray-resources.json": { + "sha256": "8c5031bb537472738d9d1b2b71d112a5ae91e4998fc59fdfc2f86b40383e2d6b", + "size_bytes": 1569 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-2gpu/attempt-5055848/device-evidence/trainer-cuda.json": { + "sha256": "fd3d3912570b8136ac41be1fcab01e7364cb159775cfe59e82e550900d1fa9fc", + "size_bytes": 585 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-2gpu/attempt-5055848/device-evidence/trainer-exec.json": { + "sha256": "cdef3e6e9d334a02ce252d012e9923a9da930b3c3a45de85da6c6ebe9de64d82", + "size_bytes": 377 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-2gpu/attempt-5055848/device-evidence/trainer-terminal.json": { + "sha256": "b0216e4334a15b5abcf32e4af453caf9f2fd46dfdecb0b767e58bf9e3147c47f", + "size_bytes": 175 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-2gpu/attempt-5055848/device-evidence/worker-actor_rollout-0-3478886.json": { + "sha256": "708edf03ab020a3363dddc33ac3d5282c089be892f034ea0df819674a80c832c", + "size_bytes": 667 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-2gpu/attempt-5055848/device-evidence/worker-actor_rollout-1-3478887.json": { + "sha256": "ce6f403dbb4e71b91278fa3bf5cba8423c1b10e1f5bea1f64d06f379f887ddf9", + "size_bytes": 667 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-2gpu/attempt-5055848/dispatch.json": { + "sha256": "fe5c73ba28b4f32167071aa50e54e73a87642ebec6871e26d93222fe0c5c70c9", + "size_bytes": 4630 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-2gpu/attempt-5055848/environment-5055848.json": { + "sha256": "9924d3cc2a3eb8d5e6fc4c194bf9cb1ba898b84c84e31aa07546adfd4e59a8e3", + "size_bytes": 4480 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-2gpu/attempt-5055848/gpu-device-map.json": { + "sha256": "417db269bf5669e7758705ec1fe0e4aa7fcc5007e50b6c888710040ceb680a25", + "size_bytes": 481 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-2gpu/attempt-5055848/gpu-memory-5055848.csv": { + "sha256": "fa0f3a4c885dab4b89e360a20e48698ba7f71d8d820201ab9fccafef3829ce84", + "size_bytes": 47315 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-2gpu/attempt-5055848/gpu-telemetry-5055848.csv": { + "sha256": "521a962b610439a1ec06f91f32d43b796c0cf4175f1efc2fc7b24cc139952f69", + "size_bytes": 141810 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-2gpu/attempt-5055848/launcher.log": { + "sha256": "0f1caf14cff11a808f3a6aa0c9a631ebf4c1790409b599f5ed42315c35d17e17", + "size_bytes": 153606 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-2gpu/attempt-5055848/phase-memory-5055848.csv": { + "sha256": "0320c4bc9b1f30f162c484ac4af8ba2585a0d2f3f79bea50d0a1daa12b65ecbb", + "size_bytes": 71705 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-2gpu/attempt-5055848/training-5055848.log": { + "sha256": "1909dd87471aed1c70fe69af46d054cca0d240cc7bb9cb0031df948d2ab633e1", + "size_bytes": 153429 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-2gpu/attempt-5055848/trial-5055848.json": { + "sha256": "0945a4ede73353e833330875e9251560a1626701fb3de1ff22edcf08921eec27", + "size_bytes": 1499 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-4gpu/attempt-5055848/allocator-traces/allocator-3495668.jsonl": { + "sha256": "b64368980849bbc3b593e271bcea4cb533a8be5ec4aaa4694a7456a08245e354", + "size_bytes": 1404 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-4gpu/attempt-5055848/allocator-traces/allocator-3495669.jsonl": { + "sha256": "0924639d3c389620652ced866db7eeef372c60fa15e7a105feb445e172477000", + "size_bytes": 1412 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-4gpu/attempt-5055848/allocator-traces/allocator-3495670.jsonl": { + "sha256": "91600e16dbfa0a040a38bd69a9d9629d0a0b2c2c4ace62e66c10b8617c3a8f8b", + "size_bytes": 1412 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-4gpu/attempt-5055848/allocator-traces/allocator-3495671.jsonl": { + "sha256": "5652d1290211e5d919dfe0dc39aeb52de38bb3e23ac047579ec46cf00d62f578", + "size_bytes": 1410 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-4gpu/attempt-5055848/device-evidence/probe-exit-idle.json": { + "sha256": "2a403f722123549ece86ba038d2b641f3762ba8935309a58cf87c4ab11b0bb78", + "size_bytes": 3700 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-4gpu/attempt-5055848/device-evidence/ray-resources.json": { + "sha256": "89426c7f8c5a133bc86254e7a19e30999e2a804fcd9f38e0ebc090f903f8c4d8", + "size_bytes": 1569 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-4gpu/attempt-5055848/device-evidence/trainer-cuda.json": { + "sha256": "96f974c3689f7b68d8462fef004bc52552b944f79029b976ef99cf449346430c", + "size_bytes": 867 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-4gpu/attempt-5055848/device-evidence/trainer-exec.json": { + "sha256": "60d32be87f137fcd38bdb66a3e0d7f771913ca5368b67c8b00b49f40515131ac", + "size_bytes": 459 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-4gpu/attempt-5055848/device-evidence/trainer-terminal.json": { + "sha256": "836d138bc0b2c4873fdd171d83c554745fd3c4e8d119e92ae3579c2b661dbd02", + "size_bytes": 175 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-4gpu/attempt-5055848/device-evidence/worker-actor_rollout-0-3495668.json": { + "sha256": "e3340f3e79ee58d06d3c02aa2eaeeb6f141f600f94efe674543c6be25d00d7a9", + "size_bytes": 667 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-4gpu/attempt-5055848/device-evidence/worker-actor_rollout-1-3495669.json": { + "sha256": "a122b6b0ecea422ebbe985070f714ef261f7901c5e66e0f4e485d47fbac806cd", + "size_bytes": 667 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-4gpu/attempt-5055848/device-evidence/worker-actor_rollout-2-3495670.json": { + "sha256": "ec98aaa1fe011e54210ba5399f40aba9e5c349cd4b0aa31ca8c9dee71e100e28", + "size_bytes": 667 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-4gpu/attempt-5055848/device-evidence/worker-actor_rollout-3-3495671.json": { + "sha256": "24fb3a972eb96e8fd6c860ff7ef8f321db13c45ebc20b3cecc9c6f811685e7ac", + "size_bytes": 667 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-4gpu/attempt-5055848/dispatch.json": { + "sha256": "2264033450431d19eb64390c965c3eaef0f685a1137df6d6cf01235f05d413c3", + "size_bytes": 4812 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-4gpu/attempt-5055848/environment-5055848.json": { + "sha256": "5ab3cd1a603438812d4bd3aa841214fae60a542bfe04f71b277732424dfb299c", + "size_bytes": 4962 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-4gpu/attempt-5055848/gpu-device-map.json": { + "sha256": "440b3a97cb0f3055d6048204ee645ce35c1087c4289b5c0a62bb1dc3a880425d", + "size_bytes": 821 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-4gpu/attempt-5055848/gpu-memory-5055848.csv": { + "sha256": "fcc04ca1930bf1af4dcdfaf07036ac3fe4a33df306c8d04771e5e692e1307667", + "size_bytes": 170153 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-4gpu/attempt-5055848/gpu-telemetry-5055848.csv": { + "sha256": "bf689d971bfacb96d1ce1fe032f4d765b9769850c37299a85baf707a69beb987", + "size_bytes": 719404 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-4gpu/attempt-5055848/launcher.log": { + "sha256": "bbd29c44f2be99a05214921e90ad01ed109fe44cc53099d489a84323e085c757", + "size_bytes": 140056 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-4gpu/attempt-5055848/phase-memory-5055848.csv": { + "sha256": "bdc9512bdc1effe37cdd9123f1abd3f232ff84078f29100358ca833a31639593", + "size_bytes": 227809 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-4gpu/attempt-5055848/training-5055848.log": { + "sha256": "35b26697a53089c099ae97bebccfb9423f8b8e11c309024871493c2b6a498e40", + "size_bytes": 139879 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s162-4gpu/attempt-5055848/trial-5055848.json": { + "sha256": "a2e91ca58a8b45720caad32975802df6cda872d7c2a9dc32cd7b6da923c33652", + "size_bytes": 1499 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-2gpu/attempt-5055860/device-evidence/probe-exit-idle.json": { + "sha256": "5935d8a811f889bfbd1b26d7a7feabe261eec09fefe9cd13f94c3cc6f9462b4b", + "size_bytes": 3700 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-2gpu/attempt-5055860/device-evidence/ray-resources.json": { + "sha256": "e618a8d7fb43bbc293e2676ac6cf879820c7e77b9fec44a4ee785eda9ed19202", + "size_bytes": 1569 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-2gpu/attempt-5055860/device-evidence/trainer-cuda.json": { + "sha256": "804d1d056d7be6be3ceaa92146e7c1d1f40e1069ff60d9f5b3f185908a312581", + "size_bytes": 585 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-2gpu/attempt-5055860/device-evidence/trainer-exec.json": { + "sha256": "375478d43cc8c8ee0901cc31c39d6940b9b622bd41503ef35e32fc759c9e6a56", + "size_bytes": 377 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-2gpu/attempt-5055860/device-evidence/trainer-terminal.json": { + "sha256": "16ad8b9df50bd9a21e57fec73cef32ba3c761bc03f86eccf80ebacd419fae906", + "size_bytes": 175 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-2gpu/attempt-5055860/device-evidence/worker-actor_rollout-0-3597430.json": { + "sha256": "c71c45b092acfaaa2723dba1e03a1b1104eb7d9029ff9ad62b1ac1dfd0d1c63c", + "size_bytes": 667 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-2gpu/attempt-5055860/device-evidence/worker-actor_rollout-1-3597431.json": { + "sha256": "9f8513b16ffaa1c78e6784dc218e32915f64073e4ec67e664680941139d66d77", + "size_bytes": 667 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-2gpu/attempt-5055860/dispatch.json": { + "sha256": "a4ba020d9271c4d43a7c48eb8550fd30f62b25ac90bb1fc196adb9f91a0a03ed", + "size_bytes": 4630 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-2gpu/attempt-5055860/environment-5055860.json": { + "sha256": "32683146c80c4906929ef7fb7ec24a7997a42e276135f6b2ebf47bdee96a1635", + "size_bytes": 4480 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-2gpu/attempt-5055860/gpu-device-map.json": { + "sha256": "89c13a77db7deed706ff42ee258f71d338a71ee4e833e3eb26e23bca8e4f1ce0", + "size_bytes": 481 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-2gpu/attempt-5055860/gpu-memory-5055860.csv": { + "sha256": "feafac6e66d0ef3c280f9ebbeffdb9632b5a6209866659342bf7b341cc146a6b", + "size_bytes": 40276 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-2gpu/attempt-5055860/gpu-telemetry-5055860.csv": { + "sha256": "72c5d3af90ea3fe71d14ee447aaadb23f74e453ca127d6d2cd3e4d4ee5415aa0", + "size_bytes": 120104 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-2gpu/attempt-5055860/launcher.log": { + "sha256": "f7b4b982280434e461b137f779bd9d0a35c86478722397b9d74949984b5ead17", + "size_bytes": 145318 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-2gpu/attempt-5055860/phase-memory-5055860.csv": { + "sha256": "667c1431ddf1898711cdac4b831dad0a16c12a6c19e94f3c9d6016096dfa492f", + "size_bytes": 60871 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-2gpu/attempt-5055860/training-5055860.log": { + "sha256": "895a58c05190eba784ce5c123d9db6157cfd55f6e6cfdafb053273f3de646f33", + "size_bytes": 145141 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-2gpu/attempt-5055860/trial-5055860.json": { + "sha256": "13eaa77fd572f4f7e44c785e6b6d825e0403c341c955301c3d3c63f8ba5f4d76", + "size_bytes": 1499 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-4gpu/attempt-5055860/allocator-traces/allocator-3613579.jsonl": { + "sha256": "e34a2d8cf67428ab9502a4779a60a631d6e15e1bf8415221c950079ac61b94ce", + "size_bytes": 1399 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-4gpu/attempt-5055860/allocator-traces/allocator-3613580.jsonl": { + "sha256": "c4adc669de08f5f8f0e1d0b7c51215743f08c9ee1227ee2321b952dcb5a51da9", + "size_bytes": 1412 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-4gpu/attempt-5055860/allocator-traces/allocator-3613582.jsonl": { + "sha256": "64b58235b0429004b028bcb568cdfeba6880e0b398c9c3fa8ceafbcb28b4e521", + "size_bytes": 1411 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-4gpu/attempt-5055860/allocator-traces/allocator-3613583.jsonl": { + "sha256": "142ca5ca266f061fac4393828f4e685c3ff4c900b65450a918757c94cdadfc4a", + "size_bytes": 1410 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-4gpu/attempt-5055860/device-evidence/probe-exit-idle.json": { + "sha256": "9c2a06024e063e250922387d554b699c05c531ce02383334e78a25707701de3d", + "size_bytes": 3700 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-4gpu/attempt-5055860/device-evidence/ray-resources.json": { + "sha256": "7c354cd31a471afb64c854910f78188c4e7f335286f16223a429887bd689e68b", + "size_bytes": 1569 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-4gpu/attempt-5055860/device-evidence/trainer-cuda.json": { + "sha256": "83cba14c3f5e8b25580ebd1de2163d36209210666b9bedcd593507257dbb036b", + "size_bytes": 867 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-4gpu/attempt-5055860/device-evidence/trainer-exec.json": { + "sha256": "460457015084960e6062416a95b3b22e85173c73c22e7595364389968f0aba47", + "size_bytes": 459 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-4gpu/attempt-5055860/device-evidence/trainer-terminal.json": { + "sha256": "301f324b55a15de2ff50e14906c4b217e027e0e9fe5abd2ec77c068d5e92143c", + "size_bytes": 175 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-4gpu/attempt-5055860/device-evidence/worker-actor_rollout-0-3613579.json": { + "sha256": "6d30ce79da2c60a1144f105f00ada4263a00165bc9cc95d1233383ec34bbd56f", + "size_bytes": 667 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-4gpu/attempt-5055860/device-evidence/worker-actor_rollout-1-3613580.json": { + "sha256": "b738a4d93f6791a05bced4f7ed5a068d101fd6a238fef3c0d7944c8bee0a6b9d", + "size_bytes": 667 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-4gpu/attempt-5055860/device-evidence/worker-actor_rollout-2-3613582.json": { + "sha256": "928f5a1569f9e4cb6cbd8bba290091aff2730627d21ab144cfaf73d332692577", + "size_bytes": 667 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-4gpu/attempt-5055860/device-evidence/worker-actor_rollout-3-3613583.json": { + "sha256": "4b0dd9ce9a027a121768cc1debb550915007d0dfb7e6df59d94c2ed460fdb7d2", + "size_bytes": 667 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-4gpu/attempt-5055860/dispatch.json": { + "sha256": "b068e897364aa48576d7ea3df3d5026ee6f9fa3991be8cea01af6067df2bd05c", + "size_bytes": 4812 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-4gpu/attempt-5055860/environment-5055860.json": { + "sha256": "eacf127f7427362427f574e136c9723f53d1d64a0d359fa8e00cdd1badbc699e", + "size_bytes": 4962 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-4gpu/attempt-5055860/gpu-device-map.json": { + "sha256": "bbb64b1c5850fb05aad5f82ff05f75460d575c41ba7117b1a7725a957a777f5f", + "size_bytes": 821 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-4gpu/attempt-5055860/gpu-memory-5055860.csv": { + "sha256": "5d529432ba8ccec094d1573677cc486f25c4ac8e62056605f4a6443d921453de", + "size_bytes": 164699 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-4gpu/attempt-5055860/gpu-telemetry-5055860.csv": { + "sha256": "57283f2d1e6b11a13b5f492e83e527b02035d138a4af3585e09ca16b2575451a", + "size_bytes": 693899 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-4gpu/attempt-5055860/launcher.log": { + "sha256": "bf556f67fd0ceec92ee6b57e4513b2231de377f2bcba262949c1b9974bb88263", + "size_bytes": 140721 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-4gpu/attempt-5055860/phase-memory-5055860.csv": { + "sha256": "6e566f59f552b71135edbefd96d5ef01659bea3ca67cd4c9eb5905cee4a9b9c5", + "size_bytes": 220077 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-4gpu/attempt-5055860/training-5055860.log": { + "sha256": "fe75b20444c02557e12dca7f32c471b373c3956008c30024ccf9c71cddee8b24", + "size_bytes": 140544 + }, + "output/estimation_host_capacity/host-qwen25-7b-code_heavy_tail-c2-s163-4gpu/attempt-5055860/trial-5055860.json": { + "sha256": "90e351c33f99b8a588cd62e7ea4ae13bd305e461cdced4ffeee4e902e0037181", + "size_bytes": 1499 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-2gpu/attempt-5055845/device-evidence/probe-exit-idle.json": { + "sha256": "b777aa76bec88b8be516c9ee3075027f11a74df773673d1ad11f2b833227ab48", + "size_bytes": 3690 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-2gpu/attempt-5055845/device-evidence/ray-resources.json": { + "sha256": "8c9b1d208ed79b24efa930019ed26ad8f52d4eb4f336966f282ebe90c3f9d444", + "size_bytes": 1559 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-2gpu/attempt-5055845/device-evidence/trainer-cuda.json": { + "sha256": "2df904fc2102b1d25f7a00c8bc43e94492e004db5c7005848d88ecc04a96e683", + "size_bytes": 575 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-2gpu/attempt-5055845/device-evidence/trainer-exec.json": { + "sha256": "c3dc088929144a56e4bd35142c40dc42b438a9b2117904a0ed37f35f045ae6f3", + "size_bytes": 367 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-2gpu/attempt-5055845/device-evidence/trainer-terminal.json": { + "sha256": "428879eb9634dcbb7b5db27b597be7554460792cfd8705ca95c0e0b247bf1c9f", + "size_bytes": 165 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-2gpu/attempt-5055845/device-evidence/worker-actor_rollout-0-3641588.json": { + "sha256": "44cfbe719dc0d8fbdfaec152fe7815fccf976031ed87963aa12b1f03d3960559", + "size_bytes": 657 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-2gpu/attempt-5055845/device-evidence/worker-actor_rollout-1-3641589.json": { + "sha256": "04ec45b04a2163ebb1dd4458d7a4193b836b1507437e7ef101e97aac5640649a", + "size_bytes": 657 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-2gpu/attempt-5055845/dispatch.json": { + "sha256": "7eaf33acef90af6767f8ac2e19d76ae5b231da5e209f220f8e9f0716f302d65b", + "size_bytes": 4330 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-2gpu/attempt-5055845/environment-5055845.json": { + "sha256": "80525e90b0e1f649dab1e900e8c9145b72f9db50dc9020d4ec77fb7328dcf17e", + "size_bytes": 4418 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-2gpu/attempt-5055845/gpu-device-map.json": { + "sha256": "4df7c4857a0b18f9e095731bb78d732602c0ae8f99dbd5dc1d1c007a500f1182", + "size_bytes": 471 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-2gpu/attempt-5055845/gpu-memory-5055845.csv": { + "sha256": "d1f5bedf4ee67d9187d15ead047a69ef08a0c948f3363ee0b5b6d61d67ab4d52", + "size_bytes": 35477 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-2gpu/attempt-5055845/gpu-telemetry-5055845.csv": { + "sha256": "b0830bd5d7380bb2e45152ed8ef0ebace01554f332df6e6eada802b1417691a9", + "size_bytes": 105547 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-2gpu/attempt-5055845/launcher.log": { + "sha256": "ff20b241eeb9e8d1f93336d7795227f06cf8d749656c890dd441dedc9daac285", + "size_bytes": 152006 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-2gpu/attempt-5055845/phase-memory-5055845.csv": { + "sha256": "879e91221dd0e3e94f57fba6f9e7052ef57ad7584dfc3fddc9451230660c4290", + "size_bytes": 53552 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-2gpu/attempt-5055845/training-5055845.log": { + "sha256": "6ce908d1cc572e4527c6b03a03c677fe01dced3282eea870ddf5cab610ed0594", + "size_bytes": 151839 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-2gpu/attempt-5055845/trial-5055845.json": { + "sha256": "b0f9305d41f55be4970dac896c4f79fc857c6aaf5b4a561cfb21e72ef4f3ebfa", + "size_bytes": 1431 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-4gpu/attempt-5055845/allocator-traces/allocator-3657525.jsonl": { + "sha256": "5f720a58f4c5aeb0be7c5d8df8b7f58dd073512c4265381d7c7637264d55f4f8", + "size_bytes": 1406 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-4gpu/attempt-5055845/allocator-traces/allocator-3657526.jsonl": { + "sha256": "d9f4e9e74c0e54d084ea865c3591305105762a7ea11ad8636cee5da9d8e0bf36", + "size_bytes": 1397 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-4gpu/attempt-5055845/allocator-traces/allocator-3657527.jsonl": { + "sha256": "74e6dee466e2190aa9e4cdf94fc215eca3341434bc880bc1ee6f4fedc4cef4cb", + "size_bytes": 1404 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-4gpu/attempt-5055845/allocator-traces/allocator-3657528.jsonl": { + "sha256": "7eae2e4ac22b1670c758a27db650636dbb70f95285a3ea2276c42f38a2f9c2f2", + "size_bytes": 1409 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-4gpu/attempt-5055845/device-evidence/probe-exit-idle.json": { + "sha256": "21a65252000a7efe1b2a3148f273f0a21c22ddde8e238118407ca3f182922c9a", + "size_bytes": 3690 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-4gpu/attempt-5055845/device-evidence/ray-resources.json": { + "sha256": "854a95b6416672e6064f0e400a7d80221351e08b020e3339edf08a61f42d6abb", + "size_bytes": 1559 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-4gpu/attempt-5055845/device-evidence/trainer-cuda.json": { + "sha256": "31c9aa6cda5d406866faeb7df88bc89cf0d5dd51a7a336b42603bdf144507a01", + "size_bytes": 857 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-4gpu/attempt-5055845/device-evidence/trainer-exec.json": { + "sha256": "afbced505a2027444a52dd02d87ba99a42f1a0c06ed7409b9b17e07ca76bb274", + "size_bytes": 449 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-4gpu/attempt-5055845/device-evidence/trainer-terminal.json": { + "sha256": "7367a540c02e6f0e4434e0c6cb80e0f5323eec893470d2b76a4256a4ed0bf0b0", + "size_bytes": 165 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-4gpu/attempt-5055845/device-evidence/worker-actor_rollout-0-3657525.json": { + "sha256": "c75c55ad82f61af078822825c919fa5bfb1f8bcd92a567cba268e98087379bcc", + "size_bytes": 657 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-4gpu/attempt-5055845/device-evidence/worker-actor_rollout-1-3657526.json": { + "sha256": "4febd2f4e518cbd006b32db932ca3eec7f49ce1c21679939e39d4d2b9dfe9b9e", + "size_bytes": 657 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-4gpu/attempt-5055845/device-evidence/worker-actor_rollout-2-3657527.json": { + "sha256": "c87b0c9f3993ac0410fd3e21f42e62a4e2d0d730fb27024974f80089eef1b2ee", + "size_bytes": 657 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-4gpu/attempt-5055845/device-evidence/worker-actor_rollout-3-3657528.json": { + "sha256": "54ce75a2b95c43338bf43d395386b6531340e7960a14c07b1a4cf24da016bbbc", + "size_bytes": 657 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-4gpu/attempt-5055845/dispatch.json": { + "sha256": "fe1cc5a4f214e0b9dea73b9981f85f347e9547ac335f206e56d01525fcabad6b", + "size_bytes": 4512 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-4gpu/attempt-5055845/environment-5055845.json": { + "sha256": "8805e099802d7ca91e054aaf89e60ddbd6b1d86c503e22f912316c679859a34c", + "size_bytes": 4900 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-4gpu/attempt-5055845/gpu-device-map.json": { + "sha256": "bc42ded93e9417f9128d5a4af520f71b6e5f21f122321aa9cb4217860b317f9f", + "size_bytes": 811 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-4gpu/attempt-5055845/gpu-memory-5055845.csv": { + "sha256": "85d9151174af04a750b998c837255ef53bacf34cbf0fb1b6a2d4a794b49d8b92", + "size_bytes": 182114 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-4gpu/attempt-5055845/gpu-telemetry-5055845.csv": { + "sha256": "de50f4cdb6122f4fe399a8a5aa5cbb7883471dc9c79795ebd87f5ee8f65b735a", + "size_bytes": 766634 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-4gpu/attempt-5055845/launcher.log": { + "sha256": "c4fe24195da0af548b6059745ed893366918275c0fa865ac54c6f79faea193fd", + "size_bytes": 140327 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-4gpu/attempt-5055845/phase-memory-5055845.csv": { + "sha256": "0290361e6a2cb16cd0f959ab2c2c1dfca819f6a53f94a1e0151494a5bbaceca3", + "size_bytes": 242905 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-4gpu/attempt-5055845/training-5055845.log": { + "sha256": "b2cb503abf3e0d17245ddeda9c1cec45ffae1179113362f6137376c36638862a", + "size_bytes": 140160 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s161-4gpu/attempt-5055845/trial-5055845.json": { + "sha256": "37849cf5979192992d63ed3fce8f8f9bb5855810a530b1034955caaf1a1e9315", + "size_bytes": 1431 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-2gpu/attempt-5055858/device-evidence/probe-exit-idle.json": { + "sha256": "e802ca892d4dfc28d9bae6008e38fc731b2042c4bbbf5facf949ecd8d01a0f57", + "size_bytes": 3688 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-2gpu/attempt-5055858/device-evidence/ray-resources.json": { + "sha256": "e4b3dc8fc8cf84958f6f2f14ced8c61fff40d8d3e10c3233372e5352a23d15b8", + "size_bytes": 1552 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-2gpu/attempt-5055858/device-evidence/trainer-cuda.json": { + "sha256": "05f7055153eb05c2e119c135040d10e310afe80b23be322cb2574496e1bc9266", + "size_bytes": 573 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-2gpu/attempt-5055858/device-evidence/trainer-exec.json": { + "sha256": "ad0535cfee8cfde17b052696331ee58c6e033aca38946cd72949be2dfbf4f418", + "size_bytes": 366 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-2gpu/attempt-5055858/device-evidence/trainer-terminal.json": { + "sha256": "9bf35637417621fbe8fb9af4c54df54c534991c428fe70d0ff4d46ea1f845e71", + "size_bytes": 164 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-2gpu/attempt-5055858/device-evidence/worker-actor_rollout-0-827887.json": { + "sha256": "9dfc40055ea34958345a55afe90660aa1a325d6b941a71bf720351fb8fcf074a", + "size_bytes": 656 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-2gpu/attempt-5055858/device-evidence/worker-actor_rollout-1-827894.json": { + "sha256": "2c5fbf758d4c422c92188fbf613d5ba18f3671ab827cb1016ad30e9502ba01e7", + "size_bytes": 656 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-2gpu/attempt-5055858/dispatch.json": { + "sha256": "1269d5a582067af091b32cc2af4d36db9c36791a06836fecc78f5891b91dc807", + "size_bytes": 4330 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-2gpu/attempt-5055858/environment-5055858.json": { + "sha256": "6cffbbed13a57719bd8495bcc3cb66c9519227682e2e4bf0a2488497dcd60e44", + "size_bytes": 4418 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-2gpu/attempt-5055858/gpu-device-map.json": { + "sha256": "00546dc5355adfee836479837ad15809afda43feee21f4d13d633fb037f80541", + "size_bytes": 471 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-2gpu/attempt-5055858/gpu-memory-5055858.csv": { + "sha256": "60c0aaab71a9802d32b7a0e3532b4fa65e448129bcd3f4e8c48c6cf68e20ec13", + "size_bytes": 41747 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-2gpu/attempt-5055858/gpu-telemetry-5055858.csv": { + "sha256": "cba80250d271a8153249b8a3a458613eed44adc3403a93b3c334648767802d05", + "size_bytes": 124693 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-2gpu/attempt-5055858/launcher.log": { + "sha256": "b97b985f1fb19852d33562e9b2f5c469ef7dedf70499afabf826834cbfc0e67e", + "size_bytes": 150723 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-2gpu/attempt-5055858/phase-memory-5055858.csv": { + "sha256": "f3bc8669f0bb261d8f29dd1e978ba9b58272b20c43ebd7b10389959027778771", + "size_bytes": 63152 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-2gpu/attempt-5055858/training-5055858.log": { + "sha256": "0a6455c49cd49c8533ca30f7c39e6b5eaceb57aadbf308fdc2efe5ad8cd3f928", + "size_bytes": 150556 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-2gpu/attempt-5055858/trial-5055858.json": { + "sha256": "1b6208ddd37a043883baa8e4aa762cfe1119e4d82e03b6754e7a1491ae47bc55", + "size_bytes": 1431 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-4gpu/attempt-5055858/allocator-traces/allocator-802123.jsonl": { + "sha256": "f90571cc23c7d711a020d93027779f2747728ef208968d110eeefd8c69f0957f", + "size_bytes": 1398 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-4gpu/attempt-5055858/allocator-traces/allocator-802124.jsonl": { + "sha256": "08f1ee26e4509d3cb8f81f74e7bdf7eebc3a2fa9604b65d5bcd0c12f6df3bcce", + "size_bytes": 1404 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-4gpu/attempt-5055858/allocator-traces/allocator-802125.jsonl": { + "sha256": "27fce5cdc6d40e04cf28b8a9588417fa4d31d71d25a0f6448e2fde36dff24059", + "size_bytes": 1404 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-4gpu/attempt-5055858/allocator-traces/allocator-802126.jsonl": { + "sha256": "c433f20090d90f9a1e1b48cf608830a712fbd4d4b2b2abbc259096e8fa8a0b24", + "size_bytes": 1405 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-4gpu/attempt-5055858/device-evidence/probe-exit-idle.json": { + "sha256": "a758c491b843983e36475e411900d02e8cfe90cd43e70454db802d7f0970998b", + "size_bytes": 3688 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-4gpu/attempt-5055858/device-evidence/ray-resources.json": { + "sha256": "c377cfc3411bac36b667c6e7a6cd3911a489adb3dbaed2140ab241c00f59624a", + "size_bytes": 1552 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-4gpu/attempt-5055858/device-evidence/trainer-cuda.json": { + "sha256": "3e7ceebf4bc857dbe8d997addf297a242d5929997cddb3b954a39edbae19e8d7", + "size_bytes": 855 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-4gpu/attempt-5055858/device-evidence/trainer-exec.json": { + "sha256": "4dd27278c99d1db3736774d9dd258eb90aab8dceb4bcbb8e64fef8068ff99994", + "size_bytes": 448 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-4gpu/attempt-5055858/device-evidence/trainer-terminal.json": { + "sha256": "10fa0454f1463af1a30b115bdf45620cdc3045447f28c215f6070fe69ab886c1", + "size_bytes": 164 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-4gpu/attempt-5055858/device-evidence/worker-actor_rollout-0-802123.json": { + "sha256": "3216c1337dfffe6313abc84ea51abd8a9e396dc177a7e8dfa41eaabf6c209aa4", + "size_bytes": 656 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-4gpu/attempt-5055858/device-evidence/worker-actor_rollout-1-802124.json": { + "sha256": "e950e452136d75ba43a1ea45176cec9a1e10d1346bc1285a66da86cd500f8a6a", + "size_bytes": 656 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-4gpu/attempt-5055858/device-evidence/worker-actor_rollout-2-802125.json": { + "sha256": "94e4eff45487753971d2277b257b09871006e7593a43e1deadee3661c9ef9ff2", + "size_bytes": 656 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-4gpu/attempt-5055858/device-evidence/worker-actor_rollout-3-802126.json": { + "sha256": "8a94205f733595bb4f894a4beaab8f90801d2f7c75c00b2bafb95e38a2fab40f", + "size_bytes": 656 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-4gpu/attempt-5055858/dispatch.json": { + "sha256": "91e269ac53b2d3a0f95dc830478c372c2069ab150b1fe7b02d89e0b5aa3d474d", + "size_bytes": 4512 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-4gpu/attempt-5055858/environment-5055858.json": { + "sha256": "4779aa01665897ec82b1f6204f6225dda322e33ece9619aa27173ac37676e468", + "size_bytes": 4900 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-4gpu/attempt-5055858/gpu-device-map.json": { + "sha256": "b45a4d85a0d89f2226c0dff1c6697ca3d6bb444f224d61466086ae6b697ff166", + "size_bytes": 811 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-4gpu/attempt-5055858/gpu-memory-5055858.csv": { + "sha256": "5e449ad227dc376eefe37833e1e88fc6ac94796fcf7a1736fbf9f77ac1b23582", + "size_bytes": 137508 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-4gpu/attempt-5055858/gpu-telemetry-5055858.csv": { + "sha256": "a6afbe28a1df8aa2cf2c2dba25d9538ca40dc8f133d538945f4eaca1df183187", + "size_bytes": 576417 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-4gpu/attempt-5055858/launcher.log": { + "sha256": "909b6cd621ed6bf0077c3801ec09de4be2008c9fa280c73ca4fb19d2146a0e7e", + "size_bytes": 138551 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-4gpu/attempt-5055858/phase-memory-5055858.csv": { + "sha256": "aa9a9130f88ca1e39853dfd4877a8970aba8409051053638022a6809c11262f1", + "size_bytes": 183608 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-4gpu/attempt-5055858/training-5055858.log": { + "sha256": "abfbc06c66ce14ee4c8b3b100e3cf9335169a8e3162c8a87723b3a8f27b71b24", + "size_bytes": 138384 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s162-4gpu/attempt-5055858/trial-5055858.json": { + "sha256": "b46f64d2b6062b9f24d82b69f276a9a5471cd41288ac15cdf6ec218574844047", + "size_bytes": 1431 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-2gpu/attempt-5055847/device-evidence/probe-exit-idle.json": { + "sha256": "b415aa828c58dfc309ffebf6c0a6575b26b9b2430b8a54b14d34ae71a905c034", + "size_bytes": 3690 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-2gpu/attempt-5055847/device-evidence/ray-resources.json": { + "sha256": "516789ba306e38fb53f8b5458825519366e43caeb67cb17f89129d5889c3494d", + "size_bytes": 1559 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-2gpu/attempt-5055847/device-evidence/trainer-cuda.json": { + "sha256": "18200e287843a74c96abea10e8b2300f586c49a18a45cea1eca423e63040e09e", + "size_bytes": 575 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-2gpu/attempt-5055847/device-evidence/trainer-exec.json": { + "sha256": "c175ad2bb51f042c4bfdd0050a3f6aed6bf30e9c0e2e2c86ac7f2d261fdbb5b4", + "size_bytes": 367 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-2gpu/attempt-5055847/device-evidence/trainer-terminal.json": { + "sha256": "266051af79e99249a9b6922f26b8a07705a9ec3dbd1caafc5407e7a7e9b0cf35", + "size_bytes": 165 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-2gpu/attempt-5055847/device-evidence/worker-actor_rollout-0-3547555.json": { + "sha256": "37a0d7a8ae322ffcb1987b546ced6af852943a086b26241b429d7ea293a81fb9", + "size_bytes": 657 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-2gpu/attempt-5055847/device-evidence/worker-actor_rollout-1-3547556.json": { + "sha256": "e2afbf56454eff27ad2dd52bc76069436dfcb0c62c616cc7dc9f7a399dde17a2", + "size_bytes": 657 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-2gpu/attempt-5055847/dispatch.json": { + "sha256": "bcb7f7078254f8d59cbed3a4997aa844489fdc5d4e10ce7743b557c99c3a5131", + "size_bytes": 4330 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-2gpu/attempt-5055847/environment-5055847.json": { + "sha256": "18c21604af831b1b1665813cef754edb93cfa7b2f2dfc61262b5672289437c09", + "size_bytes": 4418 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-2gpu/attempt-5055847/gpu-device-map.json": { + "sha256": "7acec01171399fbfd415fbb9fdfefdf75e9000e882ef74d7ada1839b0cb17027", + "size_bytes": 471 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-2gpu/attempt-5055847/gpu-memory-5055847.csv": { + "sha256": "19c9c00a2042cf00b838944ab9d53541b3e3f6ab830847616506baf642082cd1", + "size_bytes": 44431 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-2gpu/attempt-5055847/gpu-telemetry-5055847.csv": { + "sha256": "7d894cc0dfab7932636f2f5bda81ccc26d7ee58ab832a79d2e3a128594d2a781", + "size_bytes": 133007 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-2gpu/attempt-5055847/launcher.log": { + "sha256": "60f634390988b81e710a04d37eb3d7a1705fa70ef466d420fd45e05ddcf650d0", + "size_bytes": 152445 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-2gpu/attempt-5055847/phase-memory-5055847.csv": { + "sha256": "c25ad1b9a33e6d28afffea147ef8e2f8d944961b9e99d957c22c14414070f940", + "size_bytes": 67291 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-2gpu/attempt-5055847/training-5055847.log": { + "sha256": "c9cb86baea65985262d7c08dee415bd705bbb657577b3092168dc7670d59cfa3", + "size_bytes": 152278 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-2gpu/attempt-5055847/trial-5055847.json": { + "sha256": "57dc50de8b0b2bc365e966297638655eafca869a4fcef5be84a371c2cc61222e", + "size_bytes": 1431 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-4gpu/attempt-5055847/allocator-traces/allocator-3563810.jsonl": { + "sha256": "c345803b85bb7b1007edb566d512d0d23835bcf1f86fb2f6d967a06fad35092e", + "size_bytes": 1398 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-4gpu/attempt-5055847/allocator-traces/allocator-3563811.jsonl": { + "sha256": "c8e0a83607e5769079bb984a5e91b1cf6a7318f163b3ad117448be5d19dbb7cb", + "size_bytes": 1410 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-4gpu/attempt-5055847/allocator-traces/allocator-3563812.jsonl": { + "sha256": "fdba6f8fffc9c3481e2941ee05b560499a45d672b2a7353ff5c175c7fcb22a91", + "size_bytes": 1409 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-4gpu/attempt-5055847/allocator-traces/allocator-3563813.jsonl": { + "sha256": "3bf0e61e4decb549210e8ef34dfe68ed815b3ea8b47683a4b2396ac4b00adffe", + "size_bytes": 1410 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-4gpu/attempt-5055847/device-evidence/probe-exit-idle.json": { + "sha256": "a0e794eb7188da3859f2158ddf4a03f049d136ee18decaa2a8432c8e3c34a07d", + "size_bytes": 3690 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-4gpu/attempt-5055847/device-evidence/ray-resources.json": { + "sha256": "5fa6bb9d95647cde35d976f59c86ad0ce9de4062858e5d63cb5c9e02f6a54b0e", + "size_bytes": 1559 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-4gpu/attempt-5055847/device-evidence/trainer-cuda.json": { + "sha256": "2b52867b5f81972aadc25d5cfefa86a9270966cad52d841eb1f90fde316de843", + "size_bytes": 857 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-4gpu/attempt-5055847/device-evidence/trainer-exec.json": { + "sha256": "03811e669d3623195c23a7759b50accdb05808c736506e9303b71618cf85797c", + "size_bytes": 449 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-4gpu/attempt-5055847/device-evidence/trainer-terminal.json": { + "sha256": "966e1433ea458a6ac9e78fedd703da8fd444e549bebf9356762d0c7af177a95c", + "size_bytes": 165 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-4gpu/attempt-5055847/device-evidence/worker-actor_rollout-0-3563810.json": { + "sha256": "462c3601f438c62a5e9210151a1717bb18fa4816d620ca1f747621645f661fb9", + "size_bytes": 657 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-4gpu/attempt-5055847/device-evidence/worker-actor_rollout-1-3563811.json": { + "sha256": "72fe6f42647a258775d00fd48052963ca9a55be891880970ec9c6c61f5c7b401", + "size_bytes": 657 