Add fuzzy TopK sparsity diagnostics ladder

.gitattributes

@@ -767,3 +767,45 @@ fuzzy-topk-three-arm-ladder/60m-double-fuzzy-h16d-q0p25d-k4d-l10-3tpp-bs16-lr2e-
 fuzzy-topk-three-arm-ladder/60m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
 fuzzy-topk-three-arm-ladder/60m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
 fuzzy-topk-three-arm-ladder/60m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/125m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/125m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1337/fuzzy_sparsity.rigvec filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/125m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1337/fuzzy_sparsity_lossy.rigvec filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/125m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/125m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1338/fuzzy_sparsity.rigvec filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/125m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1338/fuzzy_sparsity_lossy.rigvec filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/125m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/125m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1339/fuzzy_sparsity.rigvec filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/125m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1339/fuzzy_sparsity_lossy.rigvec filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1337/fuzzy_sparsity.rigvec filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1337/fuzzy_sparsity_lossy.rigvec filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1337/training.riglog filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1338/fuzzy_sparsity.rigvec filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1338/fuzzy_sparsity_lossy.rigvec filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1338/training.riglog filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1339/fuzzy_sparsity.rigvec filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1339/fuzzy_sparsity_lossy.rigvec filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1339/training.riglog filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/500m-fuzzy-h16d-k4d-l16-2tpp-bs16-lr2e-8-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/500m-fuzzy-h16d-k4d-l16-2tpp-bs16-lr2e-8-s1337/fuzzy_sparsity.rigvec filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/500m-fuzzy-h16d-k4d-l16-2tpp-bs16-lr2e-8-s1337/fuzzy_sparsity_lossy.rigvec filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/500m-fuzzy-h16d-k4d-l16-2tpp-bs16-lr2e-8-s1337/training.riglog filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/500m-fuzzy-h16d-k4d-l16-2tpp-bs16-lr2e-8-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/500m-fuzzy-h16d-k4d-l16-2tpp-bs16-lr2e-8-s1338/fuzzy_sparsity.rigvec filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/500m-fuzzy-h16d-k4d-l16-2tpp-bs16-lr2e-8-s1338/fuzzy_sparsity_lossy.rigvec filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/500m-fuzzy-h16d-k4d-l16-2tpp-bs16-lr2e-8-s1338/training.riglog filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/500m-fuzzy-h16d-k4d-l16-2tpp-bs16-lr2e-8-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/500m-fuzzy-h16d-k4d-l16-2tpp-bs16-lr2e-8-s1339/fuzzy_sparsity.rigvec filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/500m-fuzzy-h16d-k4d-l16-2tpp-bs16-lr2e-8-s1339/fuzzy_sparsity_lossy.rigvec filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/500m-fuzzy-h16d-k4d-l16-2tpp-bs16-lr2e-8-s1339/training.riglog filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/60m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1337/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/60m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1337/fuzzy_sparsity.rigvec filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/60m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1337/fuzzy_sparsity_lossy.rigvec filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/60m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1338/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/60m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1338/fuzzy_sparsity.rigvec filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/60m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1338/fuzzy_sparsity_lossy.rigvec filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/60m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1339/diagnostics.riglog filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/60m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1339/fuzzy_sparsity.rigvec filter=lfs diff=lfs merge=lfs -text
+fuzzy-topk-sparsity-diagnostics-ladder/60m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1339/fuzzy_sparsity_lossy.rigvec filter=lfs diff=lfs merge=lfs -text

README.md

@@ -12,7 +12,7 @@ pretty_name: rig training logs
 # rig training logs
 
 Complete training logs for every GPT pretraining run in
-[honglu2875/rig](https://github.com/honglu2875/rig) — 451 runs across nineteen
+[honglu2875/rig](https://github.com/honglu2875/rig) — 463 runs across twenty
 studies, at full recorded resolution. Loss and learning-rate curves at every
 optimizer step; per-layer parameter, gradient, and update statistics at every
 diagnostic step. The raw logs themselves are not downsampled.
