RLVRAMBench / benchmark /host_capacity /execution_history.json
kobzaond's picture
Release audited measurement study and larger-model evidence
d8321f8 verified
Raw History Blame Contribute Delete
3.09 kB
{
"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"
}
}