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-4gpu/attempt-5055847/device-evidence/worker-actor_rollout-2-3563812.json": { + "sha256": "56e43743ba141f80cbc576203b785c4154910e176ff2a96f81255f297ccfc7be", + "size_bytes": 657 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-4gpu/attempt-5055847/device-evidence/worker-actor_rollout-3-3563813.json": { + "sha256": "389efebd3a4ea66a3d2faf5e9ea46a6b9cb3b9124e347da23d37835e03bf515c", + "size_bytes": 657 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-4gpu/attempt-5055847/dispatch.json": { + "sha256": "f1dd993d2f3f8d9dbb6bf2986bd0b3f097b0a744ce84d70ddbbe8cd3b5b2bdb2", + "size_bytes": 4512 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-4gpu/attempt-5055847/environment-5055847.json": { + "sha256": "4e03274b2a7fa9234085017249c2822d5480a06ffa085209fdc1a6aa2fa5b162", + "size_bytes": 4900 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-4gpu/attempt-5055847/gpu-device-map.json": { + "sha256": "91f43ee075a1abe7320de6c73f9a96fed3b69c8086db6abf068f2fe3f9211e73", + "size_bytes": 811 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-4gpu/attempt-5055847/gpu-memory-5055847.csv": { + "sha256": "926af37d182966cf3d826acb08f698ba9b4137c284709b42357450dba4e3ebc3", + "size_bytes": 224585 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-4gpu/attempt-5055847/gpu-telemetry-5055847.csv": { + "sha256": "3db078630b08f1cf337a3b279c4df79369f076e25b087aef032a74bb52cb7f9b", + "size_bytes": 945675 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-4gpu/attempt-5055847/launcher.log": { + "sha256": "0909dc5b195fd2a5d991f433505f7dffcab889c0ecdc5668a10aa703374d0ead", + "size_bytes": 141166 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-4gpu/attempt-5055847/phase-memory-5055847.csv": { + "sha256": "ba545bca34e70cfa866919860dd7ef2d2da7f09ef3b23dac6846f39e6c5ecc20", + "size_bytes": 299506 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-4gpu/attempt-5055847/training-5055847.log": { + "sha256": "a4a40b83284c8b0561cd1a1d6de4298b37c701f1248239535f02b795b8991d19", + "size_bytes": 140999 + }, + "output/estimation_host_capacity/host-qwen25-7b-gsm8k-c2-s163-4gpu/attempt-5055847/trial-5055847.json": { + "sha256": "e23b2997b7a855ef1601efd231f694876dd6fa9c85d36128e5cd90fcda1910b8", + "size_bytes": 1431 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-2gpu/attempt-5055864/device-evidence/probe-exit-idle.json": { + "sha256": "c085339a3bb6a8423a99a4b5e6f5a59ce53a571552d8b560201aef9e8a34118e", + "size_bytes": 3689 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-2gpu/attempt-5055864/device-evidence/ray-resources.json": { + "sha256": "e67148eb3d351b840e83b84957ee13d79ed32b06b0d4fd4aa828389d72024720", + "size_bytes": 1558 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-2gpu/attempt-5055864/device-evidence/trainer-cuda.json": { + "sha256": "397a8a6aef1a20331f9c25cb502516a5d1880a30bc64eeacffb87e4262a59c30", + "size_bytes": 574 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-2gpu/attempt-5055864/device-evidence/trainer-exec.json": { + "sha256": "97dc64e38e4ce043e80985bc33e6072a3ff0d16c47a31bb7380988d8ff77909a", + "size_bytes": 366 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-2gpu/attempt-5055864/device-evidence/trainer-terminal.json": { + "sha256": "fc15df3659f0e15ff682ea80a05d6031773c1e8a9a4fa56e5066ffbb8c617b42", + "size_bytes": 164 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-2gpu/attempt-5055864/device-evidence/worker-actor_rollout-0-3597684.json": { + "sha256": "93b2b62213c276bb87f97433b34bd7a23e008f066169379a41252a7f6be34731", + "size_bytes": 656 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-2gpu/attempt-5055864/device-evidence/worker-actor_rollout-1-3597685.json": { + "sha256": "53d6a9bc99fb89ce017e31ef0fa0491435f2505586fe74333126dfb2ff44f615", + "size_bytes": 656 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-2gpu/attempt-5055864/dispatch.json": { + "sha256": "f645c1a9516d69a0b8bc7e061b88369be40fa544c537a4f0e3c36677e467d1f1", + "size_bytes": 4318 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-2gpu/attempt-5055864/environment-5055864.json": { + "sha256": "e912b4eddc048e6c275e84b8baab706df188c3263a7ceccfb1454d39fcf1a6fa", + "size_bytes": 4414 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-2gpu/attempt-5055864/gpu-device-map.json": { + "sha256": "82b5db5ed4b8f850e2d60268f09937819d3516adc2bda514dcf28fd30721aae1", + "size_bytes": 470 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-2gpu/attempt-5055864/gpu-memory-5055864.csv": { + "sha256": "4be4879326e0b303927533f28f746e706e7a0ab7f90a5636b00170f190c3b2c7", + "size_bytes": 39563 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-2gpu/attempt-5055864/gpu-telemetry-5055864.csv": { + "sha256": "2003ff9dac6e9a4d9f37ca439a2c2198cec058af530a713492eb1d90f78b557e", + "size_bytes": 117517 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-2gpu/attempt-5055864/launcher.log": { + "sha256": "585dc88e6962db949729714f29234a8ac354c435fbec2d116a33cdd097b1ee89", + "size_bytes": 151969 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-2gpu/attempt-5055864/phase-memory-5055864.csv": { + "sha256": "d2d6ece196ec15e92bd7f0505a3dad2a67f3f96a7b35262b2e88496ded4645dc", + "size_bytes": 59678 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-2gpu/attempt-5055864/training-5055864.log": { + "sha256": "80cdb67c540fa19967d9c5627fa0d8b33c024cbcb7ae23d7e58d4535d01d8053", + "size_bytes": 151803 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-2gpu/attempt-5055864/trial-5055864.json": { + "sha256": "3942ac0b5bc951c9a9c22b6ab7145ebc2d241c5e78ac112e77cec363bf74d3bd", + "size_bytes": 1424 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-4gpu/attempt-5055864/allocator-traces/allocator-3571156.jsonl": { + "sha256": "ad805b8eee80dc4b7b13169579f06e5f8bfc9e265517a177c7d06b014f412882", + "size_bytes": 1404 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-4gpu/attempt-5055864/allocator-traces/allocator-3571157.jsonl": { + "sha256": "d9f68dccb3b9f936695f18f5509fe99cb05599bc52d69430ef55b364d197e09b", + "size_bytes": 1409 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-4gpu/attempt-5055864/allocator-traces/allocator-3571158.jsonl": { + "sha256": "9fecba1ca02906a161e139724431b292134a52c547391a5e8b9007783de9e82d", + "size_bytes": 1412 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-4gpu/attempt-5055864/allocator-traces/allocator-3571159.jsonl": { + "sha256": "c22b43989ec3bca8f6f62f8ff48f0af654b475b96a20dfc1f680cb560b19103b", + "size_bytes": 1410 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-4gpu/attempt-5055864/device-evidence/probe-exit-idle.json": { + "sha256": "ce04ee6f11c8a9c4d83a8b1f9b6df352f0c91da33aea814be6dfcda32862687e", + "size_bytes": 3689 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-4gpu/attempt-5055864/device-evidence/ray-resources.json": { + "sha256": "a6f1fa5a7ed1b81891e799d09ee8f6c6ea31bdcc70d738325d62eb0c429ced9b", + "size_bytes": 1558 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-4gpu/attempt-5055864/device-evidence/trainer-cuda.json": { + "sha256": "9c07c2184bc56597d8e58dca03bbd42e5a658708e8d17874110d586accbe4b58", + "size_bytes": 856 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-4gpu/attempt-5055864/device-evidence/trainer-exec.json": { + "sha256": "3cd6eae16780f0879274834b811f948d14ab0406a0b74aaca81ede9867cb321e", + "size_bytes": 448 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-4gpu/attempt-5055864/device-evidence/trainer-terminal.json": { + "sha256": "16cc70afa31fc1247c745fb6499bf991b4e70100444dce8044a023f54942b1bd", + "size_bytes": 164 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-4gpu/attempt-5055864/device-evidence/worker-actor_rollout-0-3571156.json": { + "sha256": "2a8dffb74119c328f1defc94f17137189c611818e08e91bcf2e2c33eb9565fb3", + "size_bytes": 656 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-4gpu/attempt-5055864/device-evidence/worker-actor_rollout-1-3571157.json": { + "sha256": "46e3f08ea23076bd5c8f67b3ca990fabce9a618792edaf7b009a3345d699982b", + "size_bytes": 656 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-4gpu/attempt-5055864/device-evidence/worker-actor_rollout-2-3571158.json": { + "sha256": "3bc5ec9404104f79c07ee1acbb6981a586cb3be23d30119fd48ebfa28992818a", + "size_bytes": 656 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-4gpu/attempt-5055864/device-evidence/worker-actor_rollout-3-3571159.json": { + "sha256": "fb53ac6637803eb628fcab6e08ad88f5835b2af804fe875099aa3a711a24dc4a", + "size_bytes": 656 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-4gpu/attempt-5055864/dispatch.json": { + "sha256": "538e8b0a98691fa3dd457ea212197fad1eb197a311f8a40159e1cf7d7f64b205", + "size_bytes": 4500 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-4gpu/attempt-5055864/environment-5055864.json": { + "sha256": "9fa0bc5986fc09193bbd8905f9d8386d73ac91fadf0369e4ca4f5011b772f543", + "size_bytes": 4896 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-4gpu/attempt-5055864/gpu-device-map.json": { + "sha256": "10673761c821963f041f7a7a30f4c4387d6d29cedf5214d8152663c47a43882d", + "size_bytes": 810 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-4gpu/attempt-5055864/gpu-memory-5055864.csv": { + "sha256": "91060ef947f1e5d26ea5375d591c9f11313f11226e35a5d2e7ef3ff54184fec3", + "size_bytes": 149539 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-4gpu/attempt-5055864/gpu-telemetry-5055864.csv": { + "sha256": "336a8aaf180fbca3ef765d87526d39ded5d94458dddc6399e3b3150300789eff", + "size_bytes": 630000 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-4gpu/attempt-5055864/launcher.log": { + "sha256": "c915e0421724b0ff0c94c7859e2b4212bbefb040f962f8066f0baee4e705103a", + "size_bytes": 140944 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-4gpu/attempt-5055864/phase-memory-5055864.csv": { + "sha256": "0bc8bbcf2213a9b1e1307d09b0e40de783a5543de338d67501fdf738b4a26862", + "size_bytes": 199996 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-4gpu/attempt-5055864/training-5055864.log": { + "sha256": "7e2179f00a1cc6ec75671faab917cbc8eea7454df0c07f0a275ed6e7f97b7106", + "size_bytes": 140778 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s161-4gpu/attempt-5055864/trial-5055864.json": { + "sha256": "87f26af8968fd9e52e53e2455cd9068dd711d8bcbb02c8e240b9248c76850939", + "size_bytes": 1424 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-2gpu/attempt-5055859/device-evidence/probe-exit-idle.json": { + "sha256": "d7e1aa648ee83e84de3bc1d7c58ae4b87b27bab0e7cd059f57b52ec426f9f550", + "size_bytes": 3689 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-2gpu/attempt-5055859/device-evidence/ray-resources.json": { + "sha256": "ceca2d7ffdb6c4d7e6acb7e04bb44475273b95e4d5d901abc8b18e50801321de", + "size_bytes": 1558 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-2gpu/attempt-5055859/device-evidence/trainer-cuda.json": { + "sha256": "cb94127ca0d7494b5a10de55847516ecaa125cf7f68714b38b51b8b53466a12f", + "size_bytes": 574 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-2gpu/attempt-5055859/device-evidence/trainer-exec.json": { + "sha256": "6646883863a9523ea9266df2de5a999c9646fcc371013ab5470dd67f0e6f0b46", + "size_bytes": 366 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-2gpu/attempt-5055859/device-evidence/trainer-terminal.json": { + "sha256": "38963c8ef334c8a20eeccba605cb9f9fcfa953a17ab310a4bd854ea39cb3226b", + "size_bytes": 164 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-2gpu/attempt-5055859/device-evidence/worker-actor_rollout-0-3524806.json": { + "sha256": "6a7f7a063f54cf34d8eb6d2d3bc8d59c6956e15e56fa40de44495f7eba8dc2fd", + "size_bytes": 656 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-2gpu/attempt-5055859/device-evidence/worker-actor_rollout-1-3524807.json": { + "sha256": "463403cbc813bafe388acbef24bfe849a2165574d9ae2f5a47a7279e2f8541e6", + "size_bytes": 656 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-2gpu/attempt-5055859/dispatch.json": { + "sha256": "23ad348e9a0bb4c0ebf9410cecfc1e35c9d5961185d9b804a03e53ff3f2ee025", + "size_bytes": 4318 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-2gpu/attempt-5055859/environment-5055859.json": { + "sha256": "009af079583012e7cd7f4c8e99ba5aebd723292b1093c88a7176095241642a11", + "size_bytes": 4414 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-2gpu/attempt-5055859/gpu-device-map.json": { + "sha256": "c742c595089726a43c76cafa21708c8667cbe40c17f2fc3097b613980ac8938e", + "size_bytes": 470 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-2gpu/attempt-5055859/gpu-memory-5055859.csv": { + "sha256": "9f70bec34a73f50d64f5b98d33bd4351150f6b855ae106082ee4412e4a037f09", + "size_bytes": 37641 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-2gpu/attempt-5055859/gpu-telemetry-5055859.csv": { + "sha256": "a6ad33cf45a48734d90247b4a43ff33692eaf6ae1261cfcd85a429afeab767ff", + "size_bytes": 112122 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-2gpu/attempt-5055859/launcher.log": { + "sha256": "8ef75c910c46ad8d9825ac3c3333839e29d019fc90a8802bf953f500378f1d42", + "size_bytes": 144385 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-2gpu/attempt-5055859/phase-memory-5055859.csv": { + "sha256": "a86a8dfc41e9984a7225f5e45ac48d448c65697496a3eabb3c2c445a2e00e847", + "size_bytes": 56856 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-2gpu/attempt-5055859/training-5055859.log": { + "sha256": "f83faa610986608599b0e517acf8396c8e6a1522f81d65f07d192c2793df6f6b", + "size_bytes": 144219 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-2gpu/attempt-5055859/trial-5055859.json": { + "sha256": "1a7488a0481b8bf3032cb77f194084e2cd85cde2d8c2da4dc8eab5d60c0a701a", + "size_bytes": 1424 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-4gpu/attempt-5055859/allocator-traces/allocator-3539673.jsonl": { + "sha256": "3f84ecfb3ba77e54f57f136a6e0defca7be0a4146c882dcb9be6085c4b58ff71", + "size_bytes": 1407 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-4gpu/attempt-5055859/allocator-traces/allocator-3539674.jsonl": { + "sha256": "a510bb169fa55ce5dd474fd0f10f9bdd57c5bdf3e20524e7ee20236a76c2c264", + "size_bytes": 1407 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-4gpu/attempt-5055859/allocator-traces/allocator-3539675.jsonl": { + "sha256": "5989a701dc26009a6a9927c3593ea914baf0f201e0e632fb5538437db1684e0b", + "size_bytes": 1413 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-4gpu/attempt-5055859/allocator-traces/allocator-3539676.jsonl": { + "sha256": "1bf7c0db77e488b6d62b32c74025a99665f8a98976c9c8d308beb6468b091f0e", + "size_bytes": 1411 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-4gpu/attempt-5055859/device-evidence/probe-exit-idle.json": { + "sha256": "df325c5d5dcf6d772ee453e29ef100ff2b1149bcabafa04e6075e289cc63773b", + "size_bytes": 3689 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-4gpu/attempt-5055859/device-evidence/ray-resources.json": { + "sha256": "0f8edd00fb5cb5f0cbc01b1f2b1234487c876c9b1741c3ee1504b326ec2c44ea", + "size_bytes": 1558 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-4gpu/attempt-5055859/device-evidence/trainer-cuda.json": { + "sha256": "83cf2e3d2f191a0bd7d22bc1be1867b02b2f9c9de0953119fadf5929fa39d9f1", + "size_bytes": 856 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-4gpu/attempt-5055859/device-evidence/trainer-exec.json": { + "sha256": "299b81402619b96c53575c419de06f7aea3a7499cb120cb69bf60a682a3e288c", + "size_bytes": 448 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-4gpu/attempt-5055859/device-evidence/trainer-terminal.json": { + "sha256": "e355eca03a188cc59bb1427390d7c3dbf1bd2f85b04be3cc56d6845768dd50be", + "size_bytes": 164 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-4gpu/attempt-5055859/device-evidence/worker-actor_rollout-0-3539673.json": { + "sha256": "c0dcc6b3817f1db4962f86e67dd98ae9e3928de1044ffb1eda60300dc23e062b", + "size_bytes": 656 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-4gpu/attempt-5055859/device-evidence/worker-actor_rollout-1-3539674.json": { + "sha256": "0f0a3f3b79a0196527783032a568d89c96ede1bece1fc690cee34c7fac04aada", + "size_bytes": 656 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-4gpu/attempt-5055859/device-evidence/worker-actor_rollout-2-3539675.json": { + "sha256": "c290d4472e0ba843817466a3eb90d8f2848a94f91253e45af7ec924db42a31b3", + "size_bytes": 656 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-4gpu/attempt-5055859/device-evidence/worker-actor_rollout-3-3539676.json": { + "sha256": "42e00bc12078e63cf9941ea356251cd007f94da8256f25abb6538ebcc74c0798", + "size_bytes": 656 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-4gpu/attempt-5055859/dispatch.json": { + "sha256": "e0e0590fd32a6048876b2f93e24ea174a662f1b6a1111e44b84d96bd67d7e7bf", + "size_bytes": 4500 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-4gpu/attempt-5055859/environment-5055859.json": { + "sha256": "762d6a8d8a2a3fd6c6e19fbbc1cb9d2cdef558dd78838eebaf34abe9082bc2ca", + "size_bytes": 4896 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-4gpu/attempt-5055859/gpu-device-map.json": { + "sha256": "7595b966ca546cc6d3a9f8552a47f17337b68b5df71d21ee4d607d925cea0809", + "size_bytes": 810 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-4gpu/attempt-5055859/gpu-memory-5055859.csv": { + "sha256": "0afb280d3ced36c4d7b3818cd30e102b482734d0c8c9018e3b6c6cc3d032960b", + "size_bytes": 184880 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-4gpu/attempt-5055859/gpu-telemetry-5055859.csv": { + "sha256": "3ef7237492bc0b9e74a0ab6fa28330d9e78dd8ea74e7363d36c8a5d04487ca26", + "size_bytes": 773393 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-4gpu/attempt-5055859/launcher.log": { + "sha256": "f0ae16874095d68c1a79e32fe7cbfd31a85feeb29b7db0410816934ca5ed0130", + "size_bytes": 140603 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-4gpu/attempt-5055859/phase-memory-5055859.csv": { + "sha256": "a6851d4660bbd5d2739d52e3b2865a88e6ae4e413dd5ddb31a42924dacbe4d31", + "size_bytes": 246219 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-4gpu/attempt-5055859/training-5055859.log": { + "sha256": "7902fc03bb9ad4608e8d8006d7150fac599d957d8439f7ce75b17c984793e761", + "size_bytes": 140437 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s162-4gpu/attempt-5055859/trial-5055859.json": { + "sha256": "2c198c29365b1a55a281b4d06c444124e632b9472b4e2f3f7386f54ef69c3c5f", + "size_bytes": 1424 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-2gpu/attempt-5055863/device-evidence/probe-exit-idle.json": { + "sha256": "6f49019b647d632fa0688dbe1fdd3c7cf3c8894de9b2914a5784ec23078cdcfa", + "size_bytes": 3687 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-2gpu/attempt-5055863/device-evidence/ray-resources.json": { + "sha256": "acee45cccc0405a605ced0baa6137b21aa915807a3e37b08c201956085e40593", + "size_bytes": 1551 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-2gpu/attempt-5055863/device-evidence/trainer-cuda.json": { + "sha256": "26325ab7c7afae5eaa70df35c292fa87b94429b39825155cc656c68e259f2eb3", + "size_bytes": 572 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-2gpu/attempt-5055863/device-evidence/trainer-exec.json": { + "sha256": "ae4cf78c2a118a01c7ff85f330e70e516110100823671b7f52faeace4660e1b2", + "size_bytes": 365 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-2gpu/attempt-5055863/device-evidence/trainer-terminal.json": { + "sha256": "ef6324517a61d535813dcffecb9bef82e85ae7aae3429f9eb0da855cacab067e", + "size_bytes": 163 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-2gpu/attempt-5055863/device-evidence/worker-actor_rollout-0-873976.json": { + "sha256": "25e15c6d6227c7c16dc68b64af3e03520331ca34d4625943890fd0b08ff41bd2", + "size_bytes": 655 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-2gpu/attempt-5055863/device-evidence/worker-actor_rollout-1-873977.json": { + "sha256": "ed331e46edbfbca3a0687291e9a2d3673b245119fff3bb1c9acbe737e3f4b62a", + "size_bytes": 655 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-2gpu/attempt-5055863/dispatch.json": { + "sha256": "f992efaf90e3139177fa0ffaa6d9f99582d65d6b19f53cf95cc147c451e1b734", + "size_bytes": 4318 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-2gpu/attempt-5055863/environment-5055863.json": { + "sha256": "8c6910e80b91671965264e645285ef4b6bebaf4811e704b580fcf61c1be93881", + "size_bytes": 4414 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-2gpu/attempt-5055863/gpu-device-map.json": { + "sha256": "8619b487797b3e1f722a2dbefca1834ed378b0bb2e248a74fb598a36ce3b4b7c", + "size_bytes": 470 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-2gpu/attempt-5055863/gpu-memory-5055863.csv": { + "sha256": "baa37fa35eff6146f3d5d5f651264a3b14a34dd8ccc491b247e87125ed6ed1f2", + "size_bytes": 36664 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-2gpu/attempt-5055863/gpu-telemetry-5055863.csv": { + "sha256": "c46c326d82545eafa153cdcf0cd934fff63fc461500a81a173646ffcedce68e5", + "size_bytes": 109072 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-2gpu/attempt-5055863/launcher.log": { + "sha256": "64404885773717a98e7918c7f5105cce0bf8f729472cc1b1590286fddc321d40", + "size_bytes": 150636 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-2gpu/attempt-5055863/phase-memory-5055863.csv": { + "sha256": "c6391a964849aed375ebae6401c5eb5ecf539b158fde433532add57193763dcb", + "size_bytes": 55339 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-2gpu/attempt-5055863/training-5055863.log": { + "sha256": "17c19dfa4d4ad3112381f58d6ed78a252194ac6e109f94c6d81b8e719f1b1361", + "size_bytes": 150470 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-2gpu/attempt-5055863/trial-5055863.json": { + "sha256": "d240f525d4a2dda5bb865b2c7776e06040494133d6fb0ff393ea6ad153341e9c", + "size_bytes": 1424 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-4gpu/attempt-5055863/allocator-traces/allocator-843954.jsonl": { + "sha256": "57395fe2006d7015b9ef16a947f46306d8343a95bd81a3f9c163d04a49bc9609", + "size_bytes": 1396 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-4gpu/attempt-5055863/allocator-traces/allocator-843955.jsonl": { + "sha256": "579b3bc01e48c774186f34e616df6e46693bbec9ba67fb983e1449a38804043c", + "size_bytes": 1404 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-4gpu/attempt-5055863/allocator-traces/allocator-843956.jsonl": { + "sha256": "87ecc7af97fa3c3644b00b27be9844fcb34a8dd7dde38514a52aaf39e14215cf", + "size_bytes": 1404 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-4gpu/attempt-5055863/allocator-traces/allocator-843957.jsonl": { + "sha256": "e327495b335caa7d68ab8b2db0354cbc3ee44457579d42e4fc686f691e5b7501", + "size_bytes": 1404 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-4gpu/attempt-5055863/device-evidence/probe-exit-idle.json": { + "sha256": "7fd57b5d734135814de94b6ac7521afca704a3db06387a88b7eb803cef02ce45", + "size_bytes": 3687 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-4gpu/attempt-5055863/device-evidence/ray-resources.json": { + "sha256": "1b30a754890f7ac0b31d02e03793cf08eae6def09680e96e60cb1f141864a8fd", + "size_bytes": 1551 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-4gpu/attempt-5055863/device-evidence/trainer-cuda.json": { + "sha256": "7a3fd4b406870ef39c20a069b0338a463b5d80b0ec7eb005d75471cbe9aae1ab", + "size_bytes": 854 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-4gpu/attempt-5055863/device-evidence/trainer-exec.json": { + "sha256": "6b8fd626a18b67c0475304cd620af4433a6192418b3c74f09674db9e1946806d", + "size_bytes": 447 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-4gpu/attempt-5055863/device-evidence/trainer-terminal.json": { + "sha256": "07c122d5cd4b5ec36a62906348e53c0cc02cae67758509625676905dc15c339b", + "size_bytes": 163 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-4gpu/attempt-5055863/device-evidence/worker-actor_rollout-0-843954.json": { + "sha256": "83ff37fcc1ef0bf0bfcfbaf08d423625a762054b7d3f369a15ec4996ae223ca7", + "size_bytes": 655 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-4gpu/attempt-5055863/device-evidence/worker-actor_rollout-1-843955.json": { + "sha256": "b4681c2f1044304f9c46703d0ce66ddbdddeeec1bd146db8e117126d9cb8d778", + "size_bytes": 655 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-4gpu/attempt-5055863/device-evidence/worker-actor_rollout-2-843956.json": { + "sha256": "415c5f4c909141580f36370e3fd29989ead8dbdb3f2f9cc8f47e22e41670e67e", + "size_bytes": 655 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-4gpu/attempt-5055863/device-evidence/worker-actor_rollout-3-843957.json": { + "sha256": "e6f798dc163b1ef76928f10de5239919d4fa14cee5ea92f90fef0f803f1499ce", + "size_bytes": 655 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-4gpu/attempt-5055863/dispatch.json": { + "sha256": "bf51d850168699d0ede13c8f8a6c722464545023e81f605072d4874d648ad9b0", + "size_bytes": 4500 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-4gpu/attempt-5055863/environment-5055863.json": { + "sha256": "2ac19c3094ef999e349caab05586f1ffce9c568cf99aa51824da1431ee979a63", + "size_bytes": 4896 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-4gpu/attempt-5055863/gpu-device-map.json": { + "sha256": "fe307eb340ed2932b4dadfe0c421932b281fed01bba1f40863fd4e26e1380722", + "size_bytes": 810 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-4gpu/attempt-5055863/gpu-memory-5055863.csv": { + "sha256": "73ac680c79b74b6e39fc9d3406d92b19f04347f72fd4074d6cded9261e7addda", + "size_bytes": 187686 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-4gpu/attempt-5055863/gpu-telemetry-5055863.csv": { + "sha256": "7c9735d625e31113ec70e643ef12065986c485eb16ff037a93fd4ad1aaca4377", + "size_bytes": 792219 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-4gpu/attempt-5055863/launcher.log": { + "sha256": "b85345393decf96dbf37dd352b56788b77dcb7ff9fca47f63521a1ff17ffc7f6", + "size_bytes": 139931 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-4gpu/attempt-5055863/phase-memory-5055863.csv": { + "sha256": "c5028bc7781e488666283c064d251e7d35c033c62e8b9db4daa38e2f653da022", + "size_bytes": 251256 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-4gpu/attempt-5055863/training-5055863.log": { + "sha256": "4a57f829f41ce3ed8e5d5ef25e7442157533be345b2c675eb4c8b19049dac858", + "size_bytes": 139765 + }, + "output/estimation_host_capacity/host-qwen25-7b-math-c2-s163-4gpu/attempt-5055863/trial-5055863.json": { + "sha256": "e8bb9e6c3a8e9d9d96f54c2015c592efee7865dede33a90fe94d8cfbedf082d3", + "size_bytes": 1424 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-code_heavy_tail-c2-s161/allocation-trace.jsonl": { + "sha256": "ccfd3fad0f8bb5a5efc0b5f7fd6ddc583fa31e51813fed28eb3736ba721c122f", + "size_bytes": 358865 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-code_heavy_tail-c2-s161/ledger.jsonl": { + "sha256": "333c3a9409bf517fd2ca41c6b17298d3c18ffde6d0780bbd7bd4b980953a7a54", + "size_bytes": 13600 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-code_heavy_tail-c2-s161/pair-summary.json": { + "sha256": "9589ed74c9ed4da96bcfd2bd6ec994784b44e4f57150c9cd881cb516dfd86e35", + "size_bytes": 801 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-code_heavy_tail-c2-s161/pair.json": { + "sha256": "11f432ad69e4f19a7e964601235286986f30fb97862cb358860a3b4a45b81ed4", + "size_bytes": 8824 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-code_heavy_tail-c2-s161/period-1.json": { + "sha256": "a5537abf1e57731739c04dae9526c5e5899ffd5e4e684341ef9d0124dee0421f", + "size_bytes": 6261 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-code_heavy_tail-c2-s161/period-2.json": { + "sha256": "ebfddf6918a70324f056725c6e96eb25da12ab1a7d3016fe970badfcaaa38b23", + "size_bytes": 5462 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-code_heavy_tail-c2-s162/allocation-trace.jsonl": { + "sha256": "19aaab448f0fdb85bc998cf177496fa5bf4a1d5c355dd7cac59c2f7342d5580f", + "size_bytes": 419764 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-code_heavy_tail-c2-s162/ledger.jsonl": { + "sha256": "1c00aa35420037ed210fc575999f0151051f8b4627b06d67904904e489f0c6dd", + "size_bytes": 13602 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-code_heavy_tail-c2-s162/pair-summary.json": { + "sha256": "5550e11a235fcba0459f608f88fee5dae9a301f034674cfdc4ced162d34fc5e8", + "size_bytes": 803 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-code_heavy_tail-c2-s162/pair.json": { + "sha256": "ba5822d60980ed2368ce46612583322f5385a116a98a85abeb700f1080e19e5a", + "size_bytes": 8824 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-code_heavy_tail-c2-s162/period-1.json": { + "sha256": "29ee4da55c986383b3a165192ff467a9224237d60c5a7bc9a58848e6ff2cd71b", + "size_bytes": 5465 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-code_heavy_tail-c2-s162/period-2.json": { + "sha256": "9470f412d34b3d7433d10665af7f11bc0f52b346c71f2675d52d2fee33881ca5", + "size_bytes": 6244 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-code_heavy_tail-c2-s163/allocation-trace.jsonl": { + "sha256": "369da5916ed13fce410cd1922309503f2c220849828ca994364bcaf1c3b30f50", + "size_bytes": 396639 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-code_heavy_tail-c2-s163/ledger.jsonl": { + "sha256": "9d18e012ffc9f2064dcbad10dafa59b750ce161497eec51e0e6bc176ce46c3c4", + "size_bytes": 13610 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-code_heavy_tail-c2-s163/pair-summary.json": { + "sha256": "8a8f7dec9e09fdefb9731d565895496696d87ea02d9d6ee9a9075e0c18772912", + "size_bytes": 800 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-code_heavy_tail-c2-s163/pair.json": { + "sha256": "d9475a95dfbf2600d351db83bfc49b45774c5e13c2ab968d21a2168b43d9573f", + "size_bytes": 8824 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-code_heavy_tail-c2-s163/period-1.json": { + "sha256": "a05a722b4e27d32971cd07914935b60ebe8f8f2cc329304452fde6ca5a5b016f", + "size_bytes": 5466 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-code_heavy_tail-c2-s163/period-2.json": { + "sha256": "71a56d85b9394891cdc42fc94565d97a181b3dad8a6deb6816fcd2b25c9a3e7e", + "size_bytes": 6260 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-gsm8k-c2-s161/allocation-trace.jsonl": { + "sha256": "83094617e346df15c7f7167d7732c48e714a5497dcb73aae8781a6aea141f77f", + "size_bytes": 413666 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-gsm8k-c2-s161/ledger.jsonl": { + "sha256": "0d00f6eb5b648385af3f6ce1714c02b352b0a05dad2d539984f23e9783c471ce", + "size_bytes": 13294 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-gsm8k-c2-s161/pair-summary.json": { + "sha256": "081fe31718aeeace255a1af16a0df374a240de1b90eb2d6a996f2e237b6d960f", + "size_bytes": 792 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-gsm8k-c2-s161/pair.json": { + "sha256": "cfa0de3b3cd08dfeacfdce5c3ebf3e3e2cef59bc2da204889e86042c4850708f", + "size_bytes": 8496 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-gsm8k-c2-s161/period-1.json": { + "sha256": "c3ec892da14b05f0af55e38ca97e8e9eb07cf998df59452f4510f386e80076e4", + "size_bytes": 5357 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-gsm8k-c2-s161/period-2.json": { + "sha256": "67b7c5427d7a1941fb34e5e0aa67384b1b6850fa997be9012431df9822e4331f", + "size_bytes": 6135 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-gsm8k-c2-s162/allocation-trace.jsonl": { + "sha256": "b8759a7ec6b648418323b0d14ed2d88d660bd616bf6d2fa40fdafc5a13a7eca6", + "size_bytes": 342799 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-gsm8k-c2-s162/ledger.jsonl": { + "sha256": "4dcf9a91ae6bbba137a4afab1907896accd528c333c05f9396998563518b7eaa", + "size_bytes": 13274 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-gsm8k-c2-s162/pair-summary.json": { + "sha256": "585e717cf86fe5df810bbc3d33ad0a12de04065cde207dfde5cf5976d18c9e27", + "size_bytes": 793 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-gsm8k-c2-s162/pair.json": { + "sha256": "cb05d94260ce1ee8a79015825d072921a10254d9d914a0df3e5c51f3be375fef", + "size_bytes": 8496 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-gsm8k-c2-s162/period-1.json": { + "sha256": "a56fcab542b3c7da013debc7b7b96cdb5e26d2b4c83176f403ab8fa73a2d32e5", + "size_bytes": 6123 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-gsm8k-c2-s162/period-2.json": { + "sha256": "44da63e937ba49567ca09bbcdde1aaba76fc4a9cf9e6b6522d88af173c9ac797", + "size_bytes": 5350 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-gsm8k-c2-s163/allocation-trace.jsonl": { + "sha256": "9a2e78c0bf284a4ad215c97f9e34bb68553cc92db3e381702d8c754cb55b9df3", + "size_bytes": 506203 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-gsm8k-c2-s163/ledger.jsonl": { + "sha256": "cde48950a284805dca7150f75182dab4ab0a034cd0b41efeb9fe2b83512a8dd9", + "size_bytes": 13287 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-gsm8k-c2-s163/pair-summary.json": { + "sha256": "73fb54c4e55429b4a686af60ba9459a974b6654b21d4dcde246319c3c6a5c955", + "size_bytes": 790 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-gsm8k-c2-s163/pair.json": { + "sha256": "10a666b083e54c59a9b1ac6144602a9e67fe987c366ba739ea87679570ff7d22", + "size_bytes": 8496 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-gsm8k-c2-s163/period-1.json": { + "sha256": "4a1d5241e8a0418bdce0e39b38a44556de7500c6b35a3c3f30b6e5b1e6ff1fa6", + "size_bytes": 5356 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-gsm8k-c2-s163/period-2.json": { + "sha256": "a9f64eb27668c9a5cdb98b5e5e6005ba493ebff92240b507d910bcd8eac78e8a", + "size_bytes": 6131 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-math-c2-s161/allocation-trace.jsonl": { + "sha256": "d5e51f2f9959f7f15226784bda3682940d8c96712e3e51361c998c085e5d1077", + "size_bytes": 364372 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-math-c2-s161/ledger.jsonl": { + "sha256": "35af3985e7adf8837ec64dd3380dbac41b910e49a59dcb258c793c7bf4e0b53e", + "size_bytes": 13265 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-math-c2-s161/pair-summary.json": { + "sha256": "d38e7e03e7a22eb15506a9d5c7099f4f2a6f31339d66a628997622e6bc2906fa", + "size_bytes": 789 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-math-c2-s161/pair.json": { + "sha256": "da4248ba78237ac4377771fbd810c138222c7d89cc1960cd484ada584e184ab9", + "size_bytes": 8481 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-math-c2-s161/period-1.json": { + "sha256": "7212a19d01836b08a6f071c27a99f53f30cf76820f2fc596528f2787c38844db", + "size_bytes": 6137 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-math-c2-s161/period-2.json": { + "sha256": "2377679d475af540267575d66cdde6c6d6230674d3da1c752d351e67ccd7544a", + "size_bytes": 5343 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-math-c2-s162/allocation-trace.jsonl": { + "sha256": "57034385a9e3bd4f768f9898fbc80b3eeed5b10509ac57b97fcce5ff8b7f5c4c", + "size_bytes": 429045 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-math-c2-s162/ledger.jsonl": { + "sha256": "dc8d64e4c1acaab7d56a2609e8325cf9f18b62861342caf38bd2d2458019e1bb", + "size_bytes": 13286 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-math-c2-s162/pair-summary.json": { + "sha256": "1abe4426e172d9f16111fba7f1a9f92d4853837c1e26230fcdcb6029c9808ab6", + "size_bytes": 790 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-math-c2-s162/pair.json": { + "sha256": "29d4201db5096e9951dd1c70a5df4b57c945d95dfe55d30830b1b4db32ab4186", + "size_bytes": 8481 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-math-c2-s162/period-1.json": { + "sha256": "7cca740094b97c6e82320269fcc5be6870524c88be04613d7609b93df179e6b2", + "size_bytes": 5344 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-math-c2-s162/period-2.json": { + "sha256": "fe07fc54ad949b6114c46690501a5cdbf64267f3db1523bed1a7fe2368de2af5", + "size_bytes": 6168 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-math-c2-s163/allocation-trace.jsonl": { + "sha256": "9eda9b409798a408cecb8b6be83f79dba013094c068a73df15b2f4c0f1af1b10", + "size_bytes": 426531 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-math-c2-s163/ledger.jsonl": { + "sha256": "045f78ee36bcaec336b9f1ab0a0b08096132b5eff8d76089ceaed23da8eeb80a", + "size_bytes": 13242 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-math-c2-s163/pair-summary.json": { + "sha256": "54714f312a7b5defe828075d2b4adff38fc4c49885b37f94baa6e1ff00cbae83", + "size_bytes": 790 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-math-c2-s163/pair.json": { + "sha256": "d0fb1b825aaa3e394285a542e31a674611288684d7ef2a2f5eae6a96b4659e95", + "size_bytes": 8481 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-math-c2-s163/period-1.json": { + "sha256": "d73d2e6a98e0a30b7c40640b568d53fd6f940ab5995b758559f12ff9b753a098", + "size_bytes": 6111 + }, + "output/estimation_host_capacity/pairs/host-qwen25-7b-math-c2-s163/period-2.json": { + "sha256": "e69aad8c200ebbb97e87a385322ee34c17ea0a00c7644ad0b7a29effc28f2525", + "size_bytes": 5341 + } + }, + "original_amendments_sha256": { + "amendment-01.json": "b02fe0419f865c4cc3d77fb35bb4205a54f3aa9605fdadc7b86a9a11b172d25e" + }, + "original_prediction_seal_sha256": "5ed714caff922a5109346a4242545d19f2824636d0f52bb0bbd2a09b40ed4666", + "original_protocol_sha256": "1694e447e842600612604891e7b3d036ff0791fa84c4bdbeba54ed093feefdeb", + "panel_scope": "Separate host-resource follow-up; no pooling with original slots.", + "prediction_frozen_at_utc": "2026-09-15T14:10:21.535819+00:00", + "prediction_inheritance": "Exact original sealed predictions with new configuration IDs; no fitting or outcome-dependent update.", + "prediction_seal_binding": "Effective input/output hashes and pre-claim chronology; runner verifies seal but does not store its hash per