@@ -31,7 +31,10 @@ their canonical GitHub locations.
     result.json          configuration, final metrics, provenance
     metrics.json         compact final metrics
     validation.csv       held-out loss
+    fuzzy_sparsity.rigvec        optional full per-feature vector history
+    fuzzy_sparsity_lossy.rigvec  optional full-feature, widened-time companion
   records.jsonl          one ledger line per run
+  lossy_records.jsonl    optional companion hashes and retained steps
   snapshot.json.gz       compact selected curves, for lightweight consumers
   full.json.gz           every recorded point, loaded explicitly by the browser
 ```
@@ -82,6 +85,7 @@ and the dashboards show them beside each run.
 | `moe-gumbel-local-125M` | TPU v4 — 4 processes, 16 chips |
 | `sparse-autoencoder-eqflop-60M` | TPU v4 — 4 processes, 16 chips |
 | `fuzzy-topk-three-arm-ladder` | TPU v4 — 4 processes, 16 chips |
+| `fuzzy-topk-sparsity-diagnostics-ladder` | TPU v4 — 4 processes, 16 chips |
 
 `batch-sweep-500M` is the only study spanning two chip types, and the split
 follows its 5- against 20-tokens-per-parameter arms. Those are separate
@@ -111,6 +115,13 @@ log.series("grad.l2_norm", "block", 7)         # per-layer, from diagnostics
 from the definitions the writer uses, so a reader in another language can be
 built without reading the Python.
 
+`.rigvec` is the dense companion for metrics with a feature axis too large for
+the scalar column table. Its fixed records are
+`step int32 + float32[metric, layer, feature]`; the header records permanent
+metric ids, shape, group size, tokens per step, and FLOPs per token. It is
+memory-mapped by `rig.vectorlog.read_vector_log`, so multi-gigabyte histories
+can be reduced one capture at a time.
+
 ---
 
 Every dashboard here, the runs behind it, and the command that reproduces it.
@@ -121,7 +132,7 @@ exact.
 ## The logs live on HuggingFace
 
 **[huggingface.co/datasets/quintic/rig-logs](https://huggingface.co/datasets/quintic/rig-logs)**
-— 451 runs across nineteen studies, laid out as `<study>/<run-name>/`, at full
+— 463 runs across twenty studies, laid out as `<study>/<run-name>/`, at full
 recorded resolution. That is the archive of record; its
 [dataset card](https://huggingface.co/datasets/quintic/rig-logs/blob/main/README.md)
 mirrors the GitHub report catalog and adds archive and reproduction metadata.
@@ -166,6 +177,7 @@ by-product:
 | gumbel-local-moe | exact endpoints + mechanism reductions | — | 0.03 MB |
 | sparse-autoencoder-eqflop | exact endpoints + compute derivation | — | 0.02 MB |
 | fuzzy-topk-three-arm-ladder | exact endpoints + paired-seed/compute tables | — | 0.03 MB |
+| fuzzy-topk-sparsity-diagnostics | exact mechanism reductions + 40.19 GiB raw vectors | — | 0.02 MB |
 
 The two large ones carry layer detail because gradient spikes are visible in
 it, and studying them is the point. This is deliberate discretion, not a
@@ -228,17 +240,23 @@ total active matrix FLOPs. This dataset retains every complete trajectory,
 both browser payloads, exact source/config/data provenance, and the explicit
 incomplete-250M evidence boundary.
 
+The [fuzzy TopK sparsity report](https://github.com/honglu2875/rig/blob/research/fuzzy-topk-autoencoder/docs/reports/fuzzy-topk-sparsity-diagnostics.html)
+is a static mechanism study over twelve fuzzy-only runs from 60M through 500M.
+This dataset retains 40.19 GiB of cadence-10 full-neuron vectors, 0.891 GiB of
+exact widening-step companions, and the one-off interactive dead-layer,
+ridgeline, histogram, and positive-quantile views.
+
 Which metrics get charted is a declared list in `rig/report.py`, separate from
 the metric registry, because how a quantity should be drawn is a judgement the
-registry cannot make. Everything so far is a line against the time axis; a
-distribution rather than a scalar — a routing histogram, say — wants bars
-against expert index and would arrive as a new chart kind rather than being
-bent into a timeline.
+registry cannot make. Ordinary metrics remain lines against time or layer. The
+fuzzy sparsity study is the one explicit exception: zero-aware log-frequency
+histograms and positive quantiles render as dedicated histogram, ridgeline,
+and heatmap views rather than as scalar timelines.