period.", + "prediction_seal_sha256": "5ed714caff922a5109346a4242545d19f2824636d0f52bb0bbd2a09b40ed4666", + "protocol_input_evidence": { + "matrix.csv": { + "sha256": "1d2b2af601361a3cb46b65ff3be263685b3f725f2b0e4787a8ff9966084f5332", + "size_bytes": 8931 + }, + "predictions.json": { + "sha256": "5ac9cef3c6e7d08036d29315be9cb6220f6bb85fe48a0475f97f2ebce66b9980", + "size_bytes": 10821 + }, + "protocol.json": { + "sha256": "f3f49617a7fca63ddac8923bccb83a15eaf52646e5f3db5df4aa716bb25c1ef6", + "size_bytes": 5676 + }, + "study_freeze.json": { + "sha256": "80559199c59246d0e4ce52bf268d0375c90fc5168a5246ed900c9266abd951cd", + "size_bytes": 610 + }, + "targets.csv": { + "sha256": "4aaa0913debe9425ef40664f0554b2ed15980f4cab59bf8223dc86093f313fdc", + "size_bytes": 2885 + } + }, + "protocol_relative_dir": "benchmark/host_capacity", + "protocol_sha256": "f3f49617a7fca63ddac8923bccb83a15eaf52646e5f3db5df4aa716bb25c1ef6", + "raw_statuses_used_as_labels": false, + "runner_validation_implementation_sha256": "f340fb7eec5297cb675a48e267618c15d7787a42891d4a2ae4a670ec24a0cc47", + "source_commit_expected": null, + "source_input_evidence": { + "benchmark.py": { + "sha256": "35e36caad46d2814b0ac9f8180961de5564746347cf668d5c3fc1e643fd477e8", + "size_bytes": 10897 + }, + "benchmark/attempts.csv": { + "sha256": "397c8caaf0d3abe06587fd649c181fe71d5b8b5f5e24e0591cc7e56d38e275c1", + "size_bytes": 588698 + }, + "benchmark/configurations.csv": { + "sha256": "3a77c96a67b82e01d23fb6f7a440cba23041bed1dae2749072cb9abb5b5d239d", + "size_bytes": 71349 + }, + "benchmark/estimation/amendment-01.json": { + "sha256": "b02fe0419f865c4cc3d77fb35bb4205a54f3aa9605fdadc7b86a9a11b172d25e", + "size_bytes": 4632 + }, + "benchmark/estimation/fitted_model.json": { + "sha256": "8b84aa1ae57fe56fd478817dbbce505487cbef6b26652ae2c8f45e9fa5ef5df4", + "size_bytes": 11485 + }, + "benchmark/estimation/matrix.csv": { + "sha256": "1ff175e22c76bf1af3857b4ae57cd4f78bafe0d280a295d696e96815ef85e9c8", + "size_bytes": 16755 + }, + "benchmark/estimation/model_metadata.json": { + "sha256": "2f0ce16c13e72900d2a60f9b7ad28d4dc5ed43d7915d8a88925ba9c5ffd684ad", + "size_bytes": 5599 + }, + "benchmark/estimation/prediction_freeze.json": { + "sha256": "5ed714caff922a5109346a4242545d19f2824636d0f52bb0bbd2a09b40ed4666", + "size_bytes": 3004 + }, + "benchmark/estimation/predictions.json": { + "sha256": "96d7f2ae11145f0985c0faf743333d941afbcd3b80966a857b1af0af15306ace", + "size_bytes": 21607 + }, + "benchmark/estimation/protocol.json": { + "sha256": "1694e447e842600612604891e7b3d036ff0791fa84c4bdbeba54ed093feefdeb", + "size_bytes": 11490 + }, + "benchmark/estimation/retrospective.json": { + "sha256": "8ada76da63ab91e0b1ca6276a98f2b096979fe6acd7578c1223469d55d399f3a", + "size_bytes": 228053 + }, + "benchmark/estimation/targets.csv": { + "sha256": "f2481278127fa726f24297d81fc9df46e999315fe543167ed398e671fffa522f", + "size_bytes": 5037 + }, + "benchmark/outcomes.csv": { + "sha256": "af05cd91ed1de843735cc188e5e54dbd1645b3ed91c97f503ac3de67bc5f61c9", + "size_bytes": 21219 + }, + "memory_tuner/estimation_baselines.py": { + "sha256": "8cc20f9697eb38cb6a19a2255e95712ac44469cbd147eaf93e7e7ed183360ad2", + "size_bytes": 26850 + }, + "memory_tuner/host_capacity_protocol.py": { + "sha256": "9bf811457f04e02e67e31d3c9289d1155f73ee1364a233b6b77d08d86becf44b", + "size_bytes": 16560 + }, + "memory_tuner/model_memory_metadata.py": { + "sha256": "8845a1197b49d9463ebe5dc8c953463dd2106361c54e7b12244057c48aaa2712", + "size_bytes": 6588 + }, + "memory_tuner/plan_estimation_study.py": { + "sha256": "16855f584441528870fcca38c3a19eafd1234c3eacb9b43cc9eebdd4bd6ad862", + "size_bytes": 10472 + }, + "memory_tuner/run_host_capacity.py": { + "sha256": "23f1eaf462f1825a77e91d2c8d8a4ab073f4debe0fbab134ff1863e096345d37", + "size_bytes": 1497 + }, + "memory_tuner/run_matched_gpu.py": { + "sha256": "f340fb7eec5297cb675a48e267618c15d7787a42891d4a2ae4a670ec24a0cc47", + "size_bytes": 49744 + }, + "run_host_capacity.slurm": { + "sha256": "59d3daa9499d1fc76d0b856be6c0a813472f488eb49f990e6ffa57ccf819b919", + "size_bytes": 1231 + } + }, + "study_design_validation_implementation_sha256": "9bf811457f04e02e67e31d3c9289d1155f73ee1364a233b6b77d08d86becf44b", + "study_freeze_sha256": "80559199c59246d0e4ce52bf268d0375c90fc5168a5246ed900c9266abd951cd", + "study_frozen_at_utc": "2026-09-15T18:52:11.222223+00:00", + "study_group": "estimation_host_capacity", + "study_loader_implementation_sha256": "9726f0499e7774c25752f7ce02f2dfeada03ffd5a14183621c5791f9e674abc4" +} diff --git a/benchmark/host_capacity/results/summary.json b/benchmark/host_capacity/results/summary.json new file mode 100644 index 0000000000000000000000000000000000000000..679760040eae153092f3991389a0cd7797f4aa20 --- /dev/null +++ b/benchmark/host_capacity/results/summary.json @@ -0,0 +1,13 @@ +{ + "completed_processes": 0, + "eligible_processes": 9, + "known_launched_processes": 18, + "pair_errors": {}, + "planned_configurations": 6, + "planned_pairs": 9, + "planned_processes": 18, + "resolved_configurations": 3, + "schema_version": 1, + "unknown_launch_status_processes": 0, + "unresolved_processes": 9 +} diff --git a/benchmark/host_capacity/study_freeze.json b/benchmark/host_capacity/study_freeze.json new file mode 100644 index 0000000000000000000000000000000000000000..42a6bf656d03cf108afd5ce2f39be5a024a0c075 --- /dev/null +++ b/benchmark/host_capacity/study_freeze.json @@ -0,0 +1,11 @@ +{ + "files_sha256": { + "matrix.csv": "1d2b2af601361a3cb46b65ff3be263685b3f725f2b0e4787a8ff9966084f5332", + "predictions.json": "5ac9cef3c6e7d08036d29315be9cb6220f6bb85fe48a0475f97f2ebce66b9980", + "targets.csv": "4aaa0913debe9425ef40664f0554b2ed15980f4cab59bf8223dc86093f313fdc" + }, + "frozen_at_utc": "2026-09-15T18:52:11.222223+00:00", + "original_prediction_seal_sha256": "5ed714caff922a5109346a4242545d19f2824636d0f52bb0bbd2a09b40ed4666", + "protocol_sha256": "f3f49617a7fca63ddac8923bccb83a15eaf52646e5f3db5df4aa716bb25c1ef6", + "status": "frozen_after_host_censoring_before_followup_execution" +} diff --git a/benchmark/host_capacity/targets.csv b/benchmark/host_capacity/targets.csv new file mode 100644 index 0000000000000000000000000000000000000000..94c848176ae33c14908b732e3ff42c30c0224850 --- /dev/null +++ b/benchmark/host_capacity/targets.csv @@ -0,0 +1,7 @@ +configuration_id,case_id,study,model_family,model_name,model_id,workload_id,workload_name,legacy_dataset_id,algorithm,adaptation,configuration_level,condition,gpu_count,rollout_tp_size,actor_micro_batch,rollout_logprob_micro_batch,ref_logprob_micro_batch,vllm_gpu_memory_utilization,parameter_offload,optimizer_offload,free_cache_engine,max_prompt_length,max_response_length,max_model_len,max_num_seqs,rollout_n,train_batch_size,train_max_samples,val_max_samples,total_training_steps,gpu_monitor_interval_ms,allocator_trace_enabled,allocator_trace_sync,save_freq,test_freq,val_before_train,resume_mode,max_actor_ckpt_to_keep,accelerator,device_capacity_mib,margin_limit_mib,peak_measurement +host-qwen25-7b-gsm8k-c2-2gpu,host-qwen25-7b-gsm8k-2gpu,estimation_host_capacity,qwen25_7b,Qwen2.5-7B-Instruct,Qwen/Qwen2.5-7B-Instruct,gsm8k,GSM8K mathematics,gsm8k,GRPO,LoRA,c2,,2,1,8,4,4,0.6,False,False,True,1024,2048,3072,64,2,32,32,32,1,100,True,1,-1,-1,False,disable,1,NVIDIA A100-SXM4-40GB,40960,38912,maximum externally sampled per-device memory +host-qwen25-7b-gsm8k-c2-4gpu,host-qwen25-7b-gsm8k-4gpu,estimation_host_capacity,qwen25_7b,Qwen2.5-7B-Instruct,Qwen/Qwen2.5-7B-Instruct,gsm8k,GSM8K mathematics,gsm8k,GRPO,LoRA,c2,,4,1,8,4,4,0.6,False,False,True,1024,2048,3072,64,2,32,32,32,1,100,True,1,-1,-1,False,disable,1,NVIDIA A100-SXM4-40GB,40960,38912,maximum externally sampled per-device memory +host-qwen25-7b-math-c2-2gpu,host-qwen25-7b-math-2gpu,estimation_host_capacity,qwen25_7b,Qwen2.5-7B-Instruct,Qwen/Qwen2.5-7B-Instruct,math,MATH mathematics,math,GRPO,LoRA,c2,,2,1,8,8,8,0.6,False,False,True,512,512,1024,128,2,64,64,64,1,100,True,1,-1,-1,False,disable,1,NVIDIA A100-SXM4-40GB,40960,38912,maximum externally sampled per-device memory +host-qwen25-7b-math-c2-4gpu,host-qwen25-7b-math-4gpu,estimation_host_capacity,qwen25_7b,Qwen2.5-7B-Instruct,Qwen/Qwen2.5-7B-Instruct,math,MATH mathematics,math,GRPO,LoRA,c2,,4,1,8,8,8,0.6,False,False,True,512,512,1024,128,2,64,64,64,1,100,True,1,-1,-1,False,disable,1,NVIDIA A100-SXM4-40GB,40960,38912,maximum externally sampled per-device memory +host-qwen25-7b-code_heavy_tail-c2-2gpu,host-qwen25-7b-code_heavy_tail-2gpu,estimation_host_capacity,qwen25_7b,Qwen2.5-7B-Instruct,Qwen/Qwen2.5-7B-Instruct,longer_prompt_code,Longer-prompt code,code_heavy_tail,GRPO,LoRA,c2,,2,1,8,4,4,0.6,False,False,True,1536,1024,2560,64,2,32,32,32,1,100,True,1,-1,-1,False,disable,1,NVIDIA A100-SXM4-40GB,40960,38912,maximum externally sampled per-device memory +host-qwen25-7b-code_heavy_tail-c2-4gpu,host-qwen25-7b-code_heavy_tail-4gpu,estimation_host_capacity,qwen25_7b,Qwen2.5-7B-Instruct,Qwen/Qwen2.5-7B-Instruct,longer_prompt_code,Longer-prompt code,code_heavy_tail,GRPO,LoRA,c2,,4,1,8,4,4,0.6,False,False,True,1536,1024,2560,64,2,32,32,32,1,100,True,1,-1,-1,False,disable,1,NVIDIA A100-SXM4-40GB,40960,38912,maximum externally sampled per-device memory diff --git a/paper/RLVRAMBench_IEEE_Access_source.zip b/paper/RLVRAMBench_IEEE_Access_source.zip index 6c1f22419a0e95d9456c829e0b5eb73ff8ad0217..378f46b4e31e107270ae2dd361bc9ee4483d87b6 100644 --- a/paper/RLVRAMBench_IEEE_Access_source.zip +++ b/paper/RLVRAMBench_IEEE_Access_source.zip @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:bf0fe6a258cc1d0148cb8bcdb25e33fbc7494335f47ed8b4050502bb76c933b1 -size 1156009 +oid sha256:c7eef7d90561cae3961ec4c98b4cd6d1076ff63b17cf2be2d7604af8016efc1d +size 1160514 diff --git a/paper/RLVRAMBench_IEEE_Access_submission.pdf b/paper/RLVRAMBench_IEEE_Access_submission.pdf index eb7076e1abeceadcf2da9c00287155901e469f61..0df8b9a7cc550e55bff15c610b2e94dd28c7634c 100644 --- a/paper/RLVRAMBench_IEEE_Access_submission.pdf +++ b/paper/RLVRAMBench_IEEE_Access_submission.pdf @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:2ee28b5e8d999d9af7c3553d4cd3496c38333e615d3c1b092ac8f039ab305549 -size 439029 +oid sha256:9b28d833742a1befddc4b55b67968fb449d469e95b57f3451adeb509a5909865 +size 398486 diff --git a/paper/RLVRAMBench_IEEE_Access_submission.tex b/paper/RLVRAMBench_IEEE_Access_submission.tex index 7cca2844ef4aa0ac7e67f94ad23c4c1881282d21..6315f3ddd1fb38b692eb1280e26ce3e02cff17e1 100644 --- a/paper/RLVRAMBench_IEEE_Access_submission.tex +++ b/paper/RLVRAMBench_IEEE_Access_submission.tex @@ -29,6 +29,14 @@ \patchcmd{\@maketitle} {{\vss\doifont Digital Object Identifier\space\@doi\vss}\vspace*{7.4mm}\par} {}{}{\PackageError{rlvrambench}{Cannot suppress the publisher DOI field}{}} +\patchcmd{\thebibliography} + {\section*{\refname}}{\section*{\refname}\phantomsection} + {}{\PackageError{rlvrambench}{Cannot set the bibliography navigation anchor}{}} +% Keep the two short biographies compact; the vendor class remains unchanged. +\patchcmd{\IEEEbiographynophoto} + {\vskip 4\baselineskip plus 1fil minus 0\baselineskip} + {\vskip 1.5\baselineskip} + {}{\PackageError{rlvrambench}{Cannot set local biography spacing}{}} \def\rlvramfooter{\hbox to\textwidth{% {\footervolfont RLVRAMBench}\hfill% {\footerpagefont\thepage}}} @@ -49,109 +57,98 @@ \begin{document} \history{} \doi{} -\hypersetup{pdftitle={RLVRAMBench: A Benchmark for Memory Feasibility in Colocated Language-Model Reinforcement Learning},pdfauthor={Ond\v{r}ej Kobza, Jan \v{S}ediv\'{y}}} -\title{RLVRAMBench: A Benchmark for Memory\\ -Feasibility in Colocated Language-Model\\ -Reinforcement Learning} +\hypersetup{pdftitle={Memory Feasibility in Colocated Language-Model Reinforcement Learning: A Failure-Aware Measurement Study},pdfauthor={Ond\v{r}ej Kobza, Jan \v{S}ediv\'{y}}} +\title{Memory Feasibility in Colocated\\ +Language-Model Reinforcement Learning:\\ +A Failure-Aware Measurement Study} \author{\uppercase{Ond\v{r}ej Kobza}\authorrefmark{1}, AND \uppercase{Jan \v{S}ediv\'{y}}\authorrefmark{1}} \address[1]{Czech Institute of Informatics, Robotics and Cybernetics, Czech Technical University in Prague, Jugosl\'{a}vsk\'{y}ch partyz\'{a}n\r{u} 1580/3, 160 00 Prague 6, Czech Republic} \tfootnote{ORCID: Ond\v{r}ej Kobza: \href{https://orcid.org/0000-0002-0529-9860}{0000-0002-0529-9860}; Jan \v{S}ediv\'{y}: \href{https://orcid.org/0000-0003-0626-2303}{0000-0003-0626-2303}} \markboth{Kobza and \v{S}ediv\'{y}: RLVRAMBench}{Kobza and \v{S}ediv\'{y}: RLVRAMBench} \begin{abstract} -Sharing graphics processors between language-model response generation and -training reduces hardware requirements but complicates configuration -decisions: a setting that completes one stage can fail in another. -We introduce \textbf{RLVRAMBench}, a benchmark for deciding which training -configurations complete within a specified memory margin and how much -testing those decisions require. It combines repeated outcomes, -failure records, source-to-target prediction tasks, and budgeted -configuration tests evaluated on separate random seeds. We study -low-rank adaptation with Group Relative Policy Optimization in a fixed -colocated execution stack on A100 hardware. -In a small prospective panel, copying existing labels identified every -usable candidate without new target tests; a stricter memory-margin -requirement rejected usable candidates without an observed safety benefit. -This result assumes that source measurements already exist. In contrast, -transferring labels between hardware allocations admitted configurations -that failed, showing why coverage and approval errors must be reported -together. Stage-level controls reveal training-memory growth concealed -by the overall peak, while longer executions expose margin crossings -missed by individual short tests. Public measurements and reconstruction -software make these bounded findings inspectable. The contribution is -an executable evaluation protocol and evidence suite for comparing -memory-feasibility decisions, not a universal safety predictor. +Colocating language-model generation and training places different memory +demands on the same graphics processors. We present a failure-aware +measurement study of low-rank adaptation with Group Relative Policy +Optimization in a fixed training-and-generation implementation on A100 hardware. +The study distinguishes completion within a chosen memory margin, +completion above it, and memory failure. Releasing generation state +allowed several tested configurations to complete when retaining it did +not. Same-device controls found larger training-stage peaks with larger +per-device training batches, while the overall peak changed little; logging affected +the measured magnitude. Individual short tests missed some later margin +crossings, whereas the labels combining repeated tests agreed. +Earlier labels agreed in a small workload-screening panel, but copying +across GPU counts or to a larger model admitted failures. +In held-out-family comparisons, neither empirical +memory regression nor a learned classifier uniformly improved on copying +earlier labels. The released +dataset, RLVRAMBench, links outcomes to execution stages, retains failed +and incomplete attempts, and supports reconstruction and comparison of +configuration decisions. \end{abstract} \begin{keywords} -colocated execution, failure analysis, GPU memory, Group Relative Policy Optimization, performance benchmarking, reproducibility +colocated execution, failure analysis, GPU memory, Group Relative Policy Optimization, measurement study, reproducibility \end{keywords} \titlepgskip=-21pt \maketitle \section{Introduction}\label{introduction} -Reinforcement learning for large language models alternates between -generating responses and updating the model that generated them. The -generated responses are called \emph{rollouts}, and the model being trained -is the \emph{policy}, also called the actor. Rewards evaluate the generated -answers. A fixed reference policy has a different role: it assigns response -probabilities so that training can penalize excessive departure from the -initial policy. Weight -synchronization copies updated weights to the generation engine. Colocated -execution assigns these stages to the same graphics processing units -(GPUs), allowing training and generation to share hardware. - -Sharing hardware does not make memory demand constant. The generation engine reserves space for model state and intermediate results; some of that state may remain in memory during training or weight transfer. A configuration that completes generation may therefore run out of memory during a later stage. Users need to know not only how much memory a successful run consumed, but also which configurations failed and where. An out-of-memory (OOM) failure is part of the answer, not a missing measurement to discard. - -This paper studies \emph{memory feasibility}: whether a configuration -completes its requested work, with memory failures distinguished from other -execution problems. We additionally distinguish completion that leaves a -specified memory margin. The labels combine runs that complete or have a -diagnosed memory failure. Other failed or incomplete attempts remain in a -separate record, including when a later replacement supplies the memory -result. The labels therefore describe memory outcomes, not the probability -that every job will finish. Nor do they assess output correctness or -training convergence. -Throughout the paper, \emph{usable} means satisfying this memory -criterion, not being optimal for model quality or throughput. - -RLVRAMBench evaluates these questions for Group Relative Policy Optimization (GRPO) with low-rank adaptation (LoRA). GRPO learns from the relative rewards of responses to the same input prompt, while LoRA limits training to small adapter matrices rather than updating all model weights. The implementation uses VERL, a framework for language-model reinforcement learning, with vLLM as its response-generation engine. We hold this execution design fixed and study the configuration choices available to its users. - -The intended user is a researcher or engineer evaluating a candidate -training configuration. For example, a setting measured on four GPUs may -need to run on two. Copying its earlier label is inexpensive, but can -approve a target that fails. RLVRAMBench makes such decisions testable: -it specifies which source measurements a method may use, compares its -predictions with repeated target measurements, and distinguishes harmful -approvals from rejected usable settings. -A \emph{screen} is a short execution test of a candidate. When source -evidence is insufficient, a method can request such tests within a -measurement budget. - -The contribution is a measurement protocol and evidence suite for this -decision problem, supported by three complementary investigations: +Reinforcement learning for language models alternates between generating +responses, called \emph{rollouts}, and updating the \emph{policy}, or actor, +that generated them. Rewards evaluate the answers. In the setup studied +here, a fixed reference +policy supplies response probabilities used to penalize departure from +the initial policy. Weight synchronization copies updated weights to the +generation engine. \emph{Colocated execution} assigns generation and +training to the same graphics processing units (GPUs). + +Memory demand changes between these stages. State retained by the +generation engine can remain present during training or weight transfer, +so a configuration that completes generation may fail later. A useful +measurement must therefore record out-of-memory (OOM) failures as well +as successful-run peaks. We study \emph{memory feasibility}: completion +of the requested work, with a separate label for completion that leaves +a specified memory margin. The objective is resource feasibility, not +model quality or training speed. + +The study uses synchronous Group Relative Policy Optimization (GRPO) +with low-rank adaptation (LoRA) in VERL, a language-model reinforcement +learning framework, and its vLLM generation engine. GRPO learns from the +relative rewards of responses to the same prompt; LoRA trains small +adapter matrices while keeping the base weights fixed. We examine +configuration choices within this implementation on A100 GPUs. + +The intended user is a researcher adapting an earlier configuration: +for example, moving a setting measured on four GPUs to two. Earlier +measurements may guide this decision, but a target execution is needed +to check whether their labels transfer. A \emph{screen} is a short +execution test of a candidate. Our comparisons examine what existing +measurements, fixed predictors, and additional screens establish. +We count within-margin candidates recovered, failing or above-margin +candidates approved, and the execution evidence required for each +decision. Three investigations support this comparison: \begin{enumerate} \def\labelenumi{\arabic{enumi}.} \tightlist \item - \textbf{Repeated configuration outcomes and transfer tasks.} A common - configuration grid across three model families and four workloads - distinguishes within-margin completion, completion above the margin, - and memory failure. Explicit source--target tasks expose both the - simplicity of workload transfer on this grid and the different errors - made when transferring between models or GPU counts. + \textbf{Failure-inclusive configuration outcomes.} A selected + model--workload grid distinguishes within-margin completion, + above-margin completion, and memory failure. \item - \textbf{Tests of what a memory summary leaves out.} State-release and - stage-level controls show how success-only records and whole-run peaks - can conceal relevant outcomes. Repeated short and longer executions - distinguish an individual successful test from evidence supporting a - repeated configuration label. + \textbf{Tests of what memory summaries omit.} Paired state-release + comparisons and measurement controls expose completion changes and + training-stage growth hidden by whole-run peaks. \item - \textbf{An executable, traceable evaluation.} Public task inputs, - scoring software, and linked process and attempt records support - method comparison and reconstruction. Failures and unresolved attempts - remain visible rather than disappearing from successful-run summaries. + \textbf{Limits of reuse and short testing.} Transfer comparisons, + fitted prediction baselines, a prospective decision case study, and + paired short/long executions measure agreement and decision errors + under changed settings and schedules. \end{enumerate} +RLVRAMBench releases the measurements, attempt inventory, and +evaluation software supporting these investigations. + \subsection{Research questions}\label{research-questions} The study addresses three research questions (RQs): @@ -159,98 +156,86 @@ The study addresses three research questions (RQs): \begin{itemize} \tightlist \item - \textbf{RQ1 --- Feasibility and transfer:} How do resource labels - transfer between settings, and what within-margin coverage and approval - errors do existing measurements support under a target-screen budget? + \textbf{RQ1 --- Outcomes and transfer:} Which tested settings complete + within margin, and how well do reused labels and fixed predictors + identify them when the setting changes? \item - \textbf{RQ2 --- Memory mechanisms:} How do retained generation state, training micro-batch size, and generation-memory reservation affect completion and stage-specific memory use? + \textbf{RQ2 --- Stage measurements:} What do state-release and + batching comparisons reveal beyond the whole-run peak? \item - \textbf{RQ3 --- Repeatability over time:} Do safety labels agree across fresh processes and remain valid during longer training, validation, and checkpoint writing? + \textbf{RQ3 --- Execution schedule:} How do labels from short and + longer executions agree when validation and checkpoint writing are included? \end{itemize} \section{Related work}\label{related-work} -We distinguish systems that organize execution, studies of memory -mechanisms, and methods for evaluating configuration decisions. - -\subsection{Execution and resource sharing in language-model training}\label{execution-and-resource-sharing-in-language-model-training} - -Systems for reinforcement learning from human feedback (RLHF) and related post-training methods must coordinate several models and execution stages. DeepSpeed-Chat combines training and inference optimizations \cite{yao2023deepspeedchat}. HybridFlow describes these stages as a dataflow and redistributes actor state between training and generation \cite{sheng2024hybridflow}. RLHFuse combines work within and across stages to improve utilization \cite{zhong2024rlhfuse}. OpenRLHF supports several placements through distributed scheduling \cite{hu2024openrlhf}, while ReaL searches plans that reallocate model parameters during execution \cite{mei2024realhf}. - -A second group changes the execution relationship between generation and training. AReaL permits asynchronous progress \cite{fu2025areal}, and RollArt separates rollout execution for multi-task, multi-turn training \cite{tan2026rollart}. RobustRL addresses recovery after machine faults rather than memory limits within a working allocation \cite{chen2026robustrl}. - -Outside reinforcement learning, SIRIUS shares memory between response-time-sensitive inference and background training, adjusting the training batch to protect inference service \cite{wang2025sirius}. Such resource-sharing systems motivate measurements that distinguish an individual stage's demand from the overall peak. - -RLVRAMBench measures completion and memory-margin compliance within a fixed colocated system. Repeated configuration labels and their transfer are central measurements, rather than recovery from machine faults or improved scheduling. - -\subsection{Memory management during generation and training}\label{memory-management-during-generation-and-training} - -Models process text as \emph{tokens}, the units produced by their tokenizer. During generation, a key--value (KV) cache stores intermediate attention results for previously generated tokens. vLLM's PagedAttention manages this cache in blocks to reduce fragmentation, or unusable gaps in allocated memory \cite{kwon2023pagedattention}. Its Sleep Mode can move model state out of GPU memory or discard cached state between generation periods \cite{vllm2026sleepmode}. Hydra-RLHF reduces the memory required by Proximal Policy Optimization (PPO), another policy-training algorithm, through shared base models and selectable LoRA adapters \cite{santacroce2023hydra}. +\subsection{Execution and memory management}\label{execution-and-resource-sharing-in-language-model-training} +\label{memory-management-during-generation-and-training} + +Systems for reinforcement learning from human feedback (RLHF) coordinate +generation, scoring, and updates through different placements of model +state. DeepSpeed-Chat combines training and inference optimizations +\cite{yao2023deepspeedchat}; HybridFlow redistributes actor state between +these stages \cite{sheng2024hybridflow}; and RLHFuse combines work within +and across stages \cite{zhong2024rlhfuse}. ReaL searches parameter +reallocation plans \cite{mei2024realhf}. These systems change execution +to improve efficiency, whereas our study measures configuration outcomes +within one execution design. + +Memory management also changes what remains resident between stages. +vLLM stores cached attention state from earlier tokens in blocks +\cite{kwon2023pagedattention} and provides Sleep Mode to move weights +to host memory or discard cached state +\cite{vllm2026sleepmode}. Hydra-RLHF shares base weights across selectable +adapters to reduce policy-training memory \cite{santacroce2023hydra}. Zhou et al.\ study memory use in DeepSpeed-Chat and ColossalChat, including -LoRA and state retained across inference and training -\cite{zhou2024rlhfmemory}. They evaluate releasing unused cache from the -framework's memory manager, or \emph{allocator}, between stages and show -that a training-memory control can leave the overall peak unchanged when -inference determines that peak. Neither LoRA nor this masking effect is -new to the present study. Our addition is to connect such measurements -to repeated three-state configuration labels, explicit transfer decisions, -and execution-horizon tests, with controls for stage-measurement -sensitivity. The vLLM state offload and discard tested here also differs +LoRA, retained state, and release of unused cache from the framework's +memory manager, or \emph{allocator} +\cite{zhou2024rlhfmemory}. They show that a training-memory control can +leave the overall peak unchanged when inference sets that peak. +Our contribution builds on this observation: failure-inclusive GRPO +outcomes, paired generation-state release, stage-measurement controls, +and tests of label agreement across settings and schedules. The vLLM +weight offload and cache discard studied here is a different intervention from releasing unused PyTorch allocator cache. -\subsection{Failure-aware configuration measurement}\label{failure-aware-configuration-measurement} - -SLO-Guard includes crashes when tuning standalone vLLM serving under service-level objectives (SLOs), such as response-time targets \cite{lysenstoen2026sloguard}. RLVRAMBench applies failure-inclusive measurement to synchronous GRPO, where a run can fail before its full memory peak is observed. We retain the failure location, combine repetitions before assigning a configuration label, and test transfer between complete model--workload settings. Our budgeted comparison evaluates fixed admission rules, which decide whether to use each candidate configuration. It does not introduce an adaptive search algorithm or optimize training throughput. - -\begin{table*}[!t] -\centering -\caption{Scope of the closest systems and measurement work.}\label{tab:prior} -\footnotesize -\setlength{\tabcolsep}{3pt} -\renewcommand{\arraystretch}{1.12} -\begin{tabular}{@{} - >{\raggedright\arraybackslash}p{(\linewidth - 6\tabcolsep) * \real{0.2500}} - >{\raggedright\arraybackslash}p{(\linewidth - 6\tabcolsep) * \real{0.2500}} - >{\raggedright\arraybackslash}p{(\linewidth - 6\tabcolsep) * \real{0.2500}} - >{\raggedright\arraybackslash}p{(\linewidth - 6\tabcolsep) * \real{0.2500}}@{}} -\toprule\noalign{} -\begin{minipage}[b]{\linewidth}\raggedright -Work -\end{minipage} & \begin{minipage}[b]{\linewidth}\raggedright -Evaluated setting -\end{minipage} & \begin{minipage}[b]{\linewidth}\raggedright -Primary intervention or objective -\end{minipage} & \begin{minipage}[b]{\linewidth}\raggedright -Distinction from this benchmark -\end{minipage} \\ -\midrule\noalign{} -HybridFlow / RLHFuse \cite{sheng2024hybridflow, zhong2024rlhfuse} & Distributed RLHF & Redistributing state and combining stages & Changes execution to improve efficiency \\ -SIRIUS \cite{wang2025sirius} & Independent inference and training & Memory handover under service constraints & Shares memory between separate workloads \\ -RLHF memory study \cite{zhou2024rlhfmemory} & DeepSpeed-Chat and ColossalChat & State retention, cache release, and stage-dependent peaks & Establishes mechanisms tested here with repeated, failure-inclusive labels \\ -SLO-Guard \cite{lysenstoen2026sloguard} & Standalone language-model serving & Configuration tuning that includes crashes & Selects settings under service constraints \\ -RobustRL \cite{chen2026robustrl} & Post-training machine failures & Recovery of training and generation services & Studies machine faults rather than memory limits \\ -This work & Fixed colocated LoRA GRPO & Repeated, failure-inclusive measurement & Tests configuration safety and its transfer \\ -\bottomrule -\end{tabular} -\end{table*} - -Table~\ref{tab:prior} distinguishes changes to execution from the -configuration outcomes evaluated here. - -\subsection{Repeatability and few-run uncertainty}\label{repeatability-and-few-run-uncertainty} - -Deep reinforcement-learning evaluations can be sensitive to implementation details, randomness, and small run counts \cite{henderson2018matters}. Uncertainty estimates are therefore important even when average results appear consistent \cite{agarwal2021precipice}. We apply this principle to memory measurements: configurations are specified in advance, each repetition starts in a fresh process, and comparisons preserve the grouping of runs within workloads. - -\section{RLVRAMBench}\label{rlvrambench} - -RLVRAMBench evaluates predictions and admission decisions against -measured target outcomes. This section defines the execution being -measured, its labels, and the information a method may use. - -The experiments study synchronous LoRA GRPO on NVIDIA A100 GPUs with 40 GB of memory. Generation and training share the same devices rather than running as separate services. Most experiments use two GPUs; a follow-up study uses four to test whether the observed labels transfer between allocation sizes. - -The outcomes concern memory use and completion, not reward or model accuracy. The measured boundaries apply to the selected models, workloads, and software versions. Full-parameter training, other learning algorithms, other accelerators, and alternative execution designs require separate evaluation. +\subsection{Memory estimation and configuration decisions}\label{failure-aware-configuration-measurement} + +Analytical estimators predict memory before training. DNNMem accounts +for the computation graph and framework runtime, motivated in part by +jobs that fail for lack of GPU memory \cite{gao2020dnnmem}. LLMem uses +decoder-model structure and the memory distribution of distributed +fine-tuning methods to estimate their peaks and compare configurations +\cite{kim2024llmem}. For RLHF, ReaL combines static memory---gradients +and optimizer state---with active memory, including attention caches, +intermediate layer outputs (activations), and parameters moved between +roles. It estimates each function call +and takes the maximum over the iteration and devices to assess a plan's +feasibility \cite{mei2024realhf}. Component accounting and the stage-wise +maximum thus have direct precedents. These estimators use model and +execution information; label transfer instead reuses measured outcomes +from a specified source setting. + +SLO-Guard includes crashes when tuning standalone vLLM serving under +service-level objectives, such as response-time targets +\cite{lysenstoen2026sloguard}. Here, failures may arise during generation, +updates, or weight transfer. We compare fixed rules for approving +configurations, including empirical memory regression and a learned +classifier, and preserve failed attempts alongside completed peaks. +The measured objective is completion and margin compliance, rather than +service latency or an optimized search strategy. +A direct comparison with the established memory estimators would require +implementing and validating their execution and profiling assumptions +for this stack. Here we compare fixed empirical baselines fitted from +the released configuration records; this does not test the accuracy +of DNNMem, LLMem, or ReaL. + +\section{Study setting and measurements}\label{rlvrambench} + +This section defines the execution, outcome labels, and comparisons +recorded in RLVRAMBench. Most experiments use two A100 GPUs with 40 GB +of memory; a four-GPU study tests transfer between allocation sizes. \subsection{Execution and batching}\label{cases-and-configurations} @@ -281,6 +266,10 @@ workers, whose state can remain present between generation periods. \subsection{Cases and configuration controls}\label{configuration-factors} +Models process text as \emph{tokens}, the units produced by a tokenizer. +During generation, a key--value (KV) cache stores attention results for +previous tokens. + A \emph{workload} specifies the task data, prompt and response length limits, responses per prompt, and global batch. A \emph{case} combines that workload with a model, GPU count, and requested execution duration @@ -288,7 +277,7 @@ and operation schedule. A \emph{configuration} selects the memory and micro-batching controls within that case. This hierarchy keeps a new random seed distinct from a new workload or execution setting. -The vLLM reservation specifies the fraction of device memory available to +The vLLM reservation is the requested fraction of device memory assigned to the generation engine. Length limits and the maximum number of active sequences further constrain its work. \emph{Offloading} moves state to host memory. Actor parameter and optimizer offload are separate controls @@ -303,8 +292,11 @@ and restoration behavior. Each planned \emph{slot} specifies a configuration and random seed, with a fresh process for each repetition. A complete configuration label requires eligible memory outcomes for all three slots. An eligible -outcome either completes the requested work or has a diagnosed memory -failure; the rules for other attempts appear below. +outcome either completes the requested work or has a diagnosed GPU-memory +failure. This includes the generation engine rejecting its requested +device-memory budget, not only an allocation failure. Host-memory exhaustion +and other unresolved execution failures do not supply GPU-memory labels. +We use the shorter term \emph{memory failure} for this GPU-specific class. Memory is reported in mebibytes (MiB, \(2^{20}\) bytes) and gibibytes (GiB, \(2^{30}\) bytes). The operating limit is 38,912 MiB, leaving @@ -319,65 +311,49 @@ failure point. The three configuration labels are: \item \emph{Memory failure}: at least one eligible repetition has a diagnosed memory failure. \end{itemize} -A missing eligible slot leaves the repeated label unresolved. We use -\emph{safe} as shorthand for within-margin completion only. An approval -of either other state is a \emph{false-safe} decision; rejecting a -within-margin configuration is a \emph{false-unsafe} decision. +A missing eligible slot leaves the repeated label unresolved. +For label predictors, only a within-margin prediction counts as approval. +An approval error admits an above-margin or memory-failing configuration; these two +errors are counted separately. The artifact's shorthand \emph{safe} +denotes within-margin completion only. Peak memory takes the maximum over sampled times, GPUs, and repetitions; -runtime averages completed repetitions. The highest observed safe +runtime averages completed repetitions. The highest observed within-margin configuration level summarizes the tested boundary, without assuming -that untested settings are safe. Levels are test indices, not rankings +the same outcome for untested settings. Levels are test indices, not rankings of throughput or training quality. -\subsection{Benchmark tasks and evaluation}\label{benchmark-tasks-and-evaluation} +\subsection{Configuration comparisons}\label{benchmark-tasks-and-evaluation} A task supplies the target configuration's settings and measurements from a specified source case, or \emph{donor}. The method predicts the target's three-state label; target measurements are reserved for scoring. -Consider Qwen2.5-3B on GSM8K with actor micro-batch 16 and generation -reservation 0.80. Its four-GPU measurements support within-margin -completion. Copying that label approves the corresponding two-GPU -setting, but the measured target fails. This is one concrete decision -the benchmark can check, not a recommendation to run that setting. - -The released tasks test transfer across models, workloads, GPU counts, and execution horizons. Model transfer holds workload and configuration level fixed; workload transfer holds model and level fixed. GPU-count transfer uses the matched two- and four-GPU settings. Horizon transfer uses the paired one-step and 100-step follow-up cases. Whole source and target cases remain distinct: another seed or a supporting control from the target case is not additional permitted source evidence. The source-label baseline copies the outcome of the matched source configuration. +The source-label baseline copies the matched donor configuration's outcome. +Comparisons change model, workload, GPU count, or execution horizon while +holding the other specified settings fixed. Whole cases remain separate: +another target seed or a supporting target control is not permitted donor +evidence. Scoring reports three-state accuracy together with within-margin -\emph{recall}, the fraction of usable targets approved, and approval -\emph{precision}, the fraction of approved targets that are usable. -Memory-failure approvals and above-margin approvals remain separate. -Counts, denominators, rejected usable settings, and the full confusion -matrix accompany these rates; an empty denominator is undefined. -No one of these rates establishes a useful configuration decision. - -The budgeted task lets a rule request target screens before deciding. -It starts from permitted source evidence and approves, rejects, or -abstains on each candidate. Abstention withholds approval without -asserting failure. Separate target repetitions then evaluate the -decisions using the same coverage and approval-error measures. -Section~\ref{prospective-admission} specifies the independent seed roles, -query order, and cost accounting. - -The original transfer suite has 52 tasks and 400 queries over 94 distinct -target configurations. Workload transfer contributes 216 queries, model -transfer 144, GPU-count transfer 36, and horizon transfer four. These -reused targets are not independent experiments. Results are reported by -task and transfer type. The static interface counts the distinct donor -processes and recorded attempts supplied as evidence; this is not a -measurement of how much computation a method saves. - -The labels are public, so these are inspectable tasks rather than a hidden -leaderboard. Tasks must be processed separately: a target in one task can -be a source in another, and pooling their inputs would reveal target -outcomes. Comparisons must disclose additional evidence. Workload transfer -is already solved by lookup on this grid; the repeated horizon labels -also agree in all four cases. These reference checks should not dominate -a claim about prediction difficulty or generalization. +\emph{recall}, the fraction of within-margin targets approved, and +\emph{precision}, the fraction of approvals that are within margin. +Counts of memory-failure approvals, margin-violating approvals, and +rejected within-margin settings accompany these rates. An empty +denominator is undefined. + +The budgeted comparison also permits requested target screens. +A rule approves, rejects, or abstains; abstention withholds approval +without asserting failure. Separate repetitions evaluate its decisions +(Section~\ref{prospective-admission}). + +The released tasks reuse targets and expose their labels publicly. +They are inspectable comparisons, not independent experiments or a hidden +test set. Supplementary Section S13 specifies the task inventory and +input-separation contract. \subsection{Measurements}\label{measurements} -Each trial records the configuration, random seed, software environment, elapsed time, exit status, and any failure. Experiment identifiers link these records to jobs submitted through Slurm, the cluster's job scheduler. +Each invocation records the configuration, random seed, software environment, elapsed time, exit status, and any failure. Experiment identifiers link these records to jobs submitted through Slurm, the cluster's job scheduler. External sampling uses the NVIDIA Management Library (NVML), which exposes device-level memory use and utilization, at 100 ms intervals. Host-side @@ -387,7 +363,7 @@ receive the current marker. Markers do not synchronize the GPUs: stage attribution follows the observed execution interval, not the lifetime of each allocation. Unmarked intervals remain idle or unknown. -Framework allocator traces record how PyTorch requests and manages memory. +Allocator traces record PyTorch tensor allocations and retained memory. Allocated tensor memory is not the same quantity as total device memory. A stage peak restricts the external maximum to that stage's marked interval. Whole-run peak comparisons use completed runs: a failed run's @@ -403,9 +379,13 @@ programming problems \cite{li2022alphacode}. Their length limits and responses per prompt define the requested work, not the lengths of every sequence actually generated. -For CodeContests, preparation pools its original validation and test records, excludes problems requiring file input/output or lacking test examples, and orders the remaining prompts by their tokenized length using Qwen2.5-1.5B-Instruct \cite{qwen2024qwen2515b}. The shortest and longest 128 eligible prompts form disjoint \emph{standard code} and \emph{longer-prompt code} subsets. Alternating records in each ordered subset form 64 training and 64 validation examples. The longer subset is selected by length, not by fitting a tail distribution. - -Preparation profiles use Qwen tokenizers and therefore do not describe the exact batches seen by every model. Actual runs apply the selected model's tokenizer and remove prompts above the workload limit. Section~\ref{experimental-coverage} reports these filtering counts and the realized lengths in completed boundary runs. The model comparison consequently holds the workload definition fixed, not the exact token sequences or, after filtering, necessarily the same examples. +The \emph{standard code} and \emph{longer-prompt code} subsets select +opposite ends of the same CodeContests prompt-length ordering; they are +not independent corpora. Preparation uses Qwen tokenizers, whereas each +run filters prompts using its selected model's tokenizer. Thus, a matched +workload need not supply identical token sequences or examples across +models. Supplementary Section S12 gives the selection recipe, filtering +counts, and realized lengths. For GSM8K, the reward extracts a formatted final answer and compares it with the recorded answer. For MATH, it extracts the final boxed expression @@ -413,68 +393,50 @@ and compares normalized answer strings. For code, the reward checks Python syntax and the presence of input/output operations without executing the generated program or using the CodeContests correctness tests. These functions provide the rewards needed to execute GRPO; -the benchmark does not validate their effect on answer quality. +we do not evaluate their effect on answer quality. \subsection{Failures, eligibility, and attempts}\label{failure-taxonomy} \label{artifact-validity-and-retry-policy} -An \emph{attempt} is an invocation of a slot's runner; a queued job that -never invokes it is not a process attempt. We distinguish diagnosed -setup errors, diagnosed memory failures, and other or unresolved -noncompletion. Setup errors include missing dependencies, incorrect GPU -visibility, and broken instrumentation. Memory failures remain outcomes -at the stage supported by the log and trace. An allocator error alone -does not establish fragmentation as its cause. +An \emph{attempt} is an invocation of a slot's runner, not merely a queued +job. We distinguish setup errors, diagnosed memory failures, and other +or unresolved noncompletion. Failure stages follow the logs and traces; +an allocator error alone does not show that free memory split into +unusable pieces, or \emph{fragmentation}, caused the failure. -An eligible outcome requires a distinct trial record, environment record, -execution log, stage trace, and per-GPU measurements. Completion requires +Eligibility requires a trial record, environment record, execution log, +stage trace, and per-GPU measurements. Completion additionally requires a successful exit and evidence of the requested final work. A nonzero -exit enters the memory corpus when its evidence establishes a memory -failure. All other attempts retain their observed status and exclusion -reason. A missing trial record establishes neither completion nor OOM. - -Allocator logging begins when the instrumented worker starts. Early engine failures can therefore occur before any allocator event is recorded. We assess them using the execution log, external measurements, stage trace, and error stack. Allocator traces are also absent by design in the control that disables allocator instrumentation. - -Historical repairs repeat a slot without changing its scientific -settings. After the documented exclusions, selection requires exactly -one eligible outcome. An eligible replacement supplies that slot's -memory result; it does not erase an earlier attempt or establish its -cause. Historical repairs followed setup errors and some later -noncompletions. The admission panel uses the stricter predeclared policy -in Section~\ref{prospective-admission}. Intentional seed repetitions -are distinct slots, not retries. - -One historical memory-failure record is incomplete because its launcher -was changed during execution. Its identical replacement also failed, so -including the known earlier failure would not change the label. -Unresolved timeouts and late noncompletions are retained separately. -Supplementary Section S1 details these histories; the result sections -report the attempt denominators relevant to each comparison. +exit enters the memory analysis only with diagnosed memory failure. +Other attempts retain their status and exclusion reason. Early failures +and external-only controls can lack allocator traces; their remaining +evidence determines eligibility. + +Historical repairs repeated unchanged scientific settings after setup +errors or some later noncompletions. Exactly one eligible outcome is +selected per slot after documented exclusions; earlier attempts remain +in the ledger. Repeated seeds are separate slots, not retries. The +result sections report relevant attempt denominators, and Supplementary +Section S1 records the repair histories, including an incomplete +memory-failure record. The labels are conditional memory outcomes, +not estimates of all-cause job completion. \subsection{Experiment planning and selection}\label{experiment-planning-and-selection} -Pilot runs developed configuration ranges and instrumentation. Subsequent -\emph{experiment matrices} fixed settings and seeds before execution. -They and their dated amendments are development records, not independent -preregistrations. Historical pilot outcomes do not enter the repeated -estimates. - -Restriction to standard GRPO was a post-outcome scope decision after -exploratory work with a custom objective. Those excluded observations -remain archived but support none of this paper's counts or conclusions. -The original transfer baselines, stage-specific reanalysis, margin -sensitivity, and attempt/workload audits are retrospective. Later -controls were specified before their own execution, using earlier -results to select cases. Supplementary Section S1 records this sequence -without treating successive controls as independent replications. +Pilots developed the ranges and instrumentation; subsequent matrices +fixed settings and seeds before execution. Restriction to standard GRPO +followed exploratory outcomes with a custom objective. Those observations +and pilots remain archived but do not enter the reported estimates. +The original transfer baselines, stage reanalysis, and margin and +attempt audits are retrospective. Later controls were specified before +their own execution using earlier results to select cases. Supplementary +Section S1 records this sequence and its amendments. \section{Experimental protocol}\label{experimental-protocol} -The design separates three questions: whether resource labels transfer, -why measured memory changes, and whether a short test remains informative -over time. Table~\ref{tab:design} maps those questions to the comparisons. -The mechanism and instrumentation controls explain measurements; they -are not extra target evidence available to a prediction method. +Table~\ref{tab:design} links the questions to their comparisons. +Measurement controls support interpretation of the traces; they are +not additional target evidence available to prediction methods. \begin{table*}[!t] \centering @@ -486,9 +448,10 @@ are not extra target evidence available to a prediction method. Question & Comparison & What changes & Primary evidence \\ \midrule RQ1 & Boundary and transfer & Model, workload, or GPU count & Repeated three-state labels and directional approval errors \\ +RQ1 & Estimation and matched GPU count & Held-out model family, then a larger model on matched allocations & Frozen predictions, approval errors, and paired outcomes \\ RQ1 & Budgeted admission & Available donor evidence and target-screen budget & Within-margin recall and approval errors on separate evaluation seeds \\ RQ2 & State release & Generation state released or resident on the same GPUs & Paired completion and completed-run peaks \\ -RQ2 & Batch and reservation & Actor micro-batch and generation reservation & Whole-run versus actor-stage effects \\ +RQ2 & Batch and reservation & Actor micro-batch and generation reservation & Whole-run versus actor-stage contrasts \\ Measurement & Sampling and allocator logging & Sampling interval or logging condition & Label agreement, peak differences, and incomplete attempts \\ RQ3 & Short and longer execution & Requested steps and scheduled operations & Individual and repeated label agreement; later peak growth \\ \bottomrule @@ -499,81 +462,43 @@ RQ3 & Short and longer execution & Requested steps and scheduled operations & In The experiments ran on Karolina, a supercomputer operated by IT4Innovations National Supercomputing Center in Ostrava, Czech Republic \cite{it4i2026karolina}. Its accelerated nodes contain eight NVIDIA A100 GPUs; our jobs used allocations of two or four A100-SXM4-40GB devices. Each device reported 40,960 MiB of memory, with driver version 610.43.02. -The container environment used Python 3.12.3, PyTorch 2.10.0 for model -computation, NVIDIA's CUDA execution platform version 12.9, and vLLM 0.18.0 -for generation. The VERL source includes measurement instrumentation and -documented compatibility repairs. Thus, the execution design is held fixed, -but the source is not an unmodified upstream release. Environment -records retain repository commits, working-copy status, package versions, -and the container checksum. Later controls use immutable source copies. -Some historical runs used modified working trees whose per-run changes -were not captured as patches; their commit IDs alone do not fully -identify the executing source. This limitation does not prevent -reconstruction of the analyses from their saved measurements. +The container used Python 3.12.3, PyTorch 2.10.0, NVIDIA's CUDA execution +platform 12.9, and vLLM 0.18.0. VERL includes measurement instrumentation +and documented compatibility repairs. Environment records preserve +versions, source identifiers, working-copy status, and the container +checksum. Later controls use immutable source copies; the historical +source-record limitation is stated in Section~\ref{reproducibility}. \subsection{Training setup and configuration levels}\label{training-setup-and-configuration-levels} -All reported runs use the standard GRPO estimator \cite{shao2024deepseekmath} +All reported runs use standard GRPO training \cite{shao2024deepseekmath} with LoRA adapters \cite{hu2022lora}. The primary allocation places one vLLM generation worker on each of two GPUs; a worker does not split its model across GPUs. Fully Sharded Data Parallel (FSDP) partitions the actor's -training state across the allocation. Reference scoring uses the same -frozen base weights, as described in Section~\ref{cases-and-configurations}. +training state across the allocation. The boundary study uses Qwen2.5-3B-Instruct \cite{qwen2024qwen253b}, Phi-4-mini-instruct \cite{microsoft2025phi4mini}, and Granite-3.3-2B-Instruct \cite{ibm2025granite33}, with approximately 3.1, -3.8, and two billion parameters, respectively. State-release and original +3.8, and 2.5 billion stored floating-point parameters, respectively. +The names identify the released models; their nominal size labels are +not exact parameter counts. State-release and original temporal tests also use Qwen2.5-1.5B-Instruct. Tables omit the instruction-tuned suffix; prose uses Qwen, Phi, and Granite when the size is unambiguous. -Table~\ref{tab:training} lists the fixed training controls. Model parameters and generation use the 16-bit bfloat16 format (BF16), while gradient reduction uses 32-bit floating point (FP32). Gradient checkpointing recomputes intermediate activations during training to reduce stored state. Kullback--Leibler (KL) divergence penalizes departure from the reference policy; it is applied as an actor loss rather than added to the reward. - -AdamW is adaptive moment estimation with decoupled weight decay; its beta -parameters control running averages of gradients and their squares. -LoRA's rank sets the adapter dimension, and alpha scales its update. -Generation samples from token probabilities, whereas greedy validation -selects the most probable token. Temperature rescales these probabilities; -top-p and top-k restrict the probability mass and number of candidates, -respectively. The chosen values leave training sampling unrestricted. - -Adapters are attached to the model's linear transformations. Padding removal avoids computing on padding tokens introduced to align variable-length sequences. Policy clipping limits the probability-ratio change used in the update objective, and gradient clipping bounds the gradient norm. The KL loss uses VERL's low-variance estimate of divergence. None of these controls is varied in the experiments. - -\begin{table*}[!t] -\centering -\caption{Fixed training controls in the retained LoRA GRPO experiments.}\label{tab:training} -\footnotesize -\setlength{\tabcolsep}{3pt} -\renewcommand{\arraystretch}{1.12} -\begin{tabular}{@{} - >{\raggedright\arraybackslash}p{(\linewidth - 2\tabcolsep) * \real{0.5000}} - >{\raggedright\arraybackslash}p{(\linewidth - 2\tabcolsep) * \real{0.5000}}@{}} -\toprule\noalign{} -\begin{minipage}[b]{\linewidth}\raggedright -Fixed training control -\end{minipage} & \begin{minipage}[b]{\linewidth}\raggedright -Setting -\end{minipage} \\ -\midrule\noalign{} -LoRA & Rank 64; alpha 32; all linear transformations \\ -Gradient checkpointing & Enabled \\ -Padding removal & Enabled \\ -Actor FSDP mixed precision & BF16 parameters; FP32 gradient reduction \\ -vLLM inference precision & BF16 \\ -Optimizer & PyTorch AdamW; learning rate \(3 \times 10^{-6}\) \\ -Optimizer controls & Betas (0.9, 0.999); weight decay 0.01; gradient clip 1.0 \\ -Update schedule & Constant learning rate; no initial ramp; one pass over the update batch \\ -Policy clipping / micro-batch sizing & Clip ratio 0.2 / fixed sizes \\ -KL regularization & Actor loss enabled; coefficient 0.001; low-variance estimator \\ -KL in reward / entropy bonus & Disabled / no bonus encouraging varied token choices \\ -Data order & Fixed order; overlong prompts filtered \\ -Reference computation & Frozen actor base with adapters disabled; no separate model \\ -Training generation & Sampling enabled; temperature 1; top-p 1; top-k disabled \\ -Validation generation & Greedy decoding; one response per prompt \\ -\bottomrule -\end{tabular} -\end{table*} +Adapters of rank 64 and scale parameter alpha 32 attach to the selected +linear layers, excluding the output head. Training computations and +generation use 16-bit bfloat16 (BF16), while the sharded actor parameters +and gradient reduction use 32-bit floating point (FP32). The distinction +between stored parameters and computation precision matters when +estimating resident memory. +Gradient checkpointing recomputes intermediate activations to reduce +stored state, and padding removal skips padding tokens. Training samples +responses, while validation selects the most probable next token. +Kullback--Leibler (KL) divergence penalizes departure from the reference +policy through the actor loss, not the reward. Supplementary Section S11 +lists the unchanged optimizer, decoding, and regularization settings. Table~\ref{tab:workloads} gives the workloads used for the repeated boundary and batch/reservation experiments. Prompt and response lengths are upper limits; individual samples may be shorter. The log-probability micro-batch applies to policy and reference log-probability scoring. @@ -619,11 +544,11 @@ Longer-prompt code & 1,536 & 1,024 & 2,560 & 64 & 2 & 32 & 4 \\ \end{tabular} \end{table*} -The model-length cap bounds the combined prompt and response; the sequence cap bounds the number of simultaneously active generation sequences. Each boundary and batch/reservation process performs one training iteration. Section~\ref{experimental-coverage} reports preparation-subset sizes and filtering. Other experiments use the settings specified in their respective protocols below. +The model-length cap bounds the combined prompt and response; the sequence cap bounds the number of simultaneously active generation sequences. Each boundary and batch/reservation process performs one training iteration. Supplementary Section S12 reports preparation-subset sizes and filtering. Other experiments use the settings specified in their respective protocols below. \begin{table}[!t] \centering -\caption{Compound configuration levels. Their order increases the selected controls; it does not rank throughput or usefulness.}\label{tab:levels} +\caption{Compound configuration levels used in the boundary study.}\label{tab:levels} \footnotesize \setlength{\tabcolsep}{3pt} \renewcommand{\arraystretch}{1.12} @@ -653,13 +578,12 @@ c5 & 32 & 0.85 & Disabled \\ \end{tabular} \end{table} -Table~\ref{tab:levels} defines the six configuration levels, c0--c5. They jointly increase actor micro-batch size and generation-memory reservation, with actor parameter offload disabled above c1. They are ordered test settings, not equally spaced values of one variable, so the comparison cannot isolate the effect of an individual control. Optimizer offload is disabled and generation-state release is enabled throughout. +Table~\ref{tab:levels} defines the six configuration levels, c0--c5. They jointly increase actor micro-batch size and generation-memory reservation, with actor parameter offload disabled above c1. They are ordered test settings, not equally spaced values of one variable, so the comparison cannot isolate the effect of an individual control. Optimizer offload is disabled and generation-state release is enabled throughout the boundary grid. \subsection{Experimental coverage}\label{experimental-coverage} -We account separately for requested slots, actual attempts, and -realized sequence lengths. The first two establish which outcomes enter -each comparison; the latter checks what work the configured caps produced. +Table~\ref{tab:attempts} separates planned slots, actual attempts, and +eligible outcomes in the original studies. \begin{table*}[!t] \centering @@ -685,54 +609,29 @@ Four-GPU transfer & 54 & 57 & 54 & 0 & 3 \\ \end{tabular} \end{table*} -Before the prospective admission comparison, the release contains 612 -eligible processes from 683 recorded attempts: the 588 outcomes in -Table~\ref{tab:attempts}, plus 24 outcomes from 26 attempts in the targeted -batch-by-logging control. Sixteen supporting historical screens remain -separate, as do abandoned-scope observations. The original 69 exclusions -comprise two completed processes from invalid pairs, one incomplete -memory-failure record, 55 other recorded failures, and eleven attempts -without a trial record. The attempt inventory gives their evidence and -exclusion reasons. -The admission panel adds 48 eligible outcomes from 48 attempts, giving -660 eligible processes from 731 attempts across these studies. Its -separate tables preserve the original static benchmark's denominators. - -\begin{table*}[!t] -\centering -\caption{Realized work in the boundary experiment. Lengths are tokens; means give equal weight to completed processes. ``Cap batches'' counts completed batches with at least one response reaching its configured limit, not the fraction of individual responses.}\label{tab:realized} -\small -\setlength{\tabcolsep}{3pt} -\renewcommand{\arraystretch}{1.12} -\begin{tabular}{@{}llrrrrr@{}} -\toprule\noalign{} -Model & Workload & \shortstack{Examples before/\\after filtering} & -\shortstack{Completed\\processes} & \shortstack{Prompt\\mean / max.} & -\shortstack{Response\\mean / max.} & Cap batches \\ -\midrule\noalign{} -Qwen2.5-3B & GSM8K & 32 / 32 & 12 & 108 / 154 & 298 / 830 & 0 / 12 \\ -Qwen2.5-3B & MATH & 64 / 64 & 12 & 104 / 437 & 392 / 512 & 12 / 12 \\ -Qwen2.5-3B & Standard code & 32 / 32 & 12 & 421 / 518 & 286 / 958 & 0 / 12 \\ -Qwen2.5-3B & Longer-prompt code & 32 / 32 & 12 & 720 / 772 & 352 / 1,024 & 7 / 12 \\ -Phi-4-mini & GSM8K & 32 / 32 & 12 & 78 / 121 & 226 / 2,048 & 2 / 12 \\ -Phi-4-mini & MATH & 64 / 64 & 12 & 76 / 401 & 367 / 512 & 12 / 12 \\ -Phi-4-mini & Standard code & 32 / 32 & 12 & 379 / 480 & 326 / 1,024 & 12 / 12 \\ -Phi-4-mini & Longer-prompt code & 32 / 32 & 12 & 664 / 746 & 412 / 1,024 & 12 / 12 \\ -Granite-3.3-2B & GSM8K & 128 / 128 & 15 & 146 / 198 & 285 / 2,048 & 5 / 15 \\ -Granite-3.3-2B & MATH & 256 / 252 & 15 & 132 / 460 & 376 / 512 & 15 / 15 \\ -Granite-3.3-2B & Standard code & 64 / 47 & 15 & 492 / 588 & 228 / 1,024 & 1 / 15 \\ -Granite-3.3-2B & Longer-prompt code & 64 / 63 & 15 & 807 / 902 & 296 / 1,024 & 1 / 15 \\ -\bottomrule -\end{tabular} -\end{table*} +The targeted batch-by-logging control adds 24 eligible outcomes from +26 attempts; the admission panel adds 48 from 48. Before the larger-model +panels, these studies therefore contain 660 eligible processes from +731 attempts. The larger-model panels have separate planned-slot and +eligibility counts in Section~\ref{larger-model-results}. Sixteen supporting +historical screens and abandoned-scope observations remain separate. +Supplementary Section S1 gives the exclusions and missing records. -Table~\ref{tab:realized} reports the realized work in completed boundary runs. Each executes one global training batch. The larger Granite preparation subsets do not increase that batch, but model-specific filtering can change which examples reach it. All 156 completed runs have batch-length summaries; the 60 failed runs do not. Missing summaries are not interpreted as zero-length work. Machine-readable summaries cover all original studies; the supplement presents the completed 100-step studies and the new control, including fixed-order data reuse. +Every completed boundary run executes one global batch and has a length +summary; the sixty failures do not. Model-specific filtering can change +which examples reach that batch. Supplementary Section S12 reports these +counts and realized lengths without treating missing summaries as zero +work. \subsection{Transfer between execution settings}\label{transfer-between-models} -Each model is evaluated on all four workloads at the six levels in Table~\ref{tab:levels}. Development runs checked that each case included both within-margin and unsuitable settings. Qwen and Phi use seeds 41--43 for the mathematics workloads and 51--53 for code, giving 144 trials. Granite adds 72 trials on the same workload and configuration grid, using seeds 91--93 throughout. - -The outcomes are repeatability of labels, the highest observed safe level, and directional transfer errors. Peak-memory differences match workload and configuration level, subtracting the models' mean peaks where all three processes complete for each model. Seeds are not matched between Granite and the other models. Because the model choices differ in architecture, parameter count, and tokenizer, these comparisons evaluate transfer between practical model choices rather than isolating any one of those properties. +Each model uses all four workloads and six levels +(Table~\ref{tab:levels}), with three seeds per configuration. Qwen and +Phi use seeds 41--43 for mathematics and 51--53 for code; Granite uses +91--93. Development runs checked that each case spanned within-margin +and unsuitable settings. Peak contrasts match workload and level and +require all three runs to complete for both models. Model choice also +changes architecture, parameter count, and tokenizer. The retrospective workload baseline uses one donor's six-level profile to predict the other workloads for that model; collecting it requires @@ -742,6 +641,106 @@ the task contract in Section~\ref{benchmark-tasks-and-evaluation}. The four-GPU study retains one generation worker per GPU without splitting an individual worker's model. Three models, two mathematics workloads, levels c2--c4, and seeds 111--113 give 54 processes. Global batch, rollout count, and the remaining settings are unchanged. We compare its labels in both directions with the earlier two-GPU results. Because GPU count was not randomized within a common experimental period, this is a transfer test rather than an isolated causal estimate of adding GPUs. +\subsection{Estimators beyond label copying}\label{estimation-methods} + +We also test whether model structure and configuration settings help +predict memory outcomes. This comparison uses a separate fitting protocol, +not the task-local donor restriction above. Its historical pool comprises +the 72 boundary configurations and eighteen four-GPU configurations. +Each configuration is one observation combining three eligible seeds. +In each of three retrospective tests, all thirty configurations from one +model family are withheld and the other sixty supply fitting evidence. +Thus, no seed, workload, or GPU-count variant of the test family enters +the fit. The public historical results were visible during method design; +this is held-out evaluation, not blind method development. + +The first baseline copies a donor label. Donors must match workload and +compound level; selection then prefers the same GPU count, the same +model family, and the nearest parameter count, in that order. A fixed +identifier resolves ties. This rule uses settings, not target outcomes. +The two fitted baselines instead use eight size-related features derived +from checkpoint headers (tensor shape and type metadata), model +architecture, and execution settings. +They describe actor shards, adapter optimizer state, generation reservation +and replicas, large model units, activation-related volumes, and +actor residency during generation. Supplementary Section S16 gives +their definitions. + +\emph{Component regression with a startup guard} fits the completed-run +peak as an intercept plus a nonnegative weighted sum of these features. +Only configurations with three completions supply numerical peak targets. +Failure peaks are neither set to capacity nor replaced by a survivor's +peak. Coefficients summarize this dataset; they are not separately +identified physical memory components. The method is an empirical +baseline, not a reproduction of DNNMem, LLMem, or ReaL. + +An additional one-sided check tests whether resident actor shards plus +the requested generation budget already exceed device capacity: +\begin{equation} + \frac{4(P+R)}{G}+uC>C, + \label{eq:startup-budget} +\end{equation} +where \(P\) and \(R\) are base-model and adapter parameter counts, +\(G\) is the number of actor GPUs, \(u\) is the reservation fraction, +and \(C\) is capacity in bytes. The factor four represents FP32 storage. +The check applies only without actor parameter offload. It counts +generation storage within its budget and adds no separate reference copy. +A positive check predicts memory failure; a negative check is not an +approval. Otherwise, the fitted peak determines the three-state prediction +using the capacity and operating-limit thresholds. The release preserves +both this final label and the label before the startup check. + +The second fitted baseline is a regularized multinomial logistic +classifier: it learns a linear score for each outcome class and selects +the largest. It uses all source labels, including failures. +Feature scaling is learned within each fitting set. Regularization, +class weighting, feature definitions, and decision rules are fixed +without a parameter search; classifier scores are not calibrated +failure probabilities. Supplementary Section S16 gives the complete +specification and acquisition counts. + +For prospective size transfer, we froze all three methods' predictions +for twelve Qwen2.5-7B-Instruct configurations before any target execution. +The fit uses all ninety historical configurations. The target contains +approximately 7.6 billion stored floating-point parameters and uses +GSM8K, MATH, and longer-prompt code at levels c2 and c3 on two or four +GPUs. This tests a larger model within the already known Qwen family +and hardware setting, not transfer to a new architecture or device type. +Earlier 7B exploration used other workflows and is excluded from fitting. + +The frozen plan assigns three seeds to each configuration, giving +36 invocations. Eighteen allocations each reserve four GPUs and +192 GiB of host memory, then run a two-GPU and a four-GPU condition +sequentially. Period order is randomized and +balanced; each two-of-four device subset occurs three times. Fresh +processes, physical-device checks, and all-device idle checks separate +the periods. Model, workload, global prompt batch, level, and seed +match within each pair. Generated responses may differ, so the contrast +concerns the total execution consequence of GPU count, not pure sharding +of identical tensors. + +Each invocation requests one training step with full logging, without +validation, checkpoint writing, or restoration. No retry or replacement +is permitted. A repeated label requires all three eligible seeds; +unresolved configurations remain in the inventory, with each method's +unscored approvals reported separately. Host-memory exhaustion does +not establish a GPU-memory failure. Task-provisioned GPU time and +the four-GPU reservation cost are also separated. Supplementary +Section S16 records the specification, execution-source changes, +missingness, and cost accounting. + +The initial four-GPU c2 runs were interrupted by host-memory exhaustion. +We therefore froze a separate c2 follow-up under an expanded host-resource +condition, with 384 GiB of host memory and the processor allocation fixed +within each pair. It covers all three original workloads with three new +seeds, giving eighteen invocations in nine matched allocations. +The original predictors, donor choices, and thresholds remain unchanged. +This follow-up was motivated by the initial observations; it is not an +independently selected new workload sample. Its labels and costs are +reported separately, without replacing or pooling the original slots. +The within-pair comparison concerns GPU count under the new host +condition, not an isolated effect of increasing host memory between panels. + \subsection{Budgeted admission with