 
 ## The study browser
 
 [`study-browser.html`](https://github.com/honglu2875/rig/blob/main/docs/reports/study-browser.html)
-carries no run data at all — about 64 KB. It
+carries no run data at all — about 80 KB. It
 lists the studies, renders each one's card from the dataset, and fetches only
 that study's overview (0.05–1.1 MB) when you pick one. A separately labelled
 action loads `full.json.gz`, containing every recorded point, and states its
@@ -283,6 +301,7 @@ study that mixes the two topologies.
 | [gumbel-local-moe](https://github.com/honglu2875/rig/blob/research/gumbel-local-moe/docs/reports/gumbel-local-moe.html) | 8 | 125M | Gumbel-routed local MoE steps × seed | [`moe-gumbel-local-125M`](https://huggingface.co/datasets/quintic/rig-logs/tree/main/moe-gumbel-local-125M) |
 | [sparse-autoencoder-eqflop](https://github.com/honglu2875/rig/blob/research/sparse-autoencoder/docs/reports/sparse-autoencoder-eqflop.html) | 13 | 60M geometry | dictionary width × retained width at equal algorithmic FLOPs | [`sparse-autoencoder-eqflop-60M`](https://huggingface.co/datasets/quintic/rig-logs/tree/main/sparse-autoencoder-eqflop-60M) |
 | [fuzzy-topk-three-arm-ladder](https://github.com/honglu2875/rig/blob/research/fuzzy-topk-autoencoder/docs/reports/fuzzy-topk-three-arm-ladder.html) | 24 | 60M/125M/250M | dense vs fuzzy TopK vs double-fuzzy at matched active FLOPs | [`fuzzy-topk-three-arm-ladder`](https://huggingface.co/datasets/quintic/rig-logs/tree/main/fuzzy-topk-three-arm-ladder) |
+| [fuzzy-topk-sparsity-diagnostics](https://github.com/honglu2875/rig/blob/research/fuzzy-topk-autoencoder/docs/reports/fuzzy-topk-sparsity-diagnostics.html) | 12 | 60M/125M/250M/500M | per-feature fuzzy TopK activity over training | [`fuzzy-topk-sparsity-diagnostics-ladder`](https://huggingface.co/datasets/quintic/rig-logs/tree/main/fuzzy-topk-sparsity-diagnostics-ladder) |
 | [transfer-charts](https://github.com/honglu2875/rig/blob/main/docs/reports/transfer-charts.html) | — | — | derived figures, not a run dashboard | — |
 
 Each study also carries a compact `snapshot.json.gz` (0.05–1.1 MB of thinned
@@ -706,6 +725,38 @@ contains every endpoint, design/throughput table, interpretation, and
 limitation. This study's own card contains the source/config/data hashes,
 canonical manifest links, exclusion, and reproduction command.
 
+## fuzzy-topk-sparsity-diagnostics.html
+
+Twelve verified fuzzy-only runs record full per-feature winner and positive
+activation behavior from 60M through 500M. Every block's `H=16D` stored
+features are observed on a 131,072-token global batch at step 1, every ten
+steps, and final. The raw `.rigvec` files total 40.19 GiB. Exact full-feature
+companions keep every capture through step 200, then 300, 500, 900, 1700, and
+doubled gaps plus final, totaling 0.891 GiB.
+
+The observer is trajectory-neutral: all nine 60M–250M `training.riglog` files
+are byte-identical to the same-tier, same-seed uninstrumented fuzzy runs in the
+three-arm study. Almost every feature activates by step 1 and all do by step
+10, but persistent inactivity is created later. At final, the fraction absent
+from ten consecutive sampled batches averages 34.33%, 35.76%, 33.87%, and
+36.34% over layers and seeds at 60M, 125M, 250M, and 500M.
+
+The effect is layer-structured. Block 1's dead fraction rises from 69.69% to
+85.62% across the ladder while its normalized within-group winner entropy falls
+from 0.424 to 0.103; later blocks recover activity. The
+[static report](https://github.com/honglu2875/rig/blob/research/fuzzy-topk-autoencoder/docs/reports/fuzzy-topk-sparsity-diagnostics.html)
+summarizes these reductions. The study browser carries the one-off tier/seed/
+layer selectors, dead heatmap, activation-frequency ridgeline, draggable-step
+histogram, and positive-frequency quantiles.
+
+“Dead” is defined on sampled observer batches and is not proof that a feature
+never fires elsewhere in the corpus. The study measures under-use; it does not
+test a revival, pruning, or reallocation intervention. `records.jsonl` retains
+the final 500M seed's notebook-dependency-only `pyproject.toml`/`uv.lock` dirt;
+trainer, shared Python tree, config, data, runtime, and topology identities are
+unchanged. The study card inside the folder gives exact endpoints, hashes,
+commands, metric definitions, limitations, and loading examples.