separate evaluation runs}\label{prospective-admission} To test the value of already available measurements, we specified a @@ -755,8 +754,7 @@ Table~\ref{tab:workloads}. The full training split is used after filtering, in fixed order. Validation and checkpoint writing occur only at step 5. Validation uses the first 32 eligible test examples for GSM8K and 64 for MATH. -This changes several aspects of the workload and schedule; it is not an -isolated test of longer responses or more training steps. +This is a compound workload and schedule change, not a single-factor test. Each of the twelve target configurations has four fresh processes. Seed 141 supplies the only queryable screen, with the complete @@ -764,14 +762,13 @@ five-step schedule. Seeds 142--144 independently form the three-repeat evaluation label. Screening measurements therefore do not contribute to the outcome against which their decisions are judged. The matched original boundary configuration supplies donor evidence. -Target settings, source records, seed roles, acquisition order, rules, +Target settings, source records, seed roles, screening order, rules, and budgets were frozen before execution. Physical collection order is randomized with a recorded seed. -The acquisition order is c3, c2, c4 in both Qwen cases and Phi/MATH; -Phi/GSM8K uses c4, c3, c2. These orders were randomly assigned before -execution, then held fixed for every rule. The comparison is therefore -of illustrative fixed rules, not optimized acquisition strategies. +Screening orders were randomly assigned before execution: c3, c2, c4 +in both Qwen cases and Phi/MATH, and c4, c3, c2 in Phi/GSM8K. +Every rule uses the same order within a case. We compare three simple rules. \emph{Source copy} initially approves the donor's within-margin configurations. A \emph{headroom guard} @@ -779,42 +776,33 @@ additionally requires the donor peak to leave another 2,048 MiB below the operating limit. This is an engineering choice made with the historical corpus visible, not a calibrated risk bound. \emph{Direct screening} starts without a decision on any target and uses no donor -measurements. All rules request screens in the same preassigned order. +measurements. A within-margin screen changes that candidate to approve; above-margin completion or memory failure changes it to reject; unresolved noncompletion causes abstention. Unqueried candidates retain the initial rule's decision. -Offline replay reveals only the requested screens, charging each -invocation before its outcome becomes available. Budgets are zero -through three target attempts per case. We report within-margin recall -jointly with memory-failure and margin-violation approvals, including -unresolved targets. A predefined setup repair, if needed, consumes -another attempt; known memory failures and unresolved runtime failures -do not permit retries. The frozen cap is one repair per slot and eight -overall. Missing evaluation outcomes retain bounds over admissible -labels instead of disappearing from the denominator. - -Point recall uses established within-margin evaluation targets. -An abstention recovers none of them; unresolved evaluation labels -instead enter the reported admissible-outcome bounds. A screen that -never started is not charged, causes abstention on that candidate, -and stops acquisition for that replay. The supplement gives the full -missing-outcome contract. - -Cost has three distinct meanings. In the existing-donor use case, target -attempts are incremental acquisition effort; the nine recorded donor -attempts per case remain visible as prior investment. Cold-start -accounting adds that investment for transfer rules. The separate -evaluation runs are research cost, not free policy observations. -Attempts are not equal amounts of computation: donors execute one step, -targets five. This panel does not test a direct-screening strategy that -spends a donor-sized budget on additional target repetitions, so it -cannot establish cold-start superiority over that broader strategy. +Offline replay reveals only requested screens and charges each invocation +before revealing its outcome, at budgets of zero to three attempts per +case. Unresolved targets remain explicit, and unstarted work is not +charged. Supplementary Section S9 specifies the frozen setup-repair +limits, missing-outcome bounds, abstention, and stopping rules. + +Incremental effort counts target attempts when donor evidence already +exists. Cold-start accounting adds the nine historical donor attempts +per case for transfer rules. The full screening/evaluation panel is +research cost, not freely available method evidence. Attempts differ +in work: donors execute one step and targets five. The comparison does +not test spending a donor-sized budget on repeated direct screening. \subsection{State-release and micro-batch experiments}\label{memory-mechanism-and-100-step-experiments} -The release experiment compares removing generation state with keeping it resident. Each pair runs both conditions sequentially on the same node and GPUs. Their order is counterbalanced: half run release first and half resident first. This is a \emph{crossover design}, in which each allocation receives both conditions. Before each period, the runner requires combined idle memory below 2,048 MiB across the two GPUs and records their baselines. Eligibility additionally requires records for both runs and identical GPU identities. This aggregate idle check differs from the later controls' per-GPU check. +To test whether retaining generation state limits completion, the +release experiment runs both conditions sequentially on the same +node and GPUs, with half of the pairs release-first and half resident-first. +Before each period, combined idle memory must be below 2,048 MiB across +the two GPUs. Eligible pairs require both records and identical GPU +identities. This aggregate check differs from the later per-GPU check. Release uses vLLM Sleep Mode level 1 to suspend generation workers after initialization or generation. Weights move to host memory and the KV @@ -828,125 +816,128 @@ Section~\ref{releasing-generation-state-changes-completion} gives all six \emph{base configurations}, the shared settings before release is varied. Each uses three seeds. -A separate \emph{factorial experiment} tests every combination of two factors: actor micro-batch size (8 or 16) and vLLM reservation (0.60 or 0.70). It covers three models, four workloads, and three common seeds. A four-condition block consists of the runs sharing a model, workload, and seed. The main effect of a factor averages its difference over the two values of the other factor; their interaction measures whether that difference depends on the other factor. - -These original blocks are analytical groupings, not same-allocation crossovers. Their 144 processes ran in separate scheduler allocations across twenty nodes during the same experimental session. Placement and execution order were not randomized within blocks. Environment records identify each allocation and source state, including modified working copies. Thus, fixed scientific settings do not by themselves eliminate placement, order, or source-history effects. The new control below repeats the batch contrast in immutable, same-GPU blocks. +A separate \emph{factorial experiment} combines actor micro-batches 8 and +16 with vLLM reservations 0.60 and 0.70 across three models, four +workloads, and three seeds. Each model--workload--seed block contains +four conditions. A factor contrast averages its difference over the +other factor; an interaction measures dependence of that difference +on the other factor. -Safety effects use all blocks. A pairwise peak contrast requires both runs to complete, while a peak main effect or interaction requires all four peaks. The separate stage-specific reanalysis uses the same blocks and is identified as post hoc. +These 144 processes ran in separate allocations across twenty nodes, +without randomized placement or order within blocks. They are analytical +groupings, not same-GPU comparisons. Label comparisons use all blocks; +pairwise peaks require both completions and factorial peak contrasts +require all four. The retrospective stage reanalysis uses the same +blocks; the control below repeats the batch contrast on the same GPUs. \subsection{Measurement controls}\label{follow-up-controls-and-transfer-tests} +These controls test whether sampling and allocator logging alter the +reported outcomes or stage contrasts. Sampling calibration compares 10 ms and 100 ms external intervals in -six matched configuration pairs, using one process at each interval and -seed 53 throughout. The cases use Qwen2.5-1.5B and Qwen2.5-3B on the -two mathematics workloads, with below-limit and near-boundary settings. -Each process executes one step without validation or checkpoint writing. -The supplementary settings table specifies the six configurations; -these pairs are not same-allocation crossovers. Separately, the 10 ms -traces are resampled at 20, 50, 100, and 200 ms, with starting offsets -in 10 ms increments. This checks loss of observed peaks in the recorded -traces, not the unobserved behavior between their original samples. - -The instrumentation control tests whether allocator logging changes the measurement. It covers the three models on GSM8K and MATH at each model's previous highest safe and next unsafe level, using seeds 111--113. Each of 36 allocations runs two fresh processes on the same GPUs. The \emph{external-only} condition retains NVML sampling and stage markers. The \emph{full} condition additionally synchronizes GPU work before each traced worker operation, resets allocator peak counters, and records allocator statistics when the operation returns. The wrapper adds no separate synchronization after that return. Waiting for outstanding GPU operations can affect timing and stage attribution. - -The two orders each occur in 18 allocations. Each period starts only after all GPUs return below 1,024 MiB of idle memory use. Completion and safety disagreements are primary outcomes; peak and elapsed-time differences use only pairs in which both conditions complete. All twelve base configurations are reported. This comparison measures the additional effect of allocator instrumentation, not the cost of all measurement activity. - -The new batch-by-logging control uses Phi/GSM8K, where the existing stage measurements vary most, and Qwen2.5-3B/longer-prompt code, which adds code coverage. Each of three seeds (121--123) receives four fresh processes on the same two GPUs: micro-batch 8 or 16, each with external-only or full logging. Reservation is fixed at 0.60 and other workload settings follow Table~\ref{tab:workloads}. The six blocks use fixed counterbalanced condition orders, idle-device checks, an immutable source checkout, and no validation or checkpoints. The actor-stage batch effect subtracts the batch-8 peak from the batch-16 peak; the logging interaction subtracts the external-only batch effect from the full-logging effect. These are the primary outcomes. A pre-execution 512 MiB reference flags potentially material differences, not a certified equivalence margin. Every block and any incomplete attempt are retained. +six matched configuration pairs, not same-allocation pairs. Separately, +the 10 ms traces are resampled at intervals up to 200 ms with different +starting offsets. Supplementary Section S8 gives the settings. + +The instrumentation control covers the three models on GSM8K and MATH +at each model's highest within-margin level and the next level, using +seeds 111--113. Each of 36 allocations runs two fresh processes on the +same GPUs. The \emph{external-only} condition retains NVML sampling and +stage markers. The \emph{full} condition additionally synchronizes GPU +work before each traced worker operation, resets allocator peak counters, +and records statistics when the operation returns, without another +synchronization. Waiting for outstanding work can change timing and +stage attribution. + +The two orders each occur in eighteen allocations. Each period requires +idle memory below 1,024 MiB on every GPU. Completion and margin-label +disagreements use all eligible pairs; peak and time contrasts require +both completions. This comparison concerns added allocator logging, +not the cost of all measurement activity. + +The new batch-by-logging control uses Phi/GSM8K, where the existing stage +measurements vary most, and Qwen2.5-3B/longer-prompt code, which adds code +coverage. Each of three seeds (121--123) receives four fresh processes on +the same two GPUs: micro-batch 8 or 16, each with external-only or full +logging. Reservation is fixed at 0.60 and other workload settings follow +Table~\ref{tab:workloads}. The six blocks use fixed counterbalanced +condition orders, idle-device checks, an immutable source checkout, +and no validation or checkpoints. + +The actor-stage batch effect subtracts the batch-8 peak from the batch-16 +peak; the logging interaction subtracts the external-only batch effect +from the full-logging effect. These are the primary outcomes. +A pre-execution 512 MiB reference flags potentially material differences, +not a certified equivalence margin. Every block and any incomplete +attempt are retained. \subsection{Testing one-step screens over longer runs}\label{testing-one-step-screens-over-longer-runs} -One-step runs permit broad configuration screening, but do not establish -behavior over a longer schedule. The original duration test uses two -cases and three regimes: below the operating limit, near it, and unsafe -in the available historical screen. Three seeds per regime give eighteen -forty-step runs. The sixteen supporting screens comprise three processes -for each regime except unsafe MATH, which has one; their logs verify -one-step execution despite an early trace format with a constant step -counter. - -These cases differ from the boundary workloads. Table~\ref{tab:duration-settings} -specifies their shape and event schedule. Both use the full training -split after prompt-length filtering, fixed data order, global batch -128, five responses per prompt, at most 256 active sequences, and -log-probability micro-batch 8. Reservations are 0.50 below the limit, -0.65 near it, and 0.80 for the unsafe regime. Actor micro-batches are -16 for Qwen2.5-3B and 4 for Qwen2.5-1.5B, except that the unsafe regime -uses 16 for both. - -\begin{table*}[!t] -\centering -\caption{Duration-test settings and scheduled operations. Prompt and response caps are tokens. Every listed final-step event is required for completion.}\label{tab:duration-settings} -\footnotesize -\setlength{\tabcolsep}{4pt} -\begin{tabular}{@{}p{0.24\linewidth}p{0.15\linewidth}p{0.20\linewidth}p{0.31\linewidth}@{}} -\toprule -Study and case & Prompt / response cap & Training steps & Validation and checkpoint schedule \\ -\midrule -Original Qwen2.5-3B / GSM8K & 256 / 1,024 & 40; 100 & 40-step: neither operation. 100-step: validation before training and every 20 steps; checkpoints at 80 and 100. \\ -Original Qwen2.5-1.5B / MATH & 1,792 / 1,024 & 40; 100 & Same schedule; validation subset at most 128 prompts. \\ -Paired Phi and Granite / GSM8K & 1,024 / 2,048 & 1; 100 & Validation at 0 and every 20 steps; checkpoints every 25; both at the final step, including step 1. \\ -Paired Phi and Granite / MATH & 512 / 512 & 1; 100 & Same schedule; validation subsets 32 prompts for GSM8K and 64 for MATH. \\ -\bottomrule -\end{tabular} -\end{table*} - -The outcomes are completion, agreement with the one-step safety label, and the largest later increase over the memory peak at step 1. Reward and optimization statistics are retained as checks that training executed. - -Actor parameter and optimizer offload are disabled, generation-state release is enabled, and the original forty-step runs have no validation or checkpoint writing. The historical label comparison uses the available source screens, including the unsafe MATH regime's single-screen exception; it is not a new three-seed screen for every regime. - -The original 100-step extension retains the two cases and the below-limit -and near-boundary regimes, using three new seeds (91--93). Its scheduled -operations appear in Table~\ref{tab:duration-settings}. Their peaks are -compared with the maximum over the remaining stages. - -The paired duration test covers Phi and Granite on GSM8K and MATH with +To test whether short-run labels hold over a longer schedule, the +paired duration test covers Phi and Granite on GSM8K and MATH with seeds 111--113. Each allocation runs a fresh one-step screen and a separate 100-step process. Actor micro-batch is 16; reservations 0.65 for Phi and 0.75 for Granite interpolate between their earlier tested -levels and were fixed before these outcomes. Both horizons use the full -training split after prompt-length filtering and the schedule in -Table~\ref{tab:duration-settings}. Runs start without resuming a checkpoint; -only the latest saved state is kept. Long runs remain in the comparison -even when their paired screen is unsafe. Completion checks verify every -requested step and scheduled validation and checkpoint event. The -primary comparison combines three repetitions; individual pairs are -reported separately. +levels and were fixed before these outcomes. Both horizons use the +filtered full training split in fixed order. Validation occurs before +training and every twenty steps, checkpoint writing every 25 steps, +and both at the final step, including step 1. Runs start without +checkpoint restoration; only the latest saved state is kept. Long runs +remain in the comparison even when their screen exceeds the margin. +Completion checks cover every requested update and scheduled operation. +The primary comparison combines three repetitions; individual pairs +are reported separately. + +Earlier tests cover Qwen2.5-3B/GSM8K and Qwen2.5-1.5B/MATH. Three +settings per case and three seeds give eighteen forty-step runs; the +below-limit and near-boundary settings also receive twelve 100-step +runs. Their sixteen historical screens include only one for the +unsuitable MATH setting, rather than three. The forty-step runs omit +validation and checkpoints, while the 100-step extension includes both. +Supplementary Section S14 gives these distinct workloads and schedules. + +Besides completion and label agreement, we report later growth relative +to each long run's own first-step peak and compare validation/checkpoint +peaks with the maximum over other stages. \subsection{Statistics}\label{statistics} -Repeated seeds vary generation and execution randomness, not the sampled -workload. We therefore keep related measurements together: processes -support attempt accounting, configurations support labels, and complete -cases support transfer comparisons. - -A \emph{cluster bootstrap} resamples groups rather than individual -processes to describe sensitivity to the measured case composition. -The groups are six base configurations for state release, four workloads -for model and factorial comparisons, and twelve base configurations for -the instrumentation control. The reported 95\% descriptive resampling -intervals use 20,000 resamples with recorded seeds. Only three base -configurations contribute completed pairs to the release peak contrast. - -These selected groups are neither numerous nor independent samples of a deployment population. We therefore emphasize per-case outcomes, effect ranges, and agreement in direction. A sensitivity analysis gives the two code subsets half weight each, so that the three source corpora contribute equally. The two-case original temporal study and the new two-case control are reported by case without population-risk bounds or bootstrap precision claims. - -The post hoc stage-specific intervals are descriptive and have no adjustment for testing multiple effects. They are not evidence from a prospectively registered hypothesis test. +The analysis is descriptive. We report configuration counts, per-case +results, and differences within matched allocations. Repeated seeds +test execution repeatability; they are not new workloads or independent +samples of a deployment population. Transfer comparisons keep complete +model--workload cases together, following the emphasis on implementation +and few-run uncertainty in reinforcement-learning evaluation +\cite{henderson2018matters,agarwal2021precipice}. + +The main paper emphasizes observed ranges and agreement in direction. +The supplement retains the earlier group-resampling summaries and a +sensitivity analysis that gives the two code subsets half weight each. +Those calculations describe dependence on the selected cases; they do +not provide population-risk bounds or compensate for the small number +of distinct workloads. \section{Results}\label{results} -We first evaluate configuration labels and their transfer, including -budgeted admission (RQ1). We then examine state release, micro-batch -effects, and their measurement controls (RQ2), before testing short -screens against longer execution (RQ3). +We first report configuration outcomes (RQ1), then examine state release +and stage measurements (RQ2), followed by short/long agreement (RQ3). +Transfer, budgeted configuration decisions, and fitted prediction +comparisons complete RQ1. -\subsection{Completion and memory safety}\label{completion-and-memory-safety} +\subsection{Completion and memory margin}\label{completion-and-memory-safety} -The three-model boundary experiment contains 216 eligible processes representing 72 configurations. Table~\ref{tab:boundary} distinguishes completion from safety: 156 processes complete, but only 132 leave the specified margin. The sixty failures occur during generation initialization (51), generation itself (six), or weight synchronization (three). +Of the 72 configurations in the three-model boundary experiment, 44 +complete within margin, eight complete above it, and twenty have memory +failures (Table~\ref{tab:boundary}). Their three repetitions give 216 +eligible processes. The sixty failing processes stop +during generation initialization (51), generation (six), or weight +synchronization (three). -All three repetitions agree within each configuration. There are 44 safe configurations and 28 unsafe configurations; the latter include both memory failures and runs that complete above the operating limit. +All three repetitions agree within each configuration. \begin{table*}[!t] \centering -\caption{Completion and safety in the original boundary experiment.}\label{tab:boundary} +\caption{Completion and margin compliance in the original boundary experiment.}\label{tab:boundary} \footnotesize \setlength{\tabcolsep}{3pt} \renewcommand{\arraystretch}{1.12} @@ -969,7 +960,7 @@ Within margin \end{minipage} & \begin{minipage}[b]{\linewidth}\raggedleft Memory failures \end{minipage} & \begin{minipage}[b]{\linewidth}\raggedleft -Safe configurations +Within-margin configurations \end{minipage} \\ \midrule\noalign{} Qwen2.5-3B & 72 & 48 & 48 & 24 & 16/24 \\ @@ -988,208 +979,6 @@ Figure~\ref{fig:boundary} shows where each outcome occurs within the tested conf \caption{Three-state feasibility profiles across the boundary grid. Each cell combines three fresh processes. Identical profiles within a model explain the success of the donor-workload lookup baseline.}\label{fig:boundary} \end{figure*} -\subsection{Simple lookup and cross-model transfer}\label{model-labels-are-not-interchangeable} - -Within each model, the full six-level, three-state profile is identical across the four workloads. A donor-workload lookup therefore predicts every remaining workload label correctly. Across twelve donor choices, this gives 216 correct comparisons involving 72 distinct target configurations, not 216 independent observations. This grid does not demonstrate a need for a learned workload-specific predictor. - -The model-specific profiles differ. Qwen and Granite remain within margin through c3, whereas Phi does so through c2. Table~\ref{tab:model-transfer} shows the consequences of copying these profiles between models. - -\begin{table*}[!t] -\centering -\caption{Source-model lookup at the primary margin. Each direction compares 24 target configurations. Approval errors distinguish actual memory failure from completion above margin; rejected usable settings are reported separately.}\label{tab:model-transfer} -\small -\setlength{\tabcolsep}{3pt} -\renewcommand{\arraystretch}{1.12} -\begin{tabular}{@{}lrrrrr@{}} -\toprule\noalign{} -Source $\rightarrow$ target & \shortstack{Correct\\three-state label} & -Approved & \shortstack{Approved:\\memory failure} & -\shortstack{Approved:\\above margin} & \shortstack{Rejected:\\within margin} \\ -\midrule\noalign{} -Qwen $\rightarrow$ Phi & 20 & 16 & 0 & 4 & 0 \\ -Granite $\rightarrow$ Phi & 16 & 16 & 0 & 4 & 0 \\ -Phi $\rightarrow$ Qwen & 20 & 12 & 0 & 0 & 4 \\ -Phi $\rightarrow$ Granite & 16 & 12 & 0 & 0 & 4 \\ -Qwen $\rightarrow$ Granite & 20 & 16 & 0 & 0 & 0 \\ -Granite $\rightarrow$ Qwen & 20 & 16 & 0 & 0 & 0 \\ -\bottomrule -\end{tabular} -\end{table*} - -Transferring Qwen's labels to Phi accepts four of Phi's twelve unsafe configurations. All four complete above the operating limit: these errors concern the reserved memory margin, not missed OOMs. Qwen and Granite agree on binary safety at all 24 settings, although their three-state profiles differ because Granite completes above margin at c4, where Qwen fails. - -The four Qwen-to-Phi errors are also four of the 16 configurations approved by Qwen. Thus, the false-safe fraction among unsafe targets and the error fraction among approved settings answer different questions. Neither denominator represents a sampled deployment workload. - -At matched settings where both models complete, Phi's mean peak exceeds -Qwen's by 3,517 MiB and Granite's by 4,598 MiB. Granite's mean is -1,081 MiB below Qwen's. The Qwen/Phi boundary difference has the same -direction in every workload. Supplementary descriptive intervals summarize -variation across these selected workloads, not uncertainty about unseen -deployment workloads. - -The distinction depends on the reserved margin. Table~\ref{tab:margin} recalculates the labels under five operating thresholds; memory failures remain unsafe throughout. This sensitivity analysis was performed after the primary experiment. - -\begin{table*}[!t] -\centering -\caption{Post hoc sensitivity to the reserved memory margin.}\label{tab:margin} -\footnotesize -\setlength{\tabcolsep}{3pt} -\renewcommand{\arraystretch}{1.12} -\begin{tabular}{@{} - >{\raggedright\arraybackslash}p{(\linewidth - 8\tabcolsep) * \real{0.1765}} - >{\raggedleft\arraybackslash}p{(\linewidth - 8\tabcolsep) * \real{0.2353}} - >{\raggedright\arraybackslash}p{(\linewidth - 8\tabcolsep) * \real{0.1765}} - >{\raggedright\arraybackslash}p{(\linewidth - 8\tabcolsep) * \real{0.1765}} - >{\raggedleft\arraybackslash}p{(\linewidth - 8\tabcolsep) * \real{0.2353}}@{}} -\toprule\noalign{} -\begin{minipage}[b]{\linewidth}\raggedright -Capacity allowed -\end{minipage} & \begin{minipage}[b]{\linewidth}\raggedleft -Limit (MiB) -\end{minipage} & \begin{minipage}[b]{\linewidth}\raggedright -Qwen highest safe level -\end{minipage} & \begin{minipage}[b]{\linewidth}\raggedright -Phi highest safe level -\end{minipage} & \begin{minipage}[b]{\linewidth}\raggedleft -Qwen→Phi false-safe / unsafe target configurations -\end{minipage} \\ -\midrule\noalign{} -90\% & 36,864 & c2, all workloads & c2, all workloads & 0/12 \\ -92.5\% & 37,888 & c3, all workloads & c2, all workloads & 4/12 \\ -95\% (primary) & 38,912 & c3, all workloads & c2, all workloads & 4/12 \\ -97.5\% & 39,936 & c3, all workloads & c3 GSM8K; c2 otherwise & 3/11 \\ -100\% & 40,960 & c3, all workloads & c3, all workloads & 0/8 \\ -\bottomrule -\end{tabular} -\end{table*} - -Granite reaches c3 at the primary limit and c4 when no margin is reserved. At 90\% and 92.5\%, it reaches c3 except on MATH, where it reaches c2. Qwen and Phi have the same completion boundary at the full-capacity limit. The reproduction package includes all 60 model--workload--threshold combinations and 30 directional transfer summaries. - -\subsection{Transfer between GPU counts}\label{transfer-between-gpu-counts} - -On four GPUs, all 54 processes complete and 48 leave the specified margin. The repeated labels therefore identify sixteen of eighteen configurations as safe. - -\begin{table}[!t] -\centering -\caption{Outcomes at the same configuration levels on two and four GPUs.}\label{tab:gpu-transfer} -\footnotesize -\setlength{\tabcolsep}{3pt} -\renewcommand{\arraystretch}{1.12} -\begin{tabular}{@{}lll@{}} -\toprule\noalign{} -Model / workload & Two-GPU c2 / c3 / c4 & Four-GPU c2 / c3 / c4 \\ -\midrule\noalign{} -Granite-3.3-2B / GSM8K & safe / safe / above & safe / safe / safe \\ -Granite-3.3-2B / MATH & safe / safe / above & safe / safe / safe \\ -Phi-4-mini / GSM8K & safe / above / failure & safe / safe / above \\ -Phi-4-mini / MATH & safe / above / failure & safe / safe / above \\ -Qwen2.5-3B / GSM8K & safe / safe / failure & safe / safe / safe \\ -Qwen2.5-3B / MATH & safe / safe / failure & safe / safe / safe \\ -\bottomrule -\end{tabular} -\end{table} - -In Table~\ref{tab:gpu-transfer}, ``above'' denotes completion above the operating limit, and ``failure'' denotes at least one memory failure. Across the eighteen matched configurations, transferring two-GPU labels to four GPUs wrongly rejects six within-margin configurations and makes no approval error. Reverse transfer approves two configurations that fail and four that complete above margin. These six errors are six of eight unsafe two-GPU targets, or six of sixteen configurations approved on four GPUs. The tested c2--c4 range does not establish the four-GPU boundary beyond those levels. - -These results also show why exact label accuracy is not a sufficient -admission score. Across both GPU-transfer directions, source copying -gets 20 of 36 labels right. Always predicting within-margin completion -gets 26 right, but approves four memory-failing targets rather than two -(Table~\ref{tab:accuracy-diagnostic}). Always rejecting would avoid -approval errors while finding no usable configuration. A useful -comparison must therefore keep coverage and harmful approvals together. - -\begin{table}[!t] -\centering -\caption{Diagnostic rules on the 36 GPU-transfer queries. ``Always reject'' predicts memory failure for every target. Counts are decisions, not deployment-risk estimates.}\label{tab:accuracy-diagnostic} -\footnotesize -\setlength{\tabcolsep}{3pt} -\begin{tabular}{@{}p{0.28\linewidth}rrrr@{}} -\toprule -Rule & \shortstack{Correct\\labels} & \shortstack{Usable\\approved} & -\shortstack{Failure\\approved} & \shortstack{Above\\approved} \\ -\midrule -Source copy & 20 & 20 & 2 & 4 \\ -Always approve & 26 & 26 & 4 & 6 \\ -Always reject & 4 & 0 & 0 & 0 \\ -\bottomrule -\end{tabular} -\end{table} - -\subsection{Existing evidence and the cost of new tests}\label{admission-results} - -All 48 planned invocations supply validated outcomes: 32 complete and -16 fail for memory. There are no retries or unresolved slots. The -twelve screens and 36 separate evaluation runs agree at every candidate. -The repeated evaluation labels identify six within-margin candidates, -two above-margin candidates, and four with memory failures -(Table~\ref{tab:admission}). Every completed process performs all five -updates and the final validation and checkpoint operations. - -\begin{table*}[!t] -\centering -\caption{Admission results for all four prospective cases. Entries under -the rules count usable candidates approved / usable candidates available. -Budget is requested target attempts per case. ``Safe,'' ``above,'' and -``failure'' denote the three repeated memory labels. Every approval by -these rules is within margin at every tested budget.}\label{tab:admission} -\small -\setlength{\tabcolsep}{5pt} -\renewcommand{\arraystretch}{1.15} -\begin{tabular}{@{}llrrrrr@{}} -\toprule -& & \multicolumn{2}{c}{No target tests} & -\multicolumn{3}{c}{Direct-screening budget} \\ -\cmidrule(lr){3-4}\cmidrule(l){5-7} -Model / workload & Evaluation c2 / c3 / c4 & -Source copy & Guard & 1 & 2 & 3 \\ -\midrule -Qwen2.5-3B / GSM8K & safe / safe / failure & 2/2 & 1/2 & 1/2 & 2/2 & 2/2 \\ -Qwen2.5-3B / MATH & safe / safe / failure & 2/2 & 1/2 & 1/2 & 2/2 & 2/2 \\ -Phi-4-mini / GSM8K & safe / above / failure & 1/1 & 1/1 & 0/1 & 0/1 & 1/1 \\ -Phi-4-mini / MATH & safe / above / failure & 1/1 & 1/1 & 0/1 & 1/1 & 1/1 \\ -\bottomrule -\end{tabular} -\end{table*} - -With donor evidence already available, source copy recovers all six -usable candidates without a target test. The headroom guard rejects -Qwen c3 in both workloads, reducing case-averaged recall---the equally -weighted mean of the four case recalls---to 75\%. -It avoids no additional approval error on this panel. After one -target attempt per case, it recovers those candidates. However, -source copy and the guard then make identical decisions by construction: -they differ initially only at the two Qwen c3 candidates, which are -queried first. Their positive-budget agreement is not an independent -validation of two distinct strategies. - -Direct screening recovers no candidate before testing. Its -case-averaged recall rises to 25\%, 75\%, and 100\% at budgets of one, -two, and three attempts per case, respectively. The fixed order -explains why Phi/GSM8K finds its only usable candidate last. -All nonempty approval sets have 100\% observed precision; no rule -approves a memory failure or an above-margin target. Approving -everything would also give full recall, but would admit all four -memory-failing and both above-margin candidates. Thus, recall alone -would conceal six harmful approvals. - -At the first tested common per-case budget attaining full recall, the -incremental target-attempt totals are zero for source copy, four for -the guard, and twelve for direct screening. The transfer rules additionally -use 36 historical donor attempts, giving cold-start attributed totals of -36, 40, and 12 at those points. Attempts contain different amounts of -work, so these counts do not establish GPU-time savings or optimal -acquisition. Collecting the full panel of twelve screening and -36 evaluation invocations consumed 9.5 allocated-GPU-hours of runner -time, excluding queue and other allocation overhead. - -For RQ1, existing measurements were sufficient in this changed-workload -panel, and the additional guard was unnecessarily restrictive. The -model and GPU-transfer results establish the complementary limit: -that success cannot justify copying labels across arbitrary settings. -The benchmark's value is to measure this distinction, including when -a simple rule already solves the selected cases. - \subsection{Releasing generation state changes completion}\label{releasing-generation-state-changes-completion} In the eighteen same-GPU pairs, release succeeds and resident fails in @@ -1202,7 +991,7 @@ peaks, even though they contain the main completion result. \begin{table*}[!t] \centering \caption{All state-release base configurations. Prompt/response limits are tokens; their sum is the model-length limit. The selected training subset equals the global batch. Actor and log-probability micro-batches are per GPU; the latter applies to both policy and reference scoring. Completed counts are release / resident, each out of three.}\label{tab:release} -\footnotesize +\small \setlength{\tabcolsep}{3pt} \renewcommand{\arraystretch}{1.12} \begin{tabular}{@{}llrrrrrlr@{}} @@ -1230,30 +1019,36 @@ resident-first pairs. The supplementary plot and intervals retain the order and grouping information. Throughput and latency costs of release were not the objective of this experiment. -\subsection{The overall peak hides the micro-batch effect}\label{the-overall-peak-hides-the-training-batch-effect} +\subsection{Stage-level and overall memory contrasts}\label{the-overall-peak-hides-the-training-batch-effect} -All 144 processes in the batch/reservation experiment complete. Of these, 120 leave the specified margin and 24 cross the operating threshold. The three-repetition rule yields forty safe and eight unsafe configurations; none of the unsafe labels in this experiment is caused by a crash. +All 144 batch/reservation processes complete: 120 within margin and +24 above it, giving forty within-margin and eight above-margin +configuration labels. None fails for memory. -Increasing vLLM reservation from 0.60 to 0.70 raises mean peak memory by +The higher vLLM reservation, 0.70 rather than 0.60, is associated with a +mean peak increase of 3,998 MiB, close to the additional tenth of device capacity reserved. -All Phi settings cross the operating limit, while Qwen and Granite +All Phi configurations at reservation 0.70 exceed the operating limit; +their matched 0.60 configurations remain within it. Qwen and Granite retain their labels. The peak increase occurs across all three models. -Doubling actor micro-batch from 8 to 16 leaves the labels unchanged and -changes the mean overall peak by only \(-31\) MiB. This small difference +The batch-16 and batch-8 configurations have identical labels, and their +mean overall peaks differ by only \(-31\) MiB. This small difference does not mean that processing more sequences has little memory cost. The stage-specific reanalysis of the 36 four-condition blocks finds -much larger increases during actor update: 3,980 MiB in externally +much larger batch contrasts during actor update: 3,980 MiB in externally sampled device memory and 3,848 MiB in allocated tensor memory (Table~\ref{tab:factorial}). The latter takes the maximum across training -workers. These are observations under full instrumentation; the adjacent -control results test their sensitivity to logging. +workers. These runs used separate allocations without randomized placement or order, +so the contrasts alone do not isolate causal effects. The matched +controls below test whether the actor-stage batch contrast persists +under tighter placement control and without allocator logging. \begin{table*}[!t] \centering -\caption{Overall and actor-update effects under full instrumentation. Brackets give 95\% bootstrap resampling intervals over the four selected workloads; the new controls assess magnitude sensitivity.}\label{tab:factorial} -\footnotesize +\caption{Mean descriptive contrasts in the original batch/reservation grid under full instrumentation. Placement and order were not randomized. Per-workload values and earlier resampling summaries are retained in the supplement.}\label{tab:factorial} +\small \setlength{\tabcolsep}{3pt} \renewcommand{\arraystretch}{1.12} \begin{tabular}{@{} @@ -1263,7 +1058,7 @@ control results test their sensitivity to logging. >{\raggedleft\arraybackslash}p{(\linewidth - 6\tabcolsep) * \real{0.2667}}@{}} \toprule\noalign{} \begin{minipage}[b]{\linewidth}\raggedright -Factorial main effect +Configuration contrast \end{minipage} & \begin{minipage}[b]{\linewidth}\raggedleft Whole-run NVML peak (MiB) \end{minipage} & \begin{minipage}[b]{\linewidth}\raggedleft @@ -1272,13 +1067,16 @@ Actor-update NVML peak (MiB) Actor-update allocated peak (MiB) \end{minipage} \\ \midrule\noalign{} -Actor micro-batch 8→16 & −31 {[}−75, −3{]} & 3,980 {[}2,588, 5,563{]} & 3,848 {[}2,502, 5,383{]} \\ -Reservation 0.60→0.70 & 3,998 {[}3,918, 4,052{]} & 9 {[}−216, 170{]} & 1 {[}−200, 132{]} \\ +Actor micro-batch 8→16 & −31 & 3,980 & 3,848 \\ +Reservation 0.60→0.70 & 3,998 & 9 & 1 \\ \bottomrule \end{tabular} \end{table*} -Actor-update NVML effects differ across workloads: 2,073 MiB for GSM8K, 3,448 MiB for MATH, 4,132 MiB for standard code, and 6,268 MiB for longer-prompt code. Giving the two code subsets half weight each reduces the mean from 3,980 to 3,574 MiB; all workload effects remain positive. The reservation effect is essentially unchanged at 4,006 MiB under this weighting. +Actor-stage batch contrasts are positive in every workload, ranging +from 2,073 to 6,268 MiB in NVML memory. Giving the two related code +subsets half weight each retains this direction +(Supplementary Section S3). Actor update determines the overall peak in none of the 144 traces. Generation or weight synchronization dominates most runs; the remaining @@ -1293,106 +1091,77 @@ persists without allocator instrumentation. \caption{Targeted batch-by-logging control. Each connected pair compares the actor-stage batch-16 minus batch-8 effect under external-only and full instrumentation within one four-condition allocation. Asterisks identify additional allocations after initial noncompletion. These are observed block effects, not confidence intervals. Section~\ref{effect-of-allocator-instrumentation} reports the attempt history.}\label{fig:logging} \end{figure*} -\subsection{Effect of allocator instrumentation}\label{effect-of-allocator-instrumentation} - -The instrumentation comparison contains 72 processes in 36 eligible pairs. Completion differs in 0/36 pairs, and safety differs in 0/36 pairs. Both conditions complete in 30/36 pairs. Table~\ref{tab:logging} reports each base configuration. Completed and safe counts are given for external-only and full instrumentation, respectively; each condition has three repetitions. Peak and time differences are full minus external-only and use only pairs where both complete. - -\begin{table*}[!t] -\centering -\caption{Allocator-instrumentation control. Paired counts are external-only; full.}\label{tab:logging} -\footnotesize -\setlength{\tabcolsep}{3pt} -\renewcommand{\arraystretch}{1.12} -\begin{tabular}{@{} - >{\raggedright\arraybackslash}p{(\linewidth - 12\tabcolsep) * \real{0.2667}} - >{\raggedright\arraybackslash}p{(\linewidth - 12\tabcolsep) * \real{0.0667}} - >{\raggedleft\arraybackslash}p{(\linewidth - 12\tabcolsep) * \real{0.1333}} - >{\raggedleft\arraybackslash}p{(\linewidth - 12\tabcolsep) * \real{0.1333}} - >{\raggedleft\arraybackslash}p{(\linewidth - 12\tabcolsep) * \real{0.1333}} - >{\raggedleft\arraybackslash}p{(\linewidth - 12\tabcolsep) * \real{0.1333}} - >{\raggedleft\arraybackslash}p{(\linewidth - 12\tabcolsep) * \real{0.1333}}@{}} -\toprule\noalign{} -\begin{minipage}[b]{\linewidth}\raggedright -Model / workload -\end{minipage} & \begin{minipage}[b]{\linewidth}\raggedright -Level -\end{minipage} & \begin{minipage}[b]{\linewidth}\raggedleft -Completed -\end{minipage} & \begin{minipage}[b]{\linewidth}\raggedleft -Safe -\end{minipage} & \begin{minipage}[b]{\linewidth}\raggedleft -Label disagreements -\end{minipage} & \begin{minipage}[b]{\linewidth}\raggedleft -Peak difference (MiB) -\end{minipage} & \begin{minipage}[b]{\linewidth}\raggedleft -Time difference (s) -\end{minipage} \\ -\midrule\noalign{} -Granite-3.3-2B / GSM8K & c3 & 3; 3 & 3; 3 & 0/3 & 0.0 & -15.3 \\ -Granite-3.3-2B / GSM8K & c4 & 3; 3 & 0; 0 & 0/3 & 0.0 & -6.3 \\ -Granite-3.3-2B / MATH & c3 & 3; 3 & 3; 3 & 0/3 & 0.0 & 4.7 \\ -Granite-3.3-2B / MATH & c4 & 3; 3 & 0; 0 & 0/3 & 0.0 & -8.0 \\ -Phi-4-mini / GSM8K & c2 & 3; 3 & 3; 3 & 0/3 & -6.7 & 11.0 \\ -Phi-4-mini / GSM8K & c3 & 3; 3 & 0; 0 & 0/3 & -13.3 & -4.3 \\ -Phi-4-mini / MATH & c2 & 3; 3 & 3; 3 & 0/3 & -149.3 & 5.0 \\ -Phi-4-mini / MATH & c3 & 3; 3 & 0; 0 & 0/3 & -44.7 & 5.0 \\ -Qwen2.5-3B / GSM8K & c3 & 3; 3 & 3; 3 & 0/3 & 0.0 & -19.3 \\ -Qwen2.5-3B / GSM8K & c4 & 0; 0 & 0; 0 & 0/3 & --- & --- \\ -Qwen2.5-3B / MATH & c3 & 3; 3 & 3; 3 & 0/3 & 0.0 & 13.7 \\ -Qwen2.5-3B / MATH & c4 & 0; 0 & 0; 0 & 0/3 & --- & --- \\ -\bottomrule -\end{tabular} -\end{table*} - -There were forty external-only attempts and thirty-six full-instrumentation attempts. Two external attempts had diagnosed setup problems, and two reached the startup limit without a trial record. The latter remain unresolved; their full-instrumentation counterparts never ran. Excluding the two diagnosed setup attempts leaves 38 allocation attempts with between zero and two possible label disagreements, or 0--5.3\%, depending on the unknown outcomes. This is a missing-outcome sensitivity range, not an observed paired result or confidence interval. The zero disagreement count among eligible pairs does not establish equivalence. - -Among the thirty completed pairs, the mean whole-run difference is -\mbox{\(-21.4\) MiB} and the mean elapsed-time difference is -\mbox{\(-1.4\) seconds}. Supplementary intervals resample the twelve -base configurations in the design; ten contribute completed pairs. -A dash in Table~\ref{tab:logging} means neither condition completed, -not that its missing peak was estimated. - -Whole-run agreement is not sufficient to validate individual stages. Actor-update peaks are observed in both arms of 33 pairs, including three pairs that later fail. Median full-minus-external differences for the eleven available base configurations range from $-176$ to $400$ MiB. Individual differences range from $-1,800$ to $1,456$ MiB. The Qwen/MATH c4 pairs fail before actor update, so that stage has no comparison. The supplement gives every stage contrast. This variation motivated the direct batch-by-logging control rather than an assumption that whole-run agreement extends to actor-stage effects. - -The targeted control initially completed four of its six allocations. Sixteen processes finished within margin; two full-logging processes, each first in its assigned order, did not finish within the twenty-minute limit. One logged an accelerator argument error during weight-transfer buffer reconstruction, while the other's cause remains unresolved. Neither is established as a memory-capacity failure. Their remaining six conditions never started. A dated amendment, recorded before any additional outcomes, allowed one further unchanged allocation for each incomplete block without a diagnosed memory failure. These additional attempts do not replace the initial noncompletion or add new seeds. - -Both additional allocations completed, yielding eight further processes and six usable four-condition blocks overall. Thus, 24 of 26 attempted processes completed within margin. The actor-stage batch effect is positive under external-only measurement in every usable block: its case means are 1,966 MiB for Phi/GSM8K and 6,430 MiB for Qwen/longer-prompt code (Figure~\ref{fig:logging}). The corresponding full-logging means are 2,287 and 7,059 MiB. The full-minus-external batch contrasts range from $-220$ to 1,054 MiB across blocks; three of six exceed the 512 MiB diagnostic in absolute value. Whole-run batch effects remain between $-160$ and 132 MiB. - -These results support the direction of the actor-stage increase without allocator logging, but not an instrumentation-independent effect size. Prompt batches match within each case, whereas generated responses are not identical across conditions despite matched seeds. The contrasts therefore include execution and generation variability, not only tracing overhead on fixed tensors. Nor do the two initial full-first failures establish a logging effect: condition, order, seed, and allocation remain possible explanations. The supplement reports all block effects, realized response lengths, and initial and additional attempts separately. - -\subsection{Measurement calibration}\label{measurement-calibration} - -All twelve runs in the six sampling pairs complete. Resampling the 10 ms traces at coarser intervals and different starting offsets gives 222 derived checks up to 200 ms, not additional independent runs. None changes the observed peak or its assigned stage. Runs sampled at 10 ms are 1.8 seconds shorter on average. These results support 100 ms sampling for the sustained peaks observed here; they cannot rule out missed shorter transients elsewhere. +\subsection{Sensitivity to measurement}\label{effect-of-allocator-instrumentation} + +Completion and margin labels agree between logging conditions in the +36 eligible pairs, although stage contrasts can change. +Both conditions complete in thirty. +Their mean full-minus-external differences are \(-21.4\) MiB for the +whole-run peak and \(-1.4\) seconds for elapsed time. Supplementary +Section S15 reports each base configuration. + +There are forty external-only and 36 full-logging attempts. Two external +attempts have setup errors; two time out at startup with unresolved +cause and no full-logging counterpart. Excluding diagnosed setup errors +leaves 38 allocation attempts and zero to two possible disagreements +(0--5.3\%). This is a missing-outcome range, not an observed disagreement +rate or evidence of equivalence. + +Actor-stage peaks are available in 33 pairs, including three that fail +later. Individual full-minus-external differences range from \(-1,800\) +to 1,456 MiB; base-configuration medians range from \(-176\) to 400 MiB. +Qwen/MATH c4 fails before actor update. These stage differences motivated +the direct batch-by-logging control (Supplementary Section S2). + +That control initially completes four of six allocations: sixteen +processes finish within margin, and two full-logging processes, both +first in order, reach the twenty-minute limit. One reports an accelerator +argument error during weight transfer; the other remains unexplained. +Neither is diagnosed as OOM, and the six remaining conditions never +start. A bounded amendment permits one unchanged additional allocation +per incomplete block, retaining the initial outcomes +(Supplementary Section S6). + +Both additional allocations complete, giving six complete four-condition +blocks and 24 within-margin completions among 26 attempts. Actor-stage +batch contrasts are positive under external-only measurement in every +complete block: case means are 1,966 MiB for Phi/GSM8K and 6,430 MiB for +Qwen/longer-prompt code (Figure~\ref{fig:logging}). Full logging gives +2,287 and 7,059 MiB, respectively. The logging interaction ranges from +\(-220\) to 1,054 MiB; three of six blocks exceed the 512 MiB diagnostic +in absolute value. Whole-run batch contrasts remain between \(-160\) +and 132 MiB. + +The actor-stage increase persists without allocator logging, but its +magnitude depends on measurement and generated work: matched prompts +and seeds do not produce identical responses. The two initial failures +also cannot separate logging from order, seed, or allocation. +Supplementary Section S6 retains every block and response-length summary. + +\label{measurement-calibration} +All twelve sampling-calibration runs complete. Coarser resampling of the +10 ms traces changes neither their observed peaks nor assigned stages. +This supports 100 ms sampling for the sustained peaks in these traces +(Supplementary Section S8). \FloatBarrier \subsection{Short tests and longer execution}\label{longer-runs-retain-labels-despite-memory-growth} -In the original forty-step study, all twelve runs from initially safe configurations complete and remain within margin. Of the six runs from unsafe configurations, three complete above the operating threshold and three fail during generation initialization. All eighteen labels agree with the available historical screens, including the single-screen exception described in the protocol. Among the fifteen completed runs, the maximum later increase over step 1 averages 1,717 MiB and reaches 2,404 MiB. - -The original 100-step extension has twelve completed attempts out of fourteen. Two Qwen--GSM8K attempts stop after 63 and 77 completed steps: the first reports a CUDA invalid-argument error during weight transfer, and the second stops making progress during weight synchronization with unresolved cause. Neither establishes an OOM. Their replacements complete, giving twelve eligible outcomes that all remain within margin. Thus, label agreement among eligible outcomes does not imply uninterrupted completion of every attempt. - -Among the twelve completed runs, the maximum later increase averages -1,925 MiB. Validation and checkpoint writing remain below the other -stages' maximum in these cases. The supplement gives case-level effects, -an endpoint comparison of first-step and maximum peaks, and the complete -attempt accounting. - -These are two selected model--workload cases. Their agreement does not estimate the risk of later failure in unseen settings. - -\paragraph*{Paired tests on Phi and Granite.}\label{longer-run-validation-on-phi-and-granite} +\label{longer-run-validation-on-phi-and-granite} The paired study compares twelve fresh one-step screens with separate 100-step runs. All long runs complete. Two individually within-margin screens are paired with above-margin long runs, both for Granite on -MATH. Crucially, another short repetition already exceeds the limit, -so the three-repeat rule rejects that case before the long runs. +MATH. Another short repetition already exceeds the limit, +so the three-repeat short-run label already classifies that case +as above-margin. -Table~\ref{tab:horizon} separates individual completion and safety counts from the label obtained by requiring all three repetitions to be safe. +Table~\ref{tab:horizon} distinguishes individual outcomes from the label +requiring all three repetitions to complete within margin. \begin{table*}[!t] \centering -\caption{One-step and 100-step outcomes, with memory measurements from completed long runs.}\label{tab:horizon} +\caption{One-step and 100-step outcomes, with memory measurements from completed long runs. Later growth is relative to each long run's own first-step peak, not the separate screen.}\label{tab:horizon} \footnotesize \setlength{\tabcolsep}{3pt} \renewcommand{\arraystretch}{1.12} @@ -1407,9 +1176,9 @@ Table~\ref{tab:horizon} separates individual completion and safety counts from t \begin{minipage}[b]{\linewidth}\raggedright Model / workload \end{minipage} & \begin{minipage}[b]{\linewidth}\raggedleft -Screen completed; safe +Screen completed; within margin \end{minipage} & \begin{minipage}[b]{\linewidth}\raggedleft -Long completed; safe +Long completed; within margin \end{minipage} & \begin{minipage}[b]{\linewidth}\raggedright Repeated labels \end{minipage} & \begin{minipage}[b]{\linewidth}\raggedleft @@ -1418,10 +1187,10 @@ Mean later growth (MiB) Mean validation / checkpoint peak (MiB) \end{minipage} \\ \midrule\noalign{} -Granite-3.3-2B / GSM8K & 3/3; 3/3 & 3/3; 3/3 & safe → safe & 159 & 38,443 / 13,498 \\ +Granite-3.3-2B / GSM8K & 3/3; 3/3 & 3/3; 3/3 & within → within & 159 & 38,443 / 13,498 \\ Granite-3.3-2B / MATH & 3/3; 2/3 & 3/3; 0/3 & above → above & 657 & 38,875 / 15,039 \\ -Phi-4-mini / GSM8K & 3/3; 3/3 & 3/3; 3/3 & safe → safe & 115 & 37,581 / 19,581 \\ -Phi-4-mini / MATH & 3/3; 3/3 & 3/3; 3/3 & safe → safe & 590 & 37,883 / 24,620 \\ +Phi-4-mini / GSM8K & 3/3; 3/3 & 3/3; 3/3 & within → within & 115 & 37,581 / 19,581 \\ +Phi-4-mini / MATH & 3/3; 3/3 & 3/3; 3/3 & within → within & 590 & 37,883 / 24,620 \\ \bottomrule \end{tabular} \end{table*} @@ -1431,155 +1200,386 @@ All long runs pass the required training, validation, and checkpoint checks. Later growth compares each long run with its own first step, not with the separate short-run peak. -\section{Discussion}\label{discussion} +\paragraph*{Earlier duration tests.} +In the forty-step study, all twelve runs from initially within-margin +settings complete within margin. Of six from unsuitable settings, +three complete above margin and three fail during generation +initialization. All eighteen outcomes agree with the historical screens, +including the single-screen exception. Among fifteen completions, +later growth over step 1 averages 1,717 MiB and reaches 2,404 MiB. + +The original 100-step extension completes twelve of fourteen attempts. +Two Qwen/GSM8K attempts stop after 63 and 77 steps: one reports a +CUDA invalid-argument error during weight transfer; the other stops +progressing during synchronization with unresolved cause. Neither +establishes OOM. Their replacements complete, leaving twelve eligible +within-margin outcomes. Among these completions, later growth averages +1,925 MiB, and validation/checkpoint peaks remain below other stages. +Supplementary Sections S1 and S5 preserve the attempt history and +case-level contrasts. -The practical choice is not just which configuration to run, but which -additional observation would justify that choice. The results distinguish -the roles of source evidence, measurements of individual stages, and -tests of the intended execution schedule. - -\subsection{What can be transferred between configurations}\label{what-can-be-transferred-between-configurations} - -An approval is specific to an execution setting. The transfer results -therefore support using donor measurements as evidence to be tested, -not as a certificate attached to a micro-batch or reservation value. -The benchmark makes this distinction operational by evaluating a -method's approvals against the target's own repeated runs. - -The admission result is useful specifically when donor measurements -already exist: it quantifies additional testing that did not improve -the observed set of usable configurations, and the usable settings lost -to the extra allowance. It does not establish a better cold-start -search method. Both this small panel and the saturated static workload -task leave little room to distinguish more capable predictors. Finer -settings near the boundary, additional systems, and a wider range of -workloads would be needed for that purpose. - -\subsection{Why stage-level measurements matter}\label{why-stage-level-measurements-matter} - -Stage measurements explain why a configuration changes even when its -overall label does not. The actor-stage growth matters if later changes -reduce the generation peak or increase update demand: a previously -nonlimiting stage can become the constraint. The measured contrast is -therefore useful diagnostic evidence, not a universal amount to add -to every configuration's memory estimate. Generated responses and -measurement conditions also affect its magnitude. - -\subsection{How to use a short-run screen}\label{how-to-use-a-short-run-screen} - -A short test is a filter for a specified schedule, not evidence about -every later operation. The admission panel compares separate executions -of the same five-step schedule; the duration experiments test a changed -horizon. Repeating the test caught the observed -Granite--MATH margin problem, but does not prove that three repetitions -will catch all future failures. Likewise, the measured growth across -different schedules cannot be converted into a universal allowance -to add to a one-step peak. Validation should cover the intended -duration and operations, with all-cause noncompletion recorded -separately from the conditional memory label. +\subsection{Simple lookup and cross-model transfer}\label{model-labels-are-not-interchangeable} -\section{Threats to validity}\label{threats-to-validity} +Within each model, the full six-level, three-state profile is identical +across the four workloads. Donor-workload lookup therefore predicts +every remaining workload label correctly: 216 comparisons across twelve +donor choices involve 72 distinct targets, not 216 independent +observations. This grid does not demonstrate a need for a learned +workload-specific predictor. + +The profiles differ between models. Qwen and Granite remain within +margin through c3, whereas Phi does so through c2. +Table~\ref{tab:model-transfer} shows the consequences of copying them. + +\begin{table*}[!t] +\centering +\caption{Source-model lookup at the primary margin. Each direction compares +24 target configurations. Approval errors distinguish memory failure +from completion above margin.}\label{tab:model-transfer} +\small +\setlength{\tabcolsep}{3pt} +\renewcommand{\arraystretch}{1.12} +\begin{tabular}{@{}lrrrrr@{}} +\toprule\noalign{} +Source $\rightarrow$ target & \shortstack{Correct\\three-state label} & +Approved & \shortstack{Approved:\\memory failure} & +\shortstack{Approved:\\above margin} & \shortstack{Rejected:\\within margin} \\ +\midrule\noalign{} +Qwen $\rightarrow$ Phi & 20 & 16 & 0 & 4 & 0 \\ +Granite $\rightarrow$ Phi & 16 & 16 & 0 & 4 & 0 \\ +Phi $\rightarrow$ Qwen & 20 & 12 & 0 & 0 & 4 \\ +Phi $\rightarrow$ Granite & 16 & 12 & 0 & 0 & 4 \\ +Qwen $\rightarrow$ Granite & 20 & 16 & 0 & 0 & 0 \\ +Granite $\rightarrow$ Qwen & 20 & 16 & 0 & 0 & 0 \\ +\bottomrule +\end{tabular} +\end{table*} + +Copying Qwen's labels to Phi approves four above-margin settings and no +memory failures. Those four are four of twelve unsuitable targets, or +four of sixteen approvals: the denominators answer different questions. +Qwen and Granite approve the same settings, but their three-state +profiles differ at c4, where Granite completes above margin and Qwen +fails. + +The selected margin matters. Qwen and Phi have the same completion +boundary when the full device capacity is allowed. Supplementary +Section S15 retains the five-threshold sensitivity analysis; Section S7 +gives completed-run peak contrasts between models. + +\subsection{Transfer between GPU counts}\label{transfer-between-gpu-counts} + +On four GPUs, all 54 processes complete and 48 leave the specified margin, +giving sixteen within-margin configurations out of eighteen. + +\begin{table*}[!t] +\centering +\caption{Outcomes at the same levels on two and four GPUs. ``Within'' +and ``above'' distinguish completed outcomes relative to the margin; +``failure'' denotes memory failure.}\label{tab:gpu-transfer} +\small +\setlength{\tabcolsep}{3pt} +\renewcommand{\arraystretch}{1.12} +\begin{tabular}{@{}lll@{}} +\toprule\noalign{} +Model / workload & Two-GPU c2 / c3 / c4 & Four-GPU c2 / c3 / c4 \\ +\midrule\noalign{} +Granite-3.3-2B / GSM8K & within / within / above & within / within / within \\ +Granite-3.3-2B / MATH & within / within / above & within / within / within \\ +Phi-4-mini / GSM8K & within / above / failure & within / within / above \\ +Phi-4-mini / MATH & within / above / failure & within / within / above \\ +Qwen2.5-3B / GSM8K & within / within / failure & within / within / within \\ +Qwen2.5-3B / MATH & within / within / failure & within / within / within \\ +\bottomrule +\end{tabular} +\end{table*} + +Across the eighteen matched configurations, two-to-four-GPU transfer +wrongly rejects six within-margin settings and makes no approval error. +Reverse transfer approves two memory-failing and four above-margin +settings: six of eight unsuitable targets, or six of sixteen approvals. +The c2--c4 range does not establish the four-GPU boundary beyond these +levels. + +Exact label accuracy alone can favor a worse approval rule: always +predicting within-margin completion gets more labels right than source +copying but approves more memory failures +(Table~\ref{tab:accuracy-diagnostic}). Always rejecting avoids approval +errors while recovering no within-margin setting. + +\begin{table}[!t] +\centering +\caption{Diagnostic rules on the 36 GPU-transfer queries. ``Always reject'' +predicts memory failure for every target. Counts are decisions, not +deployment-risk estimates.}\label{tab:accuracy-diagnostic} +\footnotesize +\setlength{\tabcolsep}{3pt} +\begin{tabular}{@{}p{0.28\linewidth}rrrr@{}} +\toprule +Rule & \shortstack{Correct\\labels} & \shortstack{Within-margin\\approved} & +\shortstack{Failure\\approved} & \shortstack{Above\\approved} \\ +\midrule +Source copy & 20 & 20 & 2 & 4 \\ +Always approve & 26 & 26 & 4 & 6 \\ +Always reject & 4 & 0 & 0 & 0 \\ +\bottomrule +\end{tabular} +\end{table} -The main limitations concern what the labels measure, which comparisons -isolate a change, and how far the measured cases can be generalized. +\subsection{Existing evidence and the cost of new tests}\label{admission-results} -\subsection{Construct validity}\label{construct-validity} +All 48 planned invocations supply validated outcomes: 32 complete and +16 fail for memory. There are no retries or unresolved slots. The +twelve screens and 36 separate evaluation runs agree at every candidate. +The repeated evaluation labels identify six within-margin candidates, +two above-margin candidates, and four with memory failures +(Table~\ref{tab:admission}). Every completed process performs all five +updates and the final validation and checkpoint operations. -The within-margin label depends on the chosen operating convention; -the sensitivity analysis reports that dependence. External sampling can -miss a transient shorter than its interval, despite agreement for the -sustained peaks in the calibration. +\begin{table*}[!t] +\centering +\caption{Admission results for four prospective cases. Entries under +the rules count within-margin candidates approved / available. +Budget is requested target attempts per case. Every approval by +these rules is within margin at every tested budget.}\label{tab:admission} +\small +\setlength{\tabcolsep}{5pt} +\renewcommand{\arraystretch}{1.15} +\begin{tabular}{@{}llrrrrr@{}} +\toprule +& & \multicolumn{2}{c}{No target tests} & +\multicolumn{3}{c}{Direct-screening budget} \\ +\cmidrule(lr){3-4}\cmidrule(l){5-7} +Model / workload & Evaluation c2 / c3 / c4 & +Source copy & Headroom guard & 1 & 2 & 3 \\ +\midrule +Qwen2.5-3B / GSM8K & within / within / failure & 2/2 & 1/2 & 1/2 & 2/2 & 2/2 \\ +Qwen2.5-3B / MATH & within / within / failure & 2/2 & 1/2 & 1/2 & 2/2 & 2/2 \\ +Phi-4-mini / GSM8K & within / above / failure & 1/1 & 1/1 & 0/1 & 0/1 & 1/1 \\ +Phi-4-mini / MATH & within / above / failure & 1/1 & 1/1 & 0/1 & 1/1 & 1/1 \\ +\bottomrule +\end{tabular} +\end{table*} -Stage attribution locates an observed failure; it does not identify the origin of every allocation that contributed to it. For example, memory retained from generation can contribute to a failure reported during weight synchronization. Similarly, agreement in whole-run outcomes under the instrumentation control does not establish that individual operations or stage timings are unaffected. +With donors already available, source copy recovers all six within-margin +candidates without target tests. The guard rejects Qwen c3 in both +workloads, reducing the equally weighted mean of the four case recalls +to 75\%, without avoiding an approval error. After one query per case, +copy and guard decisions agree by construction: their only initial +differences are the two Qwen c3 candidates, both queried first. -\subsection{Internal and statistical validity}\label{internal-and-statistical-validity} +Direct screening recovers no candidate before testing. Its +case-averaged recall rises to 25\%, 75\%, and 100\% at budgets of one, +two, and three attempts per case. The fixed order places Phi/GSM8K's +only within-margin candidate last. All nonempty approval sets have +100\% observed precision. Approving everything would also give full +recall, but would admit all four memory-failing and both above-margin +candidates. -The configuration levels change several controls together. Model comparisons also differ in architecture, parameter count, tokenizer, and sometimes the examples remaining after filtering. The batch/reservation experiment varies its two factors systematically but its original blocks are not controlled for allocation or execution order. The new same-GPU control addresses that concern for two selected cases, not the full factorial grid. Four-GPU measurements use a later experimental period rather than randomized GPU-count assignment. +At the first tested common per-case budget attaining full recall, the +incremental target-attempt totals are zero for source copy, four for +the guard, and twelve for direct screening. The transfer rules additionally +use 36 historical donor attempts, giving cold-start attributed totals of +36, 40, and 12 at those points. Attempts contain different amounts of +work, so these counts do not establish GPU-time savings or optimal +acquisition. The full panel's collection cost is reported in +Supplementary Section S10. -Using the same GPUs and counterbalancing execution order reduces placement and order effects, but cannot eliminate changes in host caches, generated responses, or other conditions between runs. The original external-only startup timeouts and the new full-first noncompletion occur under different designs. Neither should be erased by successful later attempts or attributed to instrumentation without further evidence. +\subsection{Held-out-family estimation}\label{estimation-results} -The workload set is small and selected rather than randomly sampled. Both code subsets come from the same corpus, and instrumentation base configurations share model--workload settings. Per-case reporting and corpus-weighting sensitivity make this dependence visible; repeated seeds and resampling cannot turn these cases into a representative population sample. +Table~\ref{tab:estimators} shows that the fitted baselines do not +uniformly improve on donor copying. Component regression predicts every +Qwen label correctly and reduces Phi's margin-violating approvals, +but it approves all four memory-failing Granite configurations. +The classifier avoids those Granite failures but rejects six +within-margin Qwen configurations. No startup check changes a label +in these retrospective tests: these differences arise from the fitted +models and donor choices, not that override. -The admission comparison uses new target runs and separate evaluation -seeds, but familiar models and workloads. Its guard was chosen with -the historical measurements visible. A fixed query order also limits -conclusions about other acquisition strategies. After publication, -the target labels are inspectable: they remain excluded from permitted -inputs, but cannot serve as a permanently hidden test set. +\begin{table*}[!t] +\centering +\caption{Held-out-family predictions. Each row evaluates thirty +configurations, with the other sixty used for fitting. ``Useful'' +counts within-margin approvals. Within-margin, above-margin, and failure +counts are 22/0/8 for Qwen, 16/6/8 for Phi, and 22/4/4 for Granite. +Seeds are not additional test examples.} +\label{tab:estimators} +\small +\setlength{\tabcolsep}{5pt} +\begin{tabular}{@{}llrrrrrr@{}} +\toprule +Held-out model & Method & Correct labels & Approved & Useful & +\shortstack{Failure\\approved} & \shortstack{Above\\approved} & +\shortstack{Within\\rejected} \\ +\midrule +Qwen & Donor copy & 26 & 22 & 22 & 0 & 0 & 0 \\ +& Regression + startup guard & 30 & 22 & 22 & 0 & 0 & 0 \\ +& Logistic classifier & 21 & 16 & 16 & 0 & 0 & 6 \\ +Phi & Donor copy & 24 & 22 & 16 & 0 & 6 & 0 \\ +& Regression + startup guard & 27 & 19 & 16 & 0 & 3 & 0 \\ +& Logistic classifier & 24 & 18 & 16 & 0 & 2 & 0 \\ +Granite & Donor copy & 26 & 22 & 22 & 0 & 0 & 0 \\ +& Regression + startup guard & 22 & 30 & 22 & 4 & 4 & 0 \\ +& Logistic classifier & 26 & 25 & 22 & 0 & 3 & 0 \\ +\bottomrule +\end{tabular} +\end{table*} -\subsection{External validity}\label{external-validity} +The component model's mean absolute peak error ranges from 1.35 to +2.57 GiB across the three test families, using only configurations +with three completions. These errors do not include failure configurations +and therefore cannot replace the categorical evaluation. +The held-out-family comparison distinguishes the baselines, whereas +workload copying is already perfect on the original boundary grid. +Three familiar families are insufficient to select a generally +reliable predictor. + +\subsection{Larger-model transfer and incomplete execution}\label{larger-model-results} + +Only six of the twelve configurations in the original 7B panel received +complete repeated labels, all for startup memory-admission failure. +No invocation supplied a validated completion. The remaining configurations +were unresolved because of setup failures, host-memory exhaustion, or +unstarted slots; Supplementary Section S16 retains the complete accounting. + +Donor copying approved all six resolved configurations incorrectly. +Both fitted methods predicted their failure labels correctly, but an +always-failure rule would do the same on this failure-only subset. +The regression's three within-margin predictions concern precisely the +four-GPU c2 configurations censored by host-memory exhaustion. +They are unscored approvals, not demonstrated successes. +The original panel thus identifies a failure of label copying under this +size change, but cannot establish useful-configuration discovery or a +completed-run memory benefit from additional GPUs. + +The expanded-host follow-up resolves the two-GPU c2 configurations +to the same startup-failure outcome. Its four-GPU invocations exit +successfully and record update and weight-synchronization calls, but +lack the required final-step record. They therefore remain unresolved; +the recorded work does not justify treating their partial peaks as +completed-run measurements. This extension supplies no validated +completion and does not resolve the regression's approvals. +Supplementary Section S16 gives the separate counts and diagnostic evidence. -The tested models contain approximately 1.5--3.8 billion parameters. Results from this compact LoRA setup need not extend to larger models, full-parameter training, other accelerators, or different execution frameworks. The workloads also omit multimodal tasks, extended agent interactions, and very long contexts. +\section{Discussion}\label{discussion} -The longest runs cover 100 steps. They test repeated validation and checkpoint writing, but not restoration from checkpoints or memory behavior over a full training schedule. Those operations and longer durations need their own evaluation. +\paragraph*{Inspect stages before interpreting a configuration control.} +\label{why-stage-level-measurements-matter} +Whole-run peaks answer a capacity question, while stage peaks reveal +where demand changes. Generation or synchronization can mask actor-stage +growth. A small overall batch-size contrast therefore does not establish +that larger training batches have little memory cost. Paired release +outcomes also show why comparisons restricted to two successful runs +can omit the important completion change. + +\paragraph*{Match the setting before reusing a profile.} +\label{what-can-be-transferred-between-configurations} +A four-GPU Qwen c4 record does not justify approving c4 on two GPUs: +the corresponding two-GPU configurations fail. A researcher can use the +archive to shortlist settings measured under closer execution conditions, +then test a changed workload or schedule. The relevant comparison is +not label accuracy alone, but how many within-margin candidates are +recovered and how many failing or above-margin candidates are approved. +The fitted baselines also require this check; their errors differ by +model. The study does not rank configurations by throughput, training +quality, or economic value. + +\paragraph*{Test the intended schedule.