+
 ## transfer-charts.html
 
 Not a run dashboard. It is a self-contained derived visualization in the

fuzzy-topk-sparsity-diagnostics-ladder/125m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1337/diagnostics.riglog

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c76ce3a3561c9ec8dacc95382e027cc5b0d2058e5db409635a906d9565340b96
+size 512740

fuzzy-topk-sparsity-diagnostics-ladder/125m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1337/fuzzy_sparsity.rigvec

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f50eb0698169aca5045ad10b0a3636632a7af98f4360b54d1dcdaeef91276c20
+size 841648004

fuzzy-topk-sparsity-diagnostics-ladder/125m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1337/fuzzy_sparsity_lossy.rigvec

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:747a2b8211685e5edcab1f62e06b0583afb571b76e54f42c2d59f462eba527c8
+size 48660644

fuzzy-topk-sparsity-diagnostics-ladder/125m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1337/metrics.json

@@ -0,0 +1,256 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.riglog",
+    "fuzzy_sparsity": "fuzzy_sparsity.rigvec",
+    "training_curve": "training.riglog",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "context_preset": "8k",
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 640,
+      "heads": 10,
+      "layers": 11,
+      "mlp_activation": "fuzzy_topk_relu",
+      "mlp_mult": 16,
+      "mlp_top_k": 2560,
+      "normalization": "rms_norm",
+      "parameterization": "completep_fixed_tpp_v1",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "125m",
+      "vocab_size": 50304
+    },
+    "model_id": "fuzzy-topk-autoencoder-gpt-v1-family",
+    "sequence_length": 8192,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 3.591915398836136,
+      "perplexity": 36.30354514272369,
+      "scored_tokens": 1048576,
+      "seconds": 0.517010869923979
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "3d77f3dec34a89456ba8284f3d7d149016d9913126698de06073a72e1ff1a550",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "context_preset": "8k",
+      "path": "dev.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "final_predictions": 1048576,
+          "probe_predictions": 0,
+          "val_every": 0,
+          "val_probe_batches": 0
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "document_masking": true,
+          "loss_backend": "tiled",
+          "sparse_mlp_backend": "choicewise",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10,
+          "sparsity_diagnostics_every": 10
+        },
+        "model": {
+          "d_model": 640,
+          "heads": 10,
+          "layers": 11,
+          "mlp_activation": "fuzzy_topk_relu",
+          "mlp_mult": 16,
+          "mlp_top_k": 2560,
+          "normalization": "rms_norm",
+          "parameterization": "completep_fixed_tpp_v1",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "125m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.00390625,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.00390625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 466,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_ratio": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 0.9166666666666666,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "ladder_data_multiplier": 1.0,
+          "name": "completep_fixed_tpp_v1",
+          "width_multiplier": 1.6666666666666667
+        },
+        "training": {
+          "batch_size": 16,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 8192,
+          "steps": 4656,
+          "target_tokens_per_parameter": 2.6913446720594294,
+          "tokens_per_parameter": 2.6913446720594294,
+          "train_tokens": 610271232
+        }
+      },
+      "schema_version": 6,
+      "sha256": "35ebd3ca279a100bf7c76299955bbcb664c4f1cc959e21f5c332270d82aad7a4"
+    },
+    "context_preset": "8k",
+    "document_masking": true,
+    "loss_backend": "tiled",
+    "sparse_mlp_backend": "choicewise",
+    "vocab_tile_size": 2048,
+    "weight_decay_policy": "weights_and_embeddings_only_v2"
+  },
+  "metrics": {
+    "achieved_tflops": 1084.1801071023378,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.00390625,
+    "diagnostic_compile_seconds": 64.66830940404907,
+    "diagnostic_point_count": 467,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 1176245951036129280,
+    "eval_compile_seconds": 13.15049199713394,
+    "final_validation_seconds": 0.517010869923979,
+    "flop_accounting": {
+      "by_site": {
+        "_choicewise_fuzzy_topk_mlp": 7086696038400,
+        "dot_general": 3033331138560,
+        "tpu_flash_causal_attention_bwd_dkv": 1889785610240,
+        "tpu_flash_causal_attention_bwd_dq": 1889785610240,
+        "tpu_flash_causal_attention_fwd": 1889785610240
+      },
+      "elementwise_per_sequence": 6690029490,
+      "matmul_per_sequence": 15789384007680,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 1927415040,
+    "mfu_estimate": 0.24640456979598588,
+    "model_tier": "125m",
+    "parameter_count": 226753280,
+    "parameters": 226753280,
+    "schedule_steps": 4656,
+    "sparsity_diagnostic_compile_seconds": 12.556366964941844,
+    "sparsity_diagnostic_point_count": 467,
+    "sparsity_diagnostics_every": 10,