} +\label{how-to-use-a-short-run-screen} +Validation and checkpoint writing belong in a screen when they belong +in the intended execution. Repetition identified the measured +Granite--MATH margin crossing that individual short tests missed, but +does not establish a universal repetition count. Retain all-cause +noncompletion separately from the conditional memory label. -\subsection{Artifact validity}\label{artifact-validity} +\section{Threats to validity}\label{threats-to-validity} -The evidence package retains excluded attempts and earlier analyses, with records identifying the material used in this paper. File checksums can detect changes in those records, while reconstruction checks the analysis. Neither establishes that the original instrumentation captured every relevant event; that limitation remains separate from reproducibility. +\paragraph*{Measurement.}\label{construct-validity} +Margin compliance depends on the chosen operating limit. Sampling can +miss shorter transients, and stage markers identify when a peak or failure +is observed, not the origin of every contributing allocation. Agreement +in whole-run outcomes under logging controls therefore does not establish +unchanged stage timings or exact operation-level memory demand. + +\paragraph*{Comparisons and selected cases.}\label{internal-and-statistical-validity} +Compound levels change several controls; model choice also changes +architecture, size, tokenizer, and sometimes filtered examples. +Historical factorial blocks and GPU-count comparisons do not control +allocation and experimental period. The targeted batch-by-logging control +addresses placement in two selected cases, not the whole grid, and still allows generated-work +and host-state variation. The workloads are selected and partly related: +both code subsets come from one corpus. Repeated seeds test repeatability, +not population representativeness. The admission guard was chosen with +historical data visible, uses a fixed query order, and is evaluated on +familiar models and workloads. Public target labels are not a permanently +hidden evaluation set. + +\paragraph*{Applicability.}\label{external-validity} +The evidence concerns one synchronous LoRA-GRPO stack on A100 GPUs. +The main comparisons use models of approximately 1.5--3.8 billion +parameters; the larger-model panels test a 7.6-billion-parameter +model from the already represented Qwen family. Incomplete executions +limit the conclusions from that extension. The longest runs +cover 100 steps with validation and checkpoint writing, but do not test +checkpoint restoration or a full training schedule. Broader execution +regimes require their own evidence. The code reward measures syntax and +response shape, not success on executable tests; these runs therefore +do not establish feasibility or learning quality for test-based code +reinforcement learning. \section{Reproducibility}\label{reproducibility} -The benchmark evidence is publicly available on Hugging Face \cite{kobza2026benchmark}, with a separate repository for reproduction software \cite{kobza2026reproduction}. Both can be accessed without authentication or approval. The software records the exact evidence version and checks its files before analysis. - -The dataset opens with configuration settings and repeated outcomes, -with separate views of processes, attempts, and execution stages. -A data dictionary defines the fields, units, and missing values. -The lightweight interface exports permitted transfer inputs and scores -user-supplied predictions. The admission replay additionally records -requested screens, revealed outcomes, and decisions at each budget. -Both run without a training allocation. - -The package contains the raw records described in the measurement protocol, experiment matrices, exclusion decisions, software environments, and scripts for the tables and figures. Reconstruction requires only a central processing unit (CPU), not a GPU training allocation. It starts from the raw evidence rather than using the published result tables as inputs, and checks run counts, seeds, stage records, and completed steps before comparing the regenerated results with the reference outputs. - -The original evidence archive was downloaded and reconstructed by the -authors in a separate environment where the source project and -precomputed tables were unavailable to the analysis. All forty original -result files matched after normalizing only file-location prefixes. -This is an isolated reconstruction check, not an independent external -replication. The current release also includes the instrumentation -controls and the separately evaluated admission panel, with its frozen -protocol, raw attempts, and acquisition transcripts. Setup commands -and version-specific integrity reports accompany the reproduction code. - -Rerunning GPU training is a separate task for users able to adapt -cluster-specific launchers. It requires the recorded execution environment, -upstream checkpoints, prepared data, and an appropriate GPU allocation; -the CPU reconstruction is not a one-command replacement for that setup. -Environment records supply launch settings and source identifiers, -subject to the historical working-copy limitation in -Section~\ref{experimental-protocol}. The -release excludes model weights and optimizer checkpoint payloads. -Users obtain upstream models under their applicable terms; version -and license records accompany the evidence. - -The authors' original benchmark measurements and reproduction software are released under the MIT license. Third-party datasets, software, and template assets retain their respective terms. +RLVRAMBench releases versioned evidence on Hugging Face +\cite{kobza2026benchmark} and separate reproduction software +\cite{kobza2026reproduction}. The package includes raw attempts, +configurations, environment records, exclusion decisions, and a data +dictionary. Reconstruction starts from raw records, checks counts, +seeds, stages, and completed steps, and regenerates tables without a +GPU allocation. Transfer scoring and admission replay likewise run on +central processing units (CPUs); replay records requested observations +and decisions at each budget. + +The authors reconstructed the original archive in an isolated environment +without the source project or precomputed input tables. All forty result +files matched after normalizing file-location prefixes. This is an +authors' reconstruction check, not independent external replication. +Version-specific checks and setup instructions accompany the release. + +\label{artifact-validity} +Exact GPU rerunning additionally requires the recorded environment, +upstream models, prepared data, and cluster-adapted launchers. Some +historical working-copy changes were not captured as per-run patches, +so their commit IDs do not fully identify the executing source. +Later controls use immutable source copies. This source-history limit +is distinct from reconstructing results from saved measurements. +Model weights and optimizer checkpoint payloads are excluded. +The authors' measurements and software use the MIT license; third-party +materials retain their own terms. \section{Conclusion}\label{conclusion} -RLVRAMBench turns memory measurements into a testable configuration -decision: how many usable settings are recovered, which unsuitable -settings are approved, and what additional testing is required? -In the prospective panel, existing labels recovered every usable -candidate without new target tests, whereas an extra allowance -discarded two of them. Copying labels between GPU counts approved -configurations that failed, so that favorable result is specific to the tested change -and available donor evidence. - -The supporting controls show why such decisions need more than a -successful-run peak. Generation can conceal changes in training memory, -and an individual short run can miss a later margin crossing. -The public protocol, repeated outcomes, and complete attempt records -allow these observations to be reconstructed and alternative decisions -to be scored. They provide a bounded reference for colocated LoRA -GRPO, not a guarantee for unmeasured hardware, workloads, or training -durations. - -\section*{Acknowledgment}\label{acknowledgment} +This study measures memory feasibility in a colocated LoRA-GRPO +implementation on A100 GPUs. Releasing generation state enables +completion in some paired cases, and stage measurements reveal training-memory growth hidden +by the overall peak. Individual short tests miss some later margin +crossings even where repeated labels agree. Earlier labels transfer +successfully to the five-step workload panel but admit failures when +copied between GPU counts or to the larger model. The larger-model panels +resolve only failing configurations, leaving recovery of usable settings +unestablished. Fitted memory regression and classification +do not uniformly improve on label copying across held-out families. +Together, these findings support using prior measurements to shortlist +candidates and using stage-complete target tests to check changed +execution conditions. The released evidence makes both useful approvals +and decision errors inspectable. + +\section*{Acknowledgment}\phantomsection\label{acknowledgment} \addcontentsline{toc}{section}{Acknowledgment} OpenAI Codex \cite{openai2026codex} was used to assist with coding and to refine and polish the manuscript text. The authors are responsible for the code, analyses, and content. \bibliographystyle{IEEEtran} \bibliography{references,ieee_access_additions} +\par\noindent\begin{minipage}{\columnwidth} \phantomsection \begin{IEEEbiographynophoto}{Ond\v{r}ej Kobza} is affiliated with the Czech Institute of Informatics, Robotics and Cybernetics, Czech Technical University in Prague, Czech Republic. His work reported in this article concerns GPU-memory feasibility and reproducible benchmarking of large-language-model reinforcement learning. @@ -1590,4 +1590,5 @@ is affiliated with the Czech Institute of Informatics, Robotics and Cybernetics, is affiliated with the Czech Institute of Informatics, Robotics and Cybernetics, Czech Technical University in Prague, Czech Republic. He supervises the research reported in this article. \end{IEEEbiographynophoto} \EOD +\end{minipage} \end{document} diff --git a/paper/RLVRAMBench_supplement.pdf b/paper/RLVRAMBench_supplement.pdf index d1153189f6043a7fcca8b99971b51622f8af48dd..6db5973c0acdbd708458e04f6ec9ca9696fc584e 100644 --- a/paper/RLVRAMBench_supplement.pdf +++ b/paper/RLVRAMBench_supplement.pdf @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:3aa5a1852a615541249dbeecfc72fdf59c2787a2b132902dcd476bf9338191df -size 266688 +oid sha256:493614368da0b65b2bc639fb5722d65cb9f39cdaae3d900b17cd7e063915a1dd +size 421805 diff --git a/paper/RLVRAMBench_supplement.tex b/paper/RLVRAMBench_supplement.tex index 0d1107c8e925cfc9311b95c97963d8f93e4907eb..d757ce9e70fc20c348be2d677532fdc3fbaa2119 100644 --- a/paper/RLVRAMBench_supplement.tex +++ b/paper/RLVRAMBench_supplement.tex @@ -10,7 +10,8 @@ \renewcommand{\thesection}{S\arabic{section}} \setlength{\parindent}{0pt} \setlength{\parskip}{6pt} -\title{RLVRAMBench\\Supplementary Methods and Evidence} +\title{Memory Feasibility in Colocated Language-Model Reinforcement Learning\\ +Supplementary Methods and Evidence} \author{Ond\v{r}ej Kobza \and Jan \v{S}ediv\'{y}} \date{} \begin{document} @@ -97,6 +98,16 @@ The original protocol and amendment are both retained. Its five-step process timeout is thirty minutes, with forty-minute allocations and at most four allocations running concurrently. +Historical repairs repeated a slot without changing its scientific +settings; documented exclusions precede selection of exactly one eligible +outcome per slot. Setup errors include missing dependencies, incorrect +GPU visibility, and broken instrumentation. A replacement supplies a +conditional memory outcome without establishing the cause of an earlier +failure. Allocator recording starts with the instrumented worker, so +earlier engine failures are assessed from execution logs, external +measurements, stage traces, and error stacks. A missing trial record +alone establishes neither completion nor memory failure. + \section{Stage-specific effects of allocator instrumentation} Both instrumentation conditions retain external device-memory sampling and @@ -212,8 +223,8 @@ in the article and Table S2.}\label{sfig:mechanisms} \section{Realized work in longer runs} -The article reports realized lengths for every model--workload group in the -boundary experiment. Table S3 extends this audit to completed 100-step runs. +Section S12 reports realized lengths for every model--workload group in the +boundary experiment. Table S3 covers completed 100-step runs. The original two cases have two regimes with three seeds each; the follow-up has one regime and three seeds per case. Means weight completed processes equally. Every completed process has a length summary and a positive @@ -411,6 +422,14 @@ base configurations, although six enter its completion comparison. Ten of the instrumentation design's twelve base configurations contribute completed pairs. +These earlier descriptive calculations use 20,000 group resamples with +recorded seeds. Model and factorial comparisons resample the four +workloads; state release resamples six base configurations; the +instrumentation comparison resamples twelve. The stage-specific +calculations are retrospective and have no multiple-comparison +adjustment. Their intervals are not inferential precision claims for +unmeasured systems. + \begin{table}[htbp] \centering \caption{Descriptive aggregate effects and 95\% resampling intervals. @@ -426,6 +445,12 @@ Granite minus Qwen completed-run peak & $-1,081$ & [$-1,144$, $-976$] & Workload Release minus resident completion (percentage points) & 50.0 & [16.7, 83.3] & Base configuration \\ Resident minus release completed-run peak & 7,061 & [5,249, 9,391] & Base configuration \\ Whole-run batch--reservation interaction & 21 & [$-57$, 135] & Workload \\ +Whole-run batch-16 minus batch-8 peak & $-31$ & [$-75$, $-3$] & Workload \\ +Actor-stage batch-16 minus batch-8 device peak & 3,980 & [2,588, 5,563] & Workload \\ +Actor-stage batch-16 minus batch-8 allocated peak & 3,848 & [2,502, 5,383] & Workload \\ +Whole-run reservation 0.70 minus 0.60 peak & 3,998 & [3,918, 4,052] & Workload \\ +Actor-stage reservation contrast, device peak & 9 & [$-216$, 170] & Workload \\ +Actor-stage reservation contrast, allocated peak & 1 & [$-200$, 132] & Workload \\ Full minus external-only whole-run peak & $-21.4$ & [$-53.7$, $-1.3$] & Base configuration \\ Full minus external-only elapsed time (s) & $-1.4$ & [$-8.0$, 5.1] & Base configuration \\ Full minus external-only elapsed time (\%) & $-0.5$ & [$-3.4$, 2.6] & Base configuration \\ @@ -448,6 +473,11 @@ checkpoint writing are disabled. Prompt and response caps, rather than realized lengths, are shown; the global batch counts prompts. The micro-batch pair gives actor-update and log-probability sizes, with the latter used by both actor and reference computations. +The 10 ms traces are also resampled at 20, 50, 100, and 200 ms with +starting offsets in 10 ms increments. This tests loss of recorded peaks, +not behavior between the original samples. All twelve runs complete; +none of the 222 derived checks changes the observed peak or assigned +stage. Runs sampled at 10 ms are 1.8 seconds shorter on average. \begin{table}[htbp] \centering @@ -484,6 +514,12 @@ The public protocol and prelaunch amendment retain the exact rules; implementation corrections made during collection do not change those settings, orders, or allowed information. +Each invocation is charged before its outcome is revealed. A predefined +setup repair consumes another attempt, with at most one repair per slot +and eight overall. Diagnosed memory failures and unresolved runtime +failures do not permit retries. Only eligible setup failures may be +repaired under the recorded rules. + \begin{table}[htbp] \centering \caption{How missing admission evidence is treated. A recorded invocation @@ -575,4 +611,564 @@ They also separate twelve screening invocations from 36 evaluation invocations. The 36 historical donor attempts are prior evidence, not additional runs in this panel. +Collecting the twelve screening and 36 evaluation invocations consumed +9.5 allocated-GPU-hours of runner time, excluding queue and other +allocation overhead. This panel cost is separate from method acquisition +cost. + +\FloatBarrier +\section{Fixed training controls} + +Table~\ref{stab:training} records the controls held fixed in the reported +LoRA GRPO studies. Rank sets the adapter dimension and alpha scales its +update. Kullback--Leibler (KL) divergence penalizes departure from the +reference policy through the actor loss. Training samples tokens; +greedy validation selects the most probable next token. + +\begin{table}[htbp] +\centering +\caption{Fixed training controls. BF16 is 16-bit bfloat16; FP32 is +32-bit floating point. Fully Sharded Data Parallel (FSDP) partitions +training state across devices.} +\label{stab:training} +\small +\begin{tabular}{@{}p{0.36\linewidth}p{0.59\linewidth}@{}} +\toprule +Control & Setting \\ +\midrule +LoRA & Rank 64; alpha 32; selected linear layers, excluding embeddings and the output head \\ +Gradient checkpointing / padding removal & Enabled / enabled \\ +Actor FSDP precision & BF16 computation; FP32 stored parameter shards and gradient reduction \\ +vLLM inference precision & BF16 \\ +Optimizer & PyTorch AdamW; learning rate $3\times10^{-6}$ \\ +Optimizer controls & Betas (0.9, 0.999); weight decay 0.01; gradient clip 1.0 \\ +Update schedule & Constant learning rate; no initial ramp; one pass over the update batch \\ +Policy clipping / micro-batch sizing & Clip ratio 0.2 / fixed sizes \\ +KL regularization & Actor loss enabled; coefficient 0.001; low-variance estimator \\ +KL in reward / entropy bonus & Disabled / no entropy bonus \\ +Data order & Fixed order; overlong prompts filtered \\ +Reference computation & Frozen actor base with adapters disabled; no separate model \\ +Training generation & Sampling; temperature 1; top-p 1; top-k disabled \\ +Validation generation & Greedy decoding; one response per prompt \\ +\bottomrule +\end{tabular} +\end{table} + +\FloatBarrier +\section{Workload preparation and boundary-run work} + +Code preparation pools the original CodeContests validation and test +records, excludes problems requiring file input/output or lacking test +examples, and sorts prompts by tokenized length using +Qwen2.5-1.5B-Instruct. The shortest and longest 128 eligible prompts +form disjoint standard-code and longer-prompt-code subsets. Alternating +records in each subset supply 64 training and 64 validation examples. +The longer subset is selected by length, not by fitting a tail +distribution. Run-time filtering then applies the selected model's +tokenizer and prompt cap. + +\begin{table}[htbp] +\centering +\caption{Realized work in the boundary experiment. Qwen, Phi, and Granite +denote Qwen2.5-3B, Phi-4-mini, and Granite-3.3-2B. Lengths are tokens; +means weight completed processes equally. ``Cap batches'' counts batches +with at least one response reaching its cap, not individual responses.} +\label{stab:realized} +\footnotesize +\setlength{\tabcolsep}{3pt} +\begin{tabular}{@{}llrrrrr@{}} +\toprule +Model & Workload & \shortstack{Examples before/\\after filtering} & +\shortstack{Completed\\processes} & \shortstack{Prompt\\mean / max.} & +\shortstack{Response\\mean / max.} & Cap batches \\ +\midrule +Qwen & GSM8K & 32 / 32 & 12 & 108 / 154 & 298 / 830 & 0 / 12 \\ +Qwen & MATH & 64 / 64 & 12 & 104 / 437 & 392 / 512 & 12 / 12 \\ +Qwen & Standard code & 32 / 32 & 12 & 421 / 518 & 286 / 958 & 0 / 12 \\ +Qwen & Longer-prompt code & 32 / 32 & 12 & 720 / 772 & 352 / 1,024 & 7 / 12 \\ +Phi & GSM8K & 32 / 32 & 12 & 78 / 121 & 226 / 2,048 & 2 / 12 \\ +Phi & MATH & 64 / 64 & 12 & 76 / 401 & 367 / 512 & 12 / 12 \\ +Phi & Standard code & 32 / 32 & 12 & 379 / 480 & 326 / 1,024 & 12 / 12 \\ +Phi & Longer-prompt code & 32 / 32 & 12 & 664 / 746 & 412 / 1,024 & 12 / 12 \\ +Granite & GSM8K & 128 / 128 & 15 & 146 / 198 & 285 / 2,048 & 5 / 15 \\ +Granite & MATH & 256 / 252 & 15 & 132 / 460 & 376 / 512 & 15 / 15 \\ +Granite & Standard code & 64 / 47 & 15 & 492 / 588 & 228 / 1,024 & 1 / 15 \\ +Granite & Longer-prompt code & 64 / 63 & 15 & 807 / 902 & 296 / 1,024 & 1 / 15 \\ +\bottomrule +\end{tabular} +\end{table} + +Each completed boundary run executes one global training batch. Larger +Granite preparation subsets do not increase that batch, although filtering +can change which examples reach it. All 156 completions have length +summaries; the sixty failures do not. Missing summaries are not +zero-length work. + +\FloatBarrier +\section{Public transfer-task contract} + +The original transfer suite contains 52 tasks and 400 queries over +94 distinct target configurations. Workload transfer contributes +216 queries, model transfer 144, GPU-count transfer 36, and horizon +transfer four. Reused targets are not independent experiments. +Model transfer holds workload and level fixed; workload transfer holds +model and level fixed. GPU-count transfer uses matched two- and four-GPU +settings; horizon transfer uses the paired one-step and 100-step cases. + +Each task must be processed separately. A target in one task may be a +source in another, so pooling task inputs would reveal target outcomes. +Another seed or supporting control from the target case is not donor +evidence. Additional evidence must be disclosed. The labels are public: +the interface enforces an input contract, not permanent test-set secrecy. + +Scoring retains the confusion matrix, counts, and denominators alongside +accuracy, within-margin recall, and approval precision. The static +interface counts distinct donor processes and actual recorded attempts; +those counts do not measure computation saved. Workload transfer is +already saturated by lookup on this grid, and all four repeated horizon +labels agree. These tasks are reference checks rather than evidence of +prediction difficulty. + +\FloatBarrier +\section{Duration-test settings and schedules} + +The original two cases use the full training split after filtering, +fixed order, global batch 128, five responses per prompt, at most +256 active sequences, and log-probability micro-batch 8. Reservations +are 0.50 below the operating limit, 0.65 near it, and 0.80 for the +historically unsuitable regime. Actor micro-batches are 16 for +Qwen2.5-3B and 4 for Qwen2.5-1.5B, except that the unsuitable regime +uses 16 for both. Actor parameter and optimizer offload are disabled; +generation-state release is enabled. + +Three seeds per regime give eighteen forty-step runs. The sixteen +historical source screens provide three repetitions per regime except +unsuitable MATH, which has one. Logs verify one-step execution despite +an early trace format with a constant step counter. The original +100-step extension retains the below-limit and near-boundary regimes +with three new seeds, 91--93. + +\begin{table}[htbp] +\centering +\caption{Duration settings. Caps are tokens; all final-step operations +listed here are required for completion.} +\label{stab:duration-settings} +\small +\setlength{\tabcolsep}{4pt} +\begin{tabular}{@{}p{0.23\linewidth}p{0.15\linewidth}p{0.10\linewidth}p{0.42\linewidth}@{}} +\toprule +Study and case & Prompt / response cap & Steps & Validation and checkpoints \\ +\midrule +Original Qwen2.5-3B / GSM8K & 256 / 1,024 & 40; 100 & +40-step: neither. 100-step: validation before training and every 20 steps; +checkpoints at 80 and 100. \\ +Original Qwen2.5-1.5B / MATH & 1,792 / 1,024 & 40; 100 & +Same schedule; validation subset at most 128 prompts. \\ +Paired Phi and Granite / GSM8K & 1,024 / 2,048 & 1; 100 & +Validation at 0 and every 20 steps; checkpoints every 25; both at the +final step, including step 1. \\ +Paired Phi and Granite / MATH & 512 / 512 & 1; 100 & +Same schedule; validation subsets 32 prompts for GSM8K and 64 for MATH. \\ +\bottomrule +\end{tabular} +\end{table} + +Both horizons in the paired study use the full training split after +filtering. Completion checks verify the scheduled updates, validation, +and checkpoint writing. Reward and optimization statistics check that +training executed, not that answer quality improved. + +\FloatBarrier +\section{Margin sensitivity and original logging comparison} + +This section checks sensitivity to the operating limit and reports +the original comparison with and without allocator logging. + +\begin{table}[htbp] +\centering +\caption{Retrospective margin sensitivity. The last column counts +Qwen-to-Phi approval errors over targets outside the chosen margin. +Diagnosed memory failures remain unsuitable at every threshold.} +\label{stab:margin} +\small +\setlength{\tabcolsep}{4pt} +\begin{tabular}{@{}p{0.13\linewidth}p{0.13\linewidth}p{0.20\linewidth}p{0.23\linewidth}p{0.18\linewidth}@{}} +\toprule +Capacity allowed & Limit (MiB) & Qwen highest within-margin level & +Phi highest within-margin level & Errors / unsuitable \\ +\midrule +90\% & 36,864 & c2, all workloads & c2, all workloads & 0/12 \\ +92.5\% & 37,888 & c3, all workloads & c2, all workloads & 4/12 \\ +95\% (primary) & 38,912 & c3, all workloads & c2, all workloads & 4/12 \\ +97.5\% & 39,936 & c3, all workloads & c3 GSM8K; c2 otherwise & 3/11 \\ +100\% & 40,960 & c3, all workloads & c3, all workloads & 0/8 \\ +\bottomrule +\end{tabular} +\end{table} + +Granite reaches c3 at the primary limit and c4 without a reserved margin. +At 90\% and 92.5\%, it reaches c3 except on MATH, where it reaches c2. +Qwen and Phi have the same completion boundary at full capacity. The +artifact retains all sixty model--workload--threshold combinations +and thirty directional transfer summaries. + +\begin{table}[htbp] +\centering +\caption{Original allocator-logging comparison. Counts are external-only; +full, each out of three. Peak and time differences are full minus +external-only, averaged over pairs with both completions. A dash is +missing completed-run evidence, not zero.} +\label{stab:logging} +\small +\setlength{\tabcolsep}{4pt} +\begin{tabular}{@{}llrrrrr@{}} +\toprule +Model / workload & Level & Completed & Within margin & +\shortstack{Label\\disagreements} & \shortstack{Peak difference\\(MiB)} & +\shortstack{Time difference\\(s)} \\ +\midrule +Granite / GSM8K & c3 & 3; 3 & 3; 3 & 0/3 & 0.0 & $-15.3$ \\ +Granite / GSM8K & c4 & 3; 3 & 0; 0 & 0/3 & 0.0 & $-6.3$ \\ +Granite / MATH & c3 & 3; 3 & 3; 3 & 0/3 & 0.0 & 4.7 \\ +Granite / MATH & c4 & 3; 3 & 0; 0 & 0/3 & 0.0 & $-8.0$ \\ +Phi / GSM8K & c2 & 3; 3 & 3; 3 & 0/3 & $-6.7$ & 11.0 \\ +Phi / GSM8K & c3 & 3; 3 & 0; 0 & 0/3 & $-13.3$ & $-4.3$ \\ +Phi / MATH & c2 & 3; 3 & 3; 3 & 0/3 & $-149.3$ & 5.0 \\ +Phi / MATH & c3 & 3; 3 & 0; 0 & 0/3 & $-44.7$ & 5.0 \\ +Qwen / GSM8K & c3 & 3; 3 & 3; 3 & 0/3 & 0.0 & $-19.3$ \\ +Qwen / GSM8K & c4 & 0; 0 & 0; 0 & 0/3 & --- & --- \\ +Qwen / MATH & c3 & 3; 3 & 3; 3 & 0/3 & 0.0 & 13.7 \\ +Qwen / MATH & c4 & 0; 0 & 0; 0 & 0/3 & --- & --- \\ +\bottomrule +\end{tabular} +\end{table} + +\FloatBarrier +\section{Estimator specification and retrospective evidence} +\label{sec:estimator-specification} + +This section specifies the baselines used to predict configuration labels +from earlier measurements. +The effective specification combines \path{benchmark/estimation/protocol.json} +with \path{amendment-01.json} in that directory. The amendment replaced +the binary parameter-offload feature with resident actor-shard volume +before the final prediction freeze and new target execution. +Historical outcomes were already public during method design. +The fitting pool contains only the 72 boundary and eighteen four-GPU +configurations. Three folds each withhold all thirty configurations of +one family---Qwen, Phi, or Granite---and fit on the other sixty. +Each configuration combines three eligible distinct seeds; seeds are +not additional test examples. Other controls, the admission panel, +and earlier 7B exploration are excluded. + +\paragraph{Configured size features.} +Let \(P\) count floating-point elements in the checkpoint headers, +which describe stored tensor shapes and types, including floating buffers. +The rank-64 adapter count \(R\) is +64 times the sum of input and output dimensions of the selected +linear decoder-layer weights; embeddings and the output head are excluded. +These counts and architecture dimensions come from +\path{model_metadata.json}, not loaded weights or target measurements. +Let \(G\) be the actor GPU count, \(T\) the generation tensor-parallel +degree (the number of GPUs sharing a generation replica, here one), +\(u\) the generation reservation fraction, and \(C\) device +capacity in bytes. Indicators \(o_p,o_o\) equal one when actor parameters +or optimizer state, respectively, are offloaded. Let \(L,H,V\) denote +decoder-layer count, hidden width, and vocabulary size. + +Let \(b\) be the global prompt batch and \(n\) responses per prompt; +\(s_{\rm model},s_{\rm prompt},s_{\rm response}\) are model, prompt, +and response length caps. Let \(m_a\) be the actor-update micro-batch +and \(m_g,m_r\) the generation-policy and reference log-probability +micro-batches. Define +\[ + s=\min(s_{\rm model},s_{\rm prompt}+s_{\rm response}),\qquad + N=bn/G,\qquad A=s\min(m_a,N),\qquad + Z=s\min\{\max(m_g,m_r),N\}. +\] +Here \(N\) counts configured sequences per GPU; \(A,Z\) are token-volume +caps, not observed response lengths. +Let \(U\) be the largest of three element counts: the largest decoder +block, all floating tensors outside decoder blocks together, and the largest individual +floating tensor. This is a proxy, not a measured FSDP wrapping unit: +the parameter group gathered together during an operation. +With \(D=2^{30}\) bytes per GiB, the ordered feature vector is: + +\begin{center} +\small +\begin{tabular}{@{}lll@{}} +\toprule +Feature & Size proxy & Value in GiB \\ +\midrule +\(x_1\) & Actor shards & \(4(P+R)/(GD)\) \\ +\(x_2\) & Adapter optimizer state & \(8(1-o_o)R/(GD)\) \\ +\(x_3\) & Generation budget & \(uC/D\) \\ +\(x_4\) & Generation replica & \(2(P+R)/(TD)\) \\ +\(x_5\) & Largest model unit & \(4U/D\) \\ +\(x_6\) & Checkpoint-boundary volume & \(2A(L+1)H/D\) \\ +\(x_7\) & Vocabulary-output volume & \(2\max(A,Z)V/D\) \\ +\(x_8\) & Resident actor shards & \((1-o_p)x_1\) \\ +\bottomrule +\end{tabular} +\end{center} + +Four-byte actor storage is FP32 despite BF16 computation; the factors +two and eight represent BF16-sized volumes and two FP32 optimizer-state +arrays, respectively. The shared reference base is counted once. +The checkpoint-boundary feature is an activation-checkpointing proxy for +intermediate representations retained for recomputation, not for model +files written to storage. The vocabulary-output feature describes a +token-by-vocabulary score-array volume. +Features can overlap: the generation budget already covers its replica +and cache. They are predictive covariates, not an exact accounting of +simultaneously live tensors. Model identity, task name, compound-level +index, measured response length, and target outcomes are not numeric inputs. + +\paragraph{Fitting and decisions.} +For the completed training subset \(F_c\), let \(p_i\) be configuration +\(i\)'s maximum external peak over its three completed runs, in GiB. +Writing \(x_{ij}\) for feature \(j\) of configuration \(i\), nonnegative +least squares minimizes squared peak error with predicted peak +\[ + \widehat p_i=\beta_0+\sum_{j=1}^{8}\beta_j x_{ij}/d_j,\qquad + d_j=\max\{1,\max_{i\in F_c}x_{ij}\},\qquad \beta_j\geq0 + \quad(j=0,\ldots,8). +\] +\(\beta_0\) is the intercept, \(\beta_1,\ldots,\beta_8\) the fitted +coefficients, and \(d_j\) the feature divisors; all use only \(F_c\). The solver uses +64-bit floating point and an iteration limit of 10,000. +Failure configurations supply no numerical peak target: neither capacity +nor a surviving seed's peak is substituted. +After conversion to MiB, predicted peaks above 40,960 receive the +memory-failure label; otherwise those above 38,912 receive above-margin, +and the remainder within-margin. +When \(o_p=0\), the additional check +\(4(P+R)/G+uC>C\) overrides this class with memory failure. +It assumes actor shards remain resident at the generation engine's +startup budget check. Passing is not an approval. +The archive retains the peak-only class and whether this guard changed it. +Intercept and coefficients are empirical, not independently identified +or causal memory components. + +The multinomial logistic classifier uses all training labels and a +separate training-only \texttt{StandardScaler}: each feature's training +mean is subtracted and its training standard deviation divides the +result, with divisor one for constant features. +It uses an L2 (squared-coefficient) penalty, inverse regularization strength +\texttt{C=1}, \texttt{class\_weight=balanced}, the \texttt{lbfgs} solver, +10,000 maximum iterations, and random state 20260915. +Balanced weighting gives each class equal total training weight. +The largest class score determines the label; scores are not calibrated +failure probabilities. Convergence warnings or nonfinite fitted parameters +stop fitting. Neither estimator uses a hyperparameter search. + +Donor copying requires the same workload and compound level. Selection +then prefers the same GPU count, the same architecture family, and the +smallest absolute logarithm of the donor-to-target \(P\) ratio, in order; +configuration identifier breaks ties. Outcomes do not choose the donor. + +\paragraph{Retrospective errors and evidence cost.} +Table~\ref{stab:estimator-cost-errors} separates historical acquisition +from the completed cases supporting numerical error. +No startup guard changes a retrospective class. +Regression nevertheless approves all four memory-failing Granite +configurations and four above-margin ones; the logistic classifier +rejects six within-margin Qwen configurations. The main article retains +the complete categorical comparison. Three selected families and overlapping +training pools do not support population confidence intervals. + +\begin{table}[htbp] +\centering +\caption{Held-out-family fitting costs and regression peak errors. +Each fitting pool has sixty configurations and 180 eligible processes; +each selected donor subset has thirty configurations and ninety eligible +processes. Peak errors use only test configurations with three completions. +``Worst under'' is the largest positive observed-minus-predicted peak.} +\label{stab:estimator-cost-errors} +\small +\setlength{\tabcolsep}{4pt} +\begin{tabular}{@{}lrrrrrr@{}} +\toprule +Held out & \shortstack{Completed\\fit cases} & +\shortstack{Pool\\attempts} & \shortstack{Donor\\attempts} & +\shortstack{Completed\\test cases} & \shortstack{Mean absolute\\error (MiB)} & +\shortstack{Worst under\\(MiB)} \\ +\midrule +Qwen & 48 & 213 & 123 & 22 & 1,383 & 1,337 \\ +Phi & 48 & 218 & 95 & 22 & 2,331 & 7,621 \\ +Granite & 44 & 185 & 95 & 26 & 2,631 & 5,738 \\ +\bottomrule +\end{tabular} +\end{table} + +The fit for the twelve prospective Qwen2.5-7B configurations uses all ninety +historical configurations: 270 eligible processes from 308 recorded attempts, +including the acquisition of failure labels. Seventy configurations +supply completed regression targets. The donor rule selects ten distinct +configurations, costing thirty eligible processes and 34 attempts. +The two four-GPU longer-prompt-code targets use disclosed two-GPU donors +because no matching four-GPU code source exists. +Reuse requires no additional source GPU execution; these acquisition +counts are not cold-start savings, target-test costs, or GPU-hours. + +The archive in \path{benchmark/estimation/} retains per-fold coefficients, +divisors, scalers, class order, predictions, confusion matrices, peak-error +coverage, and source identifiers in \path{retrospective.json}; +\path{fitted_model.json} and \path{predictions.json} preserve the all-source +fit and frozen prospective predictions. \path{prediction_freeze.json} +binds the effective input and output hashes and records software versions, +reported numerical thread pools, and fitting/prediction wall time separately. +CPU fitting used Python 3.11.4, NumPy 1.24.3, SciPy 1.10.1, and +scikit-learn 1.3.0. A one-thread limit was requested, but the thread-control +library could not inspect NumPy's numerical backend and emitted a warning. +A later read-only check identified NumPy's OpenBLAS 0.3.21; its thread +limit during fitting was not established. The recorded wall time is +therefore not a controlled single-thread performance result. +The original coefficients and prediction seal were preserved. +Replaying saved coefficients reproduced all predictions and retrospective +scores exactly. A separate fresh-fit check also reproduced them within +the numerical tolerance recorded in +\path{benchmark/estimation/numerical_diagnostic.json}, with identical class labels; +bitwise-identical refitting across numerical builds is not claimed. + +\subsection*{Original larger-model panel: outcomes and missingness} + +The eighteen matched allocations retained all 36 planned seed slots. +Thirty-four training processes launched; the final allocation failed its +allocation-identity check before either process started. +Twenty-three invocations supplied validated generation-engine +memory-admission failures, all during startup. No invocation completed +the requested training step. The other eleven launched invocations were +ineligible for a GPU-memory label: three had a device-ID import error, +one was interrupted during device preflight, and seven four-GPU c2 runs +lost workers under the 192 GiB host-memory limit. +The original attempts remain in the archive and were not replaced. + +Before the remaining, unstarted slots ran, device-identification and import +compatibility were repaired. The failed slots were not repeated, and +the configuration/seed inventory and sealed predictors remained unchanged. +The archive records the execution source and compatibility changes for +each attempt. These setup checks do not by themselves establish +equivalent execution of a complete training step. + +\begin{table}[htbp] +\centering +\caption{Original 7B panel. Each entry gives eligible seeds out of three +and the repeated label. An asterisk marks an unresolved configuration +with at least one validated memory failure. No row contains a completed +training run.} +\label{stab:original-larger-model} +\small +\setlength{\tabcolsep}{8pt} +\begin{tabular}{@{}llcc@{}} +\toprule +Workload & Level & Two task GPUs & Four task GPUs \\ +\midrule +GSM8K & c2 & 3/3; failure & 0/3; unresolved \\ +GSM8K & c3 & 3/3; failure & 3/3; failure \\ +MATH & c2 & 1/3; unresolved* & 0/3; unresolved \\ +MATH & c3 & 2/3; unresolved* & 2/3; unresolved* \\ +Longer-prompt code & c2 & 3/3; failure & 0/3; unresolved \\ +Longer-prompt code & c3 & 3/3; failure & 3/3; failure \\ +\bottomrule +\end{tabular} +\end{table} + +Six of twelve repeated configuration labels are resolved. +Donor copying approves all twelve configurations: six approvals are +observed failures and six are unscored, including three with known +partial failure evidence. Regression approves three configurations, +all unscored. The classifier approves none. +Both fitted methods predict the six resolved failures correctly, but +the resolved subset contains no within-margin or above-margin completion. +Within-margin recall, completed-peak error and completed-pair memory or +runtime differences are therefore unavailable. + +The physical collection used 4.97 task-provisioned GPU-hours and +6.15 GPU-hours of four-device reservation during training invocations. +Actual scheduler allocation spans total 6.47 GPU-hours, including the +nine-second allocation without a training process. These are alternative cost +views, not additive charges. Separate device diagnostics and preflight +jobs used another 0.20 GPU-hours. +Scheduler timestamps have one-second precision. For the interrupted +original allocation, the scheduler's outer job ended three seconds before +its batch step; the separate batch-step spans total twelve GPU-seconds +more than the outer allocation spans. The archive retains both scopes, +the subsecond execution journals, and their reconciliation. + +\subsection*{Separate follow-up with expanded host resources} + +The original panel left every four-GPU c2 configuration unresolved, +including the three configurations that regression approved. +The follow-up therefore retains c2 for all three original workloads, +with seeds 161--163 and separate configuration identifiers. +It uses eighteen fresh invocations in nine matched allocations. +Each allocation reserves four A100 GPUs and 384 GiB of host memory. +The scheduler's \texttt{cpus-per-task} request changes from 32 to 64; +its records show 64 allocated processor units in both panels. +This is a separately specified host-resource condition, not a controlled +estimate of the effect of host memory alone. + +Five pairs run the two-GPU condition first and four run the four-GPU +condition first. Each workload receives both orders. All six two-of-four +device subsets occur; three occur twice, so each device position within +the allocated four-GPU set appears four or five times. +Pair order, period order and device subsets +use separate fixed randomization seeds. +Every training and measurement setting is inherited from the original c2 +condition, including one training step and no validation or checkpoints. +The follow-up permits no retry, replacement, requeue, or further +host-memory increase. Its separate ceiling is nine allocations of +55 minutes each, or 33 reserved GPU-hours. + +All eighteen method--configuration predictions are exact copies of the +original sealed predictions, with only their target identifiers changed. +They retain numerical peaks, donor identities, and startup-check results. +The six donor-copy predictions use five distinct source configurations; +four-GPU code again retains its two-GPU donor. No new outcome enters +fitting, donor selection or a threshold adjustment. +The original prediction seal remains separate from the follow-up's +design freeze in \path{benchmark/host_capacity/}. + +Primary reporting uses the six-configuration inventory and all eighteen +seed slots. Numerical peak errors require three completed invocations +and compare the inherited prediction with the maximum completed peak. +Paired peak and runtime differences require both members to complete. +Host-kill and GPU-failure peaks are not substituted for completed peaks. +No result is pooled with the original panel. + +All eighteen slots launched once. The nine two-GPU invocations have +diagnosed generation-startup memory-admission failures, resolving all +three two-GPU configurations to memory failure. The nine four-GPU +invocations exit with code zero but lack the required one-step terminal +record. They remain unresolved, so no four-GPU repeated label, completed +peak error, or completed-pair numerical difference is available. +No scheduler host-memory exhaustion was recorded in this follow-up. + +The four-GPU traces contain returning actor-update and +weight-synchronization calls. Thus, the missing terminal record is not +evidence that training never started. The saved records do not establish +why the final message is absent; immediate shutdown of the distributed +workers is a possible logging explanation, not a demonstrated cause. +The predeclared completion rule is unchanged, and neither an exit code +nor partial operation traces replace it. + +Donor copy incorrectly approves the three resolved two-GPU targets +and has three unscored four-GPU approvals. Regression correctly predicts +the three resolved failure labels and retains three unscored approvals. +The classifier predicts failure for the two-GPU targets and above-margin +completion for the unresolved four-GPU targets; it approves none. +As in the original +panel, an always-failure rule matches the fitted methods on the +resolved subset. Within-margin recall and completed-peak errors are +undefined, not perfect scores. + +The invocations use 4.89 task-provisioned GPU-hours; weighting every +invocation by the four reserved devices gives 5.61 GPU-hours. +Actual outer scheduler allocations total 5.79 GPU-hours, with complete +accounting for all nine pairs. Batch-step spans have the same total +and are not added. The separate passing device preflight uses another +0.08 GPU-hours. These costs include unresolved execution and are not +pooled with the original panel. + \end{document} diff --git a/release.json b/release.json index 5638a258e684b056edd5d21df0004a2960b76a20..5fdf6211df02d1aa4da32158ac6ce425ad673c40 100644 --- a/release.json +++ b/release.json @@ -1,15 +1,15 @@ { - "release": "admission-20260915-7b534ef", - "source_commit": "7b534ef1d1e902ba490677acdccd8294e59e8bf5", + "release": "submission-20260915-6a7606f", + "source_commit": "6a7606fa72a162b9929d2e73a3175f1ac3ac91a4", "huggingface_repository": "kobzaond/RLVRAMBench", "code_repository": "kobzaond/RLVRAMBench-reproduce", "archive": { - "path": "releases/RLVRAMBench-review-artifact-admission-20260915-7b534ef.tar.zst", - "sha256": "a457d7169afbfb1980ec70a4af215035041d2e05eec9eb090e2c0b7c69e935b2", - "size_bytes": 177940429 + "path": "releases/RLVRAMBench-review-artifact-submission-20260915-6a7606f.tar.zst", + "sha256": "b9b62e1d94da603ba96ddc89dc362d661ca7366c3cfe798dc0a63a4fa8fc26fb", + "size_bytes": 181477577 }, - "manifest_entries": 20145, - "archive_members": 20147, + "manifest_entries": 21374, + "archive_members": 21376, "expected_derived_files": 40, "principal_processes": 588, "supporting_source_processes": 16, @@ -24,7 +24,7 @@ "review_control_attempted_processes": 26, "review_control_usable_blocks": 6, "original_recorded_attempts": 657, - "prior_immutable_release_preserved": "benchmark-20260915-478b657", + "prior_immutable_release_preserved": "admission-20260915-7b534ef", "manuscript_editing_authority": "direct_tex", "benchmark_protocol_version": "1.0", "benchmark_tasks": 52, @@ -41,5 +41,74 @@ "decision_recorded_attempts": 48, "decision_validated_memory_outcomes": 48, "expected_decision_derived_files": 11, - "expected_curated_benchmark_files": 14 + "expected_curated_benchmark_files": 14, + "estimation_protocol_version": "estimation-1.0", + "expected_estimation_derived_files": 13, + "estimation_counts": { + "planned_processes": 36, + "planned_pairs": 18, + "planned_configurations": 12, + "eligible_processes": 23, + "completed_processes": 0, + "unresolved_processes": 13, + "known_launched_processes": 34, + "unknown_launch_status_processes": 0, + "resolved_configurations": 6, + "unresolved_configurations": 6 + }, + "estimation_protocol_sha256": "1694e447e842600612604891e7b3d036ff0791fa84c4bdbeba54ed093feefdeb", + "estimation_prediction_seal_sha256": "5ed714caff922a5109346a4242545d19f2824636d0f52bb0bbd2a09b40ed4666", + "estimation_execution_commits": { + "est-qwen25-7b-code_heavy_tail-c2-s151": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-code_heavy_tail-c2-s152": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-code_heavy_tail-c2-s153": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-code_heavy_tail-c3-s151": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-code_heavy_tail-c3-s152": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-code_heavy_tail-c3-s153": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-gsm8k-c2-s151": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-gsm8k-c2-s152": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-gsm8k-c2-s153": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-gsm8k-c3-s151": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-gsm8k-c3-s152": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-gsm8k-c3-s153": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-math-c2-s151": "0bd659b3f69e2f779dbea84bb2da221bd0ff52db", + "est-qwen25-7b-math-c2-s152": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-math-c2-s153": "0bd659b3f69e2f779dbea84bb2da221bd0ff52db", + "est-qwen25-7b-math-c3-s151": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-math-c3-s152": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-math-c3-s153": "d8e0c3551d8572fb64e3b7126fbd7569a987973f" + }, + "estimation_execution_commits_sha256": "cf70aa7c361f114b2d06a5da5fbbbf5021a13b7d19c24a55023647f5b3c8301a", + "estimation_allocation_accounting_sha256": "1a91571d5bc83040c72a507602c8640faac7d6197608fcac4d17596c7ffc8483", + "expected_host_capacity_derived_files": 13, + "host_capacity_protocol_version": "host-capacity-1.0", + "host_capacity_counts": { + "planned_processes": 18, + "planned_pairs": 9, + "planned_configurations": 6, + "eligible_processes": 9, + "completed_processes": 0, + "unresolved_processes": 9, + "known_launched_processes": 18, + "unknown_launch_status_processes": 0, + "resolved_configurations": 3, + "unresolved_configurations": 3 + }, + "host_capacity_protocol_sha256": "f3f49617a7fca63ddac8923bccb83a15eaf52646e5f3db5df4aa716bb25c1ef6", + "host_capacity_study_freeze_sha256": "80559199c59246d0e4ce52bf268d0375c90fc5168a5246ed900c9266abd951cd", + "host_capacity_original_protocol_sha256": "1694e447e842600612604891e7b3d036ff0791fa84c4bdbeba54ed093feefdeb", + "host_capacity_prediction_seal_sha256": "5ed714caff922a5109346a4242545d19f2824636d0f52bb0bbd2a09b40ed4666", + "host_capacity_execution_commits": { + "host-qwen25-7b-code_heavy_tail-c2-s161": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-code_heavy_tail-c2-s162": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-code_heavy_tail-c2-s163": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-gsm8k-c2-s161": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-gsm8k-c2-s162": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-gsm8k-c2-s163": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-math-c2-s161": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-math-c2-s162": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-math-c2-s163": "120e29a146294722a6668f55a1e7fed2878eb62a" + }, + "host_capacity_execution_commits_sha256": "117b2449b2365db03f5af05aeae2c2943421ab739e5fd3c0ae16e3ce67ac5fcc", + "host_capacity_allocation_accounting_sha256": "8df8040c1cd1608abd10759dd42448d734aa1b501733c926978247d2177f12ee" } diff --git a/releases/RLVRAMBench-release-submission-20260915-6a7606f.json b/releases/RLVRAMBench-release-submission-20260915-6a7606f.json new file mode 100644 index 0000000000000000000000000000000000000000..40c0cfb3669b2568d39e7a6d1e4bd50c499ac54b --- /dev/null +++ b/releases/RLVRAMBench-release-submission-20260915-6a7606f.json @@ -0,0 +1,32 @@ +{ + "archives": [ + { + "path": "/mnt/proj3/open-35-44/verl/releases/rlvrambench/RLVRAMBench-review-artifact-submission-20260915-6a7606f.tar.zst", + "sha256": "b9b62e1d94da603ba96ddc89dc362d661ca7366c3cfe798dc0a63a4fa8fc26fb", + "size_bytes": 181477577 + }, + { + "path": "/mnt/proj3/open-35-44/verl/releases/rlvrambench/RLVRAMBench-submission-scientific-package-submission-20260915-6a7606f.zip", + "sha256": "b66fc69d3f621380b4b70a0613b5315b896d73225972e3b025149704ad3e2072", + "size_bytes": 1959056 + } + ], + "author_supplied_upload_blockers": [ + "corresponding_author_orcid", + "corresponding_author_email", + "funding_and_computing", + "competing_interests", + "ai_disclosure_approved", + "exclusive_submission_confirmed", + "Ond\u0159ej Kobza: credit_roles", + "Ond\u0159ej Kobza: submission_approval", + "Ond\u0159ej Kobza: biography_review", + "Jan \u0160ediv\u00fd: credit_roles", + "Jan \u0160ediv\u00fd: submission_approval", + "Jan \u0160ediv\u00fd: biography_review" + ], + "manifest_entries": 21374, + "review_members": 21376, + "submission_members": 3, + "tag": "submission-20260915-6a7606f" +} diff --git a/releases/RLVRAMBench-review-artifact-submission-20260915-6a7606f.tar.zst b/releases/RLVRAMBench-review-artifact-submission-20260915-6a7606f.tar.zst new file mode 100644 index 0000000000000000000000000000000000000000..344780a7e03262dbc9c9ab7fcdef2ccd883a9e15 --- /dev/null +++ b/releases/RLVRAMBench-review-artifact-submission-20260915-6a7606f.tar.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b9b62e1d94da603ba96ddc89dc362d661ca7366c3cfe798dc0a63a4fa8fc26fb +size 181477577 diff --git a/releases/RLVRAMBench-submission-scientific-package-submission-20260915-6a7606f.zip b/releases/RLVRAMBench-submission-scientific-package-submission-20260915-6a7606f.zip new file mode 100644 index 0000000000000000000000000000000000000000..6951a221526a396f3c449e3781adf6038974ecc5 --- /dev/null +++ b/releases/RLVRAMBench-submission-scientific-package-submission-20260915-6a7606f.zip @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b66fc69d3f621380b4b70a0613b5315b896d73225972e3b025149704ad3e2072 +size 1959056 diff --git a/upload-manifest.json b/upload-manifest.json index 97899665b24acce831d686892ce067d6712b6e50..dbb8a58bcd1b1b84f59d8dcf645ba26a1be2b0ef 100644 --- a/upload-manifest.json +++ b/upload-manifest.json @@ -2,20 +2,20 @@ "algorithm": "sha256", "files": { "BENCHMARK.md": { - "sha256": "09ed7d80bd70e9b548b1cb0f012501f8f66b78acf07d9e6113981d89ab893e4c", - "size_bytes": 12129 + "sha256": "fd9ccf667a854b625faba82f7ac288294838c642cb77dd7857f090fec67f2f95", + "size_bytes": 13070 }, "DATA_DICTIONARY.md": { - "sha256": "f7e2da6e55ef188650b4cef0428c6cdcc9a4986063dca19e4278fe9bbe9b0237", - "size_bytes": 16703 + "sha256": "71c90c5767a36ee80cf022860ff55ed179daa35734ff573f8064645e7a492664", + "size_bytes": 22852 }, "LICENSE": { "sha256": "7518af731dc7cf4f069effaabb5bef12dfb0142becaddbf7adf4ac873adec98f", "size_bytes": 1087 }, "README.md": { - "sha256": "ba5c5921deccd48f2177c84528169a8a95551d4708553ad5c964d0220dc363da", - "size_bytes": 14142 + "sha256": "2fe338de9dcc1548e9347c22946aed42aaeb68e9142a97705a7619bf4090d946", + "size_bytes": 18032 }, "RIGHTS.md": { "sha256": "7c1f948ecd8082ad2d0cbe334e576566d6e26a5ccfdb5ac19b69611044fafe96", @@ -101,6 +101,214 @@ "sha256": "ec270ac1d2207e8654a87bba68f9d7232a4f3788986e7dc85554d1ea8a59592e", "size_bytes": 1098 }, + "benchmark/estimation/README.md": { + "sha256": "e4b9e8cdf2c91342e683d78c5002b6a2eb934b49fdb89c303681db27fc4df85b", + "size_bytes": 6676 + }, + "benchmark/estimation/allocation_accounting.json": { + "sha256": "1a91571d5bc83040c72a507602c8640faac7d6197608fcac4d17596c7ffc8483", + "size_bytes": 33824 + }, + "benchmark/estimation/allocation_accounting_sacct.psv": { + "sha256": "224daf563d5a8dab331c60e73bbfdd21eb296cff1f0473efbe93a4c381c554d2", + "size_bytes": 8074 + }, + "benchmark/estimation/amendment-01.json": { + "sha256": "b02fe0419f865c4cc3d77fb35bb4205a54f3aa9605fdadc7b86a9a11b172d25e", + "size_bytes": 4632 + }, + "benchmark/estimation/execution_commits.json": { + "sha256": "cf70aa7c361f114b2d06a5da5fbbbf5021a13b7d19c24a55023647f5b3c8301a", + "size_bytes": 1443 + }, + "benchmark/estimation/execution_history.json": { + "sha256": "6e45860ed36ec4a5158edb481451e9783ec514539da292d469d4939d4836512c", + "size_bytes": 9570 + }, + "benchmark/estimation/fitted_model.json": { + "sha256": "8b84aa1ae57fe56fd478817dbbce505487cbef6b26652ae2c8f45e9fa5ef5df4", + "size_bytes": 11485 + }, + "benchmark/estimation/matrix.csv": { + "sha256": "1ff175e22c76bf1af3857b4ae57cd4f78bafe0d280a295d696e96815ef85e9c8", + "size_bytes": 16755 + }, + "benchmark/estimation/model_metadata.json": { + "sha256": "2f0ce16c13e72900d2a60f9b7ad28d4dc5ed43d7915d8a88925ba9c5ffd684ad", + "size_bytes": 5599 + }, + "benchmark/estimation/numerical_diagnostic.json": { + "sha256": "44134378fd5eb21f1b2466186a8fc42980a533c8c537096c61936854bab22654", + "size_bytes": 2083 + }, + "benchmark/estimation/prediction_freeze.json": { + "sha256": "5ed714caff922a5109346a4242545d19f2824636d0f52bb0bbd2a09b40ed4666", + "size_bytes": 3004 + }, + "benchmark/estimation/predictions.json": { + "sha256": "96d7f2ae11145f0985c0faf743333d941afbcd3b80966a857b1af0af15306ace", + "size_bytes": 21607 + }, + "benchmark/estimation/protocol.json": { + "sha256": "1694e447e842600612604891e7b3d036ff0791fa84c4bdbeba54ed093feefdeb", + "size_bytes": 11490 + }, + "benchmark/estimation/results/configurations.csv": { + "sha256": "9d37d28c0204df5d45d4a4f5ffecd8290800004f9d4084996c86c5e1b155a0bd", + "size_bytes": 19245 + }, + "benchmark/estimation/results/configurations.json": { + "sha256": "afd92cca168dc114138d0551933bb673160c686e3b781503b262a0c8bd03b773", + "size_bytes": 24620 + }, + "benchmark/estimation/results/configurations_flat.csv": { + "sha256": "18dd5f6ff5a123ae69b9f5ce3ef3c72d05ba51055816a0c49b02a163b6b371e7", + "size_bytes": 3146 + }, + "benchmark/estimation/results/configurations_flat.json": { + "sha256": "04271cb6baffce9523fbed16f5620bcc184a2b52c9baa18936dfdacb82b09bf8", + "size_bytes": 12494 + }, + "benchmark/estimation/results/costs.json": { + "sha256": "b3f4e98408d4afac76271fc22815beb7e1699f06c79f0774d360de23d31d1017", + "size_bytes": 40993 + }, + "benchmark/estimation/results/evaluation.json": { + "sha256": "5a2f2b4677d9d620c1a190edc9d66b61d070d77fef0e09af4eeadf137321ed22", + "size_bytes": 7320 + }, + "benchmark/estimation/results/output_manifest.json": { + "sha256": "7074b04477b92b36f99ca5944702705276c249c118df1ed785aa0759bda09b86", + "size_bytes": 1608 + }, + "benchmark/estimation/results/pairs.csv": { + "sha256": "040e9d135c0a132debc315d62dd6824b5bd826d42ec85056c722b058995d7cd1", + "size_bytes": 1810 + }, + "benchmark/estimation/results/pairs.json": { + "sha256": "f1fc7a443ef39e20bdf22ca8bc7429f7fa7a860c364c471d43bdd52e01805165", + "size_bytes": 7197 + }, + "benchmark/estimation/results/processes.csv": { + "sha256": "9a4e03a34a51a287afb983b93244662f9ac7abf366ba97bd717ba9d371fefdc2", + "size_bytes": 17770 + }, + "benchmark/estimation/results/processes.json": { + "sha256": "d41a8d9a2a24dc753e242e1e29aa3039171beaed8f61d57639b624f39c6f81d6", + "size_bytes": 47060 + }, + "benchmark/estimation/results/provenance.json": { + "sha256": "504d883a6ded17d5c56fd7a05e41378305207e5e331cce1b455095fea8621282", + "size_bytes": 160742 + }, + "benchmark/estimation/results/summary.json": { + "sha256": "ce1fc65b68d51a4b5e5fc8f72e49c0f383694fbc02fadc944c03b4639c741768", + "size_bytes": 321 + }, + "benchmark/estimation/retrospective.json": { + "sha256": "8ada76da63ab91e0b1ca6276a98f2b096979fe6acd7578c1223469d55d399f3a", + "size_bytes": 228053 + }, + "benchmark/estimation/targets.csv": { + "sha256": "f2481278127fa726f24297d81fc9df46e999315fe543167ed398e671fffa522f", + "size_bytes": 5037 + }, + "benchmark/host_capacity/README.md": { + "sha256": "5fc3433ff7acc8b5043471ef769ec0798db26a762ced65fee9d65d8b56ba35ce", + "size_bytes": 4541 + }, + "benchmark/host_capacity/allocation_accounting.json": { + "sha256": "8df8040c1cd1608abd10759dd42448d734aa1b501733c926978247d2177f12ee", + "size_bytes": 16752 + }, + "benchmark/host_capacity/allocation_accounting_sacct.psv": { + "sha256": "5c672842d7ad1ef71e6164ae76ba2e74400c45c2511a66d4139489f34664d2eb", + "size_bytes": 3957 + }, + "benchmark/host_capacity/execution_commits.json": { + "sha256": "117b2449b2365db03f5af05aeae2c2943421ab739e5fd3c0ae16e3ce67ac5fcc", + "size_bytes": 732 + }, + "benchmark/host_capacity/execution_diagnostic.md": { + "sha256": "c783240c6b714fb8225e373f6c71b6eb5248e7d2c340a1f931ac418e11ac5af5", + "size_bytes": 9255 + }, + "benchmark/host_capacity/execution_history.json": { + "sha256": "155ee86ab059a604a709a37d216eb1daa1777071132ad61679b4431e58bec926", + "size_bytes": 3094 + }, + "benchmark/host_capacity/matrix.csv": { + "sha256": "1d2b2af601361a3cb46b65ff3be263685b3f725f2b0e4787a8ff9966084f5332", + "size_bytes": 8931 + }, + "benchmark/host_capacity/predictions.json": { + "sha256": "5ac9cef3c6e7d08036d29315be9cb6220f6bb85fe48a0475f97f2ebce66b9980", + "size_bytes": 10821 + }, + "benchmark/host_capacity/protocol.json": { + "sha256": "f3f49617a7fca63ddac8923bccb83a15eaf52646e5f3db5df4aa716bb25c1ef6", + "size_bytes": 5676 + }, + "benchmark/host_capacity/results/configurations.csv": { + "sha256": "7158a2d61b35f03b457ec9987bdc5fe1e2adec93ecbc93ada04e0bcffd3a456f", + "size_bytes": 9761 + }, + "benchmark/host_capacity/results/configurations.json": { + "sha256": "df070b3e7f853925ae85c07d65e5a72f61c96a18ff548c337567bfe4282e1c47", + "size_bytes": 12334 + }, + "benchmark/host_capacity/results/configurations_flat.csv": { + "sha256": "7c2eee86ed408429abe158a82290b82b6afa2b770ad3b5bf387c7d924f9a1c69", + "size_bytes": 1865 + }, + "benchmark/host_capacity/results/configurations_flat.json": { + "sha256": "d7f475771231662c54d243734c5116765679454d4310d6a19207adeaec816532", + "size_bytes": 6251 + }, + "benchmark/host_capacity/results/costs.json": { + "sha256": "00a63241afae76e18bcd21b81b91f3a9abe6987710ee3e2eaf3c59dfc9657369", + "size_bytes": 21101 + }, + "benchmark/host_capacity/results/evaluation.json": { + "sha256": "ea7bbc7407055ab483797c937524d10748c4b41076dd1bd9bd1b847cee94c327", + "size_bytes": 6625 + }, + "benchmark/host_capacity/results/output_manifest.json": { + "sha256": "47fe58631433062547800641c4e4a4ed5136b661db1c297d6ef24d15c1012709", + "size_bytes": 1605 + }, + "benchmark/host_capacity/results/pairs.csv": { + "sha256": "f1c6ca2025a681b3f5185b7f133ec4f2028a79bfa1e0e3b6a8a0954f4478e5ce", + "size_bytes": 1006 + }, + "benchmark/host_capacity/results/pairs.json": { + "sha256": "860375b5407614a936838f51511396a1df08948a728de641345a40ade5d5fdf5", + "size_bytes": 3603 + }, + "benchmark/host_capacity/results/processes.csv": { + "sha256": "87a1603625747e96f2aa21a338b9ea86b63ade888f249489546387f646ed8dae", + "size_bytes": 9665 + }, + "benchmark/host_capacity/results/processes.json": { + "sha256": "87cf0097ef312be33ca1903052904e412ef24485aea7b348354e01e12bca0ad7", + "size_bytes": 24186 + }, + "benchmark/host_capacity/results/provenance.json": { + "sha256": "72f176399add01f5275d5fc59b5e8da0e1d34c59a56433fe65cb33e1dc82733a", + "size_bytes": 101792 + }, + "benchmark/host_capacity/results/summary.json": { + "sha256": "c8dc099edc466d15e08fae6c9fd708c7a48c1a56fd535fc96c9f8d5424b61ccc", + "size_bytes": 317 + }, + "benchmark/host_capacity/study_freeze.json": { + "sha256": "80559199c59246d0e4ce52bf268d0375c90fc5168a5246ed900c9266abd951cd", + "size_bytes": 610 + }, + "benchmark/host_capacity/targets.csv": { + "sha256": "4aaa0913debe9425ef40664f0554b2ed15980f4cab59bf8223dc86093f313fdc", + "size_bytes": 2885 + }, "benchmark/manifest.json": { "sha256": "d776ada072ca1e0357bc453f337ff0fc740ff825748b59cf8a77062b43ce6441", "size_bytes": 1517 @@ -138,24 +346,24 @@ "size_bytes": 40770 }, "paper/RLVRAMBench_IEEE_Access_source.zip": { - "sha256": "bf0fe6a258cc1d0148cb8bcdb25e33fbc7494335f47ed8b4050502bb76c933b1", - "size_bytes": 1156009 + "sha256": "c7eef7d90561cae3961ec4c98b4cd6d1076ff63b17cf2be2d7604af8016efc1d", + "size_bytes": 1160514 }, "paper/RLVRAMBench_IEEE_Access_submission.pdf": { - "sha256": "2ee28b5e8d999d9af7c3553d4cd3496c38333e615d3c1b092ac8f039ab305549", - "size_bytes": 439029 + "sha256": "9b28d833742a1befddc4b55b67968fb449d469e95b57f3451adeb509a5909865", + "size_bytes": 398486 }, "paper/RLVRAMBench_IEEE_Access_submission.tex": { - "sha256": "1c5a973ca8efa417d22bf9cb1c3296e70636a44724bfde48dd2ccb49bba2852b", - "size_bytes": 105376 + "sha256": "a9f86b81a93a377f47d84a8c8b361fd338679f48839633fca6021e21c57ff233", + "size_bytes": 84872 }, "paper/RLVRAMBench_supplement.pdf": { - "sha256": "3aa5a1852a615541249dbeecfc72fdf59c2787a2b132902dcd476bf9338191df", - "size_bytes": 266688 + "sha256": "493614368da0b65b2bc639fb5722d65cb9f39cdaae3d900b17cd7e063915a1dd", + "size_bytes": 421805 }, "paper/RLVRAMBench_supplement.tex": { - "sha256": "92427fc46634d94eb9a158438bfb36517b0715345cba92d483c1632f61eaaf96", - "size_bytes": 29073 + "sha256": "acd644a3c0b6896ce2d25de339019d491ef60ca80f2caf28690c52780f56a75c", + "size_bytes": 59253 }, "paper/figures/review_instrumentation_batch.pdf": { "sha256": "c80b1e6cd4bf5fcac419a883c46c4c01d765e15a31d1fc9f65d0964356074f60", @@ -190,20 +398,20 @@ "size_bytes": 98350 }, "release.json": { - "sha256": "2969a2636fb268f47d3d39c0086c3f243e54645d27c49a9ab0a9a0bcc39f88ae", - "size_bytes": 1700 + "sha256": "c5e499ae850b59a3a42bc766667d65e43ff22f9a381e11dab5a124082edb2985", + "size_bytes": 6007 }, - "releases/RLVRAMBench-release-admission-20260915-7b534ef.json": { - "sha256": "ef4436229c03911188f1ea05b265d9a76df977375a484ce6f43b82f9ac97fee3", - "size_bytes": 1011 + "releases/RLVRAMBench-release-submission-20260915-6a7606f.json": { + "sha256": "1f212ae1baf3e3777de7b9ac0188d3b2f30045640b1bcbf177176cbf7dff6c80", + "size_bytes": 1166 }, - "releases/RLVRAMBench-review-artifact-admission-20260915-7b534ef.tar.zst": { - "sha256": "a457d7169afbfb1980ec70a4af215035041d2e05eec9eb090e2c0b7c69e935b2", - "size_bytes": 177940429 + "releases/RLVRAMBench-review-artifact-submission-20260915-6a7606f.tar.zst": { + "sha256": "b9b62e1d94da603ba96ddc89dc362d661ca7366c3cfe798dc0a63a4fa8fc26fb", + "size_bytes": 181477577 }, - "releases/RLVRAMBench-submission-scientific-package-admission-20260915-7b534ef.zip": { - "sha256": "fb885e26e7fb16228225bf6cdc3f8281f392890436cc7e68d662ef2e6743b963", - "size_bytes": 1843095 + "releases/RLVRAMBench-submission-scientific-package-submission-20260915-6a7606f.zip": { + "sha256": "b66fc69d3f621380b4b70a0613b5315b896d73225972e3b025149704ad3e2072", + "size_bytes": 1959056 }, "tables/boundary-trials.csv": { "sha256": "6b0bda9727e04be13242e8b3eca63aa8f01a4ff9122899b59d59dd2d03d0df7a", @@ -474,11 +682,11 @@ "size_bytes": 1411 }, "verification/admission-reproduction.json": { - "sha256": "b135bb905a5474f2533fedee41b07602457e6c2b54b3da587ffebff3ff5ea061", - "size_bytes": 1359 + "sha256": "d419c8f387f071b8eab77927d722ba6c60a864f3c76c9d36f00b81bb96e6575f", + "size_bytes": 7178 }, "verification/admission-source-package.json": { - "sha256": "f7d90351425d455e061c7e3fc3a9a30d84c51403f1e822f438a5910544fad0ae", + "sha256": "48c0f52111e5a98edf02cde6b7403ebe7d28025b2a07a60eeaddd170f7d25b88", "size_bytes": 1950 } } diff --git a/verification/admission-reproduction.json b/verification/admission-reproduction.json index 741fe228f15f48aa7aa26ba6435fc82fed545fe6..b3c9095e8d159567ebc28c6c0ac56fc22e40ca8e 100644 --- a/verification/admission-reproduction.json +++ b/verification/admission-reproduction.json @@ -1,9 +1,9 @@ { "status": "passed", - "archive_sha256": "a457d7169afbfb1980ec70a4af215035041d2e05eec9eb090e2c0b7c69e935b2", - "archive_members": 20147, - "manifest_entries_verified": 20145, - "evidence_manifest_sha256": "35bea073f08b43888846b10229928b5c317da0bd8cc2cec0edf65c3347928670", + "archive_sha256": "b9b62e1d94da603ba96ddc89dc362d661ca7366c3cfe798dc0a63a4fa8fc26fb", + "archive_members": 21376, + "manifest_entries_verified": 21374, + "evidence_manifest_sha256": "003986b3df9e1782d1ac481e144e319bb82a5d0725d61d504503ee064b49fbf9", "principal_processes": 588, "supporting_source_processes": 16, "derived_csv_json_files_matched": 40, @@ -19,13 +19,106 @@ "reference_profiles_removed_from_analysis_root": true, "original_project_and_reference_tables_hidden": true, "hidden_paths": [ - "/tmp/rlvrambench-admission-reproduction-7b534ef/reference-profiles", - "/tmp/rlvrambench-admission-reproduction-7b534ef/reference-benchmark", + "/tmp/rlvrambench-final-publication-20260915.a6QcBg/local-reconstruction/reference-profiles", + "/tmp/rlvrambench-final-publication-20260915.a6QcBg/local-reconstruction/reference-benchmark", "/mnt/proj3/open-35-44/verl/rl", - "/home/kobzaond/RLVRAMBench-reproduce/benchmark", + "/mnt/proj3/open-35-44/verl/RLVRAMBench-reproduce-submission-20260915/benchmark", "/tmp/RLVRAMBench-revision-20260915", - "/tmp/rlvrambench-admission-portability.WafYHs" + "/home/kobzaond/RLVRAMBench-reproduce", + "/tmp/RLVRAMBench-reproduce-final-20260915.vvNqzM", + "/tmp/rlvrambench-admission-reproduction-7b534ef", + "/tmp/rlvrambench-backcompat-reproduction-20260915-1525", + "/tmp/rlvrambench-wrapper-tmpdir-smoke.o0h1Jt", + "/tmp/rlvrambench-original-final-collection.PsrKL9", + "/tmp/rlvrambench-collector-host-integration.XJ6iRi", + "/tmp/rlvrambench-host-final-20260915.u6ynp1", + "/tmp/rlvrambench-reproduction-20260914.swuWAw", + "/tmp/rlvrambench-final-build-corrected-20260915.m62jTj", + "/tmp/rlvrambench-ieee-extracted-20260915-status.Agzmen", + "/tmp/rlvrambench-final-publication-20260915.a6QcBg/stage", + "/mnt/proj3/open-35-44/verl/releases/rlvrambench" ], "child_namespace_reference_inaccessibility_checked": true, - "network_namespace_isolation_claimed": false + "network_namespace_isolation_claimed": false, + "estimation_panel_files_matched": 13, + "estimation_counts": { + "planned_processes": 36, + "planned_pairs": 18, + "planned_configurations": 12, + "eligible_processes": 23, + "completed_processes": 0, + "unresolved_processes": 13, + "known_launched_processes": 34, + "unknown_launch_status_processes": 0, + "resolved_configurations": 6, + "unresolved_configurations": 6 + }, + "estimation_prediction_seal_sha256": "5ed714caff922a5109346a4242545d19f2824636d0f52bb0bbd2a09b40ed4666", + "estimation_execution_commits": { + "est-qwen25-7b-code_heavy_tail-c2-s151": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-code_heavy_tail-c2-s152": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-code_heavy_tail-c2-s153": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-code_heavy_tail-c3-s151": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-code_heavy_tail-c3-s152": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-code_heavy_tail-c3-s153": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-gsm8k-c2-s151": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-gsm8k-c2-s152": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-gsm8k-c2-s153": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-gsm8k-c3-s151": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-gsm8k-c3-s152": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-gsm8k-c3-s153": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-math-c2-s151": "0bd659b3f69e2f779dbea84bb2da221bd0ff52db", + "est-qwen25-7b-math-c2-s152": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-math-c2-s153": "0bd659b3f69e2f779dbea84bb2da221bd0ff52db", + "est-qwen25-7b-math-c3-s151": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-math-c3-s152": "d8e0c3551d8572fb64e3b7126fbd7569a987973f", + "est-qwen25-7b-math-c3-s153": "d8e0c3551d8572fb64e3b7126fbd7569a987973f" + }, + "estimation_independent_raw_reconstruction_passed": true, + "estimation_reference_outcomes_hidden_during_analysis": true, + "estimation_prediction_replay_passed": true, + "estimation_prediction_replay_outcomes_hidden": true, + "estimation_prediction_replay_host_outcomes_hidden": true, + "estimation_fresh_refit_compared": true, + "estimation_source_files_verified": 154, + "estimation_output_comparison": "Exact file sets and contents after artifact-root path normalization; each output manifest independently verified against its own bytes.", + "host_capacity_panel_files_matched": 13, + "host_capacity_counts": { + "planned_processes": 18, + "planned_pairs": 9, + "planned_configurations": 6, + "eligible_processes": 9, + "completed_processes": 0, + "unresolved_processes": 9, + "known_launched_processes": 18, + "unknown_launch_status_processes": 0, + "resolved_configurations": 3, + "unresolved_configurations": 3 + }, + "host_capacity_protocol_version": "host-capacity-1.0", + "host_capacity_protocol_sha256": "f3f49617a7fca63ddac8923bccb83a15eaf52646e5f3db5df4aa716bb25c1ef6", + "host_capacity_study_freeze_sha256": "80559199c59246d0e4ce52bf268d0375c90fc5168a5246ed900c9266abd951cd", + "host_capacity_prediction_seal_sha256": "5ed714caff922a5109346a4242545d19f2824636d0f52bb0bbd2a09b40ed4666", + "host_capacity_original_protocol_sha256": "1694e447e842600612604891e7b3d036ff0791fa84c4bdbeba54ed093feefdeb", + "host_capacity_execution_commits": { + "host-qwen25-7b-code_heavy_tail-c2-s161": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-code_heavy_tail-c2-s162": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-code_heavy_tail-c2-s163": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-gsm8k-c2-s161": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-gsm8k-c2-s162": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-gsm8k-c2-s163": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-math-c2-s161": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-math-c2-s162": "120e29a146294722a6668f55a1e7fed2878eb62a", + "host-qwen25-7b-math-c2-s163": "120e29a146294722a6668f55a1e7fed2878eb62a" + }, + "host_capacity_execution_commits_sha256": "117b2449b2365db03f5af05aeae2c2943421ab739e5fd3c0ae16e3ce67ac5fcc", + "host_capacity_allocation_accounting_sha256": "8df8040c1cd1608abd10759dd42448d734aa1b501733c926978247d2177f12ee", + "host_capacity_independent_raw_reconstruction_passed": true, + "host_capacity_reference_outcomes_hidden_during_analysis": true, + "host_capacity_prediction_inheritance_replay_passed": true, + "host_capacity_prediction_replay_outcomes_hidden": true, + "host_capacity_original_prediction_replay_passed": true, + "host_capacity_fresh_refit_performed": false, + "host_capacity_source_files_verified": 154, + "host_capacity_output_comparison": "Separate exact file sets and contents after artifact-root path normalization; each output manifest independently checked against its own bytes." } diff --git a/verification/admission-source-package.json b/verification/admission-source-package.json index 8bbcb7773b82eb8b7699e57cddcdf46f1818ef4b..b73b0d7d4afff205e03d3f9fe546169e242d72db 100644 --- a/verification/admission-source-package.json +++ b/verification/admission-source-package.json @@ -2,14 +2,14 @@ "status": "clean_submission_manuscript_verified", "workflow_placeholders_in_article": 0, "journal_submission_tasks_separate": true, - "pdf_pages": 19, - "pdf_size_bytes": 439029, - "pdf_sha256": "2ee28b5e8d999d9af7c3553d4cd3496c38333e615d3c1b092ac8f039ab305549", - "tex_sha256": "1c5a973ca8efa417d22bf9cb1c3296e70636a44724bfde48dd2ccb49bba2852b", - "log_sha256": "c0facef2ad509419c0ae8f04f0646297fb2bb9f1e09eb45145cd17cdfb7c565b", - "table_captions_in_order": 18, + "pdf_pages": 16, + "pdf_size_bytes": 398486, + "pdf_sha256": "9b28d833742a1befddc4b55b67968fb449d469e95b57f3451adeb509a5909865", + "tex_sha256": "a9f86b81a93a377f47d84a8c8b361fd338679f48839633fca6021e21c57ff233", + "log_sha256": "0a7297b903630ca67a4a9bc8519ed0a94a34fca0c644b6bda2e79bc64d9b04be", + "table_captions_in_order": 13, "figure_captions_in_order": 2, - "compiled_references": 29, + "compiled_references": 25, "both_authors_and_orcids_in_pdf": true, "unmodified_official_template_files_verified": 47, "missing_glyph_warnings": 0, @@ -38,10 +38,10 @@ "source_bundle_verified_after_extraction": true, "upload_performed": false, "remote_push_performed": false, - "supplement_tex_sha256": "92427fc46634d94eb9a158438bfb36517b0715345cba92d483c1632f61eaaf96", - "supplement_pdf_sha256": "3aa5a1852a615541249dbeecfc72fdf59c2787a2b132902dcd476bf9338191df", + "supplement_tex_sha256": "acd644a3c0b6896ce2d25de339019d491ef60ca80f2caf28690c52780f56a75c", + "supplement_pdf_sha256": "493614368da0b65b2bc639fb5722d65cb9f39cdaae3d900b17cd7e063915a1dd", "source_bundle_files": 57, - "source_bundle_sha256": "bf0fe6a258cc1d0148cb8bcdb25e33fbc7494335f47ed8b4050502bb76c933b1", + "source_bundle_sha256": "c7eef7d90561cae3961ec4c98b4cd6d1076ff63b17cf2be2d7604af8016efc1d", "source_bundle_integrity_passed": true, "fresh_supplement_recompile_pdf_text_identical": true, "fresh_source_recompile_pdf_text_identical": true,