+    "stop_after_step": null,
+    "target_tokens_per_parameter": 2.6913446720594294,
+    "tokens_per_parameter": 2.6913446720594294,
+    "tokens_per_second": 562504.7457875693,
+    "tokens_processed": 610271232,
+    "total_compile_seconds": 155.68586288695224,
+    "train_compile_seconds": 65.31069452082738,
+    "train_loss": 3.641507625579834,
+    "train_seconds": 1084.9174812659621,
+    "training_data_epochs": 0.07724972079725992,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 4656,
+    "training_token_budget": 610271232,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 3.591915398836136,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1337,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

fuzzy-topk-sparsity-diagnostics-ladder/125m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1337/result.json

@@ -0,0 +1,256 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.riglog",
+    "fuzzy_sparsity": "fuzzy_sparsity.rigvec",
+    "training_curve": "training.riglog",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "context_preset": "8k",
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 640,
+      "heads": 10,
+      "layers": 11,
+      "mlp_activation": "fuzzy_topk_relu",
+      "mlp_mult": 16,
+      "mlp_top_k": 2560,
+      "normalization": "rms_norm",
+      "parameterization": "completep_fixed_tpp_v1",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "125m",
+      "vocab_size": 50304
+    },
+    "model_id": "fuzzy-topk-autoencoder-gpt-v1-family",
+    "sequence_length": 8192,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 3.591915398836136,
+      "perplexity": 36.30354514272369,
+      "scored_tokens": 1048576,
+      "seconds": 0.517010869923979
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "3d77f3dec34a89456ba8284f3d7d149016d9913126698de06073a72e1ff1a550",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "context_preset": "8k",
+      "path": "dev.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "final_predictions": 1048576,
+          "probe_predictions": 0,
+          "val_every": 0,
+          "val_probe_batches": 0
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "document_masking": true,
+          "loss_backend": "tiled",
+          "sparse_mlp_backend": "choicewise",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10,
+          "sparsity_diagnostics_every": 10
+        },
+        "model": {
+          "d_model": 640,
+          "heads": 10,
+          "layers": 11,
+          "mlp_activation": "fuzzy_topk_relu",
+          "mlp_mult": 16,
+          "mlp_top_k": 2560,
+          "normalization": "rms_norm",
+          "parameterization": "completep_fixed_tpp_v1",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "125m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.00390625,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.00390625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 466,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_ratio": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 0.9166666666666666,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "ladder_data_multiplier": 1.0,
+          "name": "completep_fixed_tpp_v1",
+          "width_multiplier": 1.6666666666666667
+        },
+        "training": {
+          "batch_size": 16,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 8192,
+          "steps": 4656,
+          "target_tokens_per_parameter": 2.6913446720594294,
+          "tokens_per_parameter": 2.6913446720594294,
+          "train_tokens": 610271232
+        }
+      },
+      "schema_version": 6,
+      "sha256": "35ebd3ca279a100bf7c76299955bbcb664c4f1cc959e21f5c332270d82aad7a4"
+    },
+    "context_preset": "8k",
+    "document_masking": true,
+    "loss_backend": "tiled",
+    "sparse_mlp_backend": "choicewise",
+    "vocab_tile_size": 2048,
+    "weight_decay_policy": "weights_and_embeddings_only_v2"
+  },
+  "metrics": {
+    "achieved_tflops": 1084.1801071023378,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.00390625,
+    "diagnostic_compile_seconds": 64.66830940404907,
+    "diagnostic_point_count": 467,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 1176245951036129280,
+    "eval_compile_seconds": 13.15049199713394,
+    "final_validation_seconds": 0.517010869923979,
+    "flop_accounting": {
+      "by_site": {
+        "_choicewise_fuzzy_topk_mlp": 7086696038400,
+        "dot_general": 3033331138560,
+        "tpu_flash_causal_attention_bwd_dkv": 1889785610240,
+        "tpu_flash_causal_attention_bwd_dq": 1889785610240,
+        "tpu_flash_causal_attention_fwd": 1889785610240
+      },
+      "elementwise_per_sequence": 6690029490,
+      "matmul_per_sequence": 15789384007680,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 1927415040,
+    "mfu_estimate": 0.24640456979598588,
+    "model_tier": "125m",
+    "parameter_count": 226753280,
+    "parameters": 226753280,
+    "schedule_steps": 4656,
+    "sparsity_diagnostic_compile_seconds": 12.556366964941844,
+    "sparsity_diagnostic_point_count": 467,
+    "sparsity_diagnostics_every": 10,
+    "stop_after_step": null,
+    "target_tokens_per_parameter": 2.6913446720594294,
+    "tokens_per_parameter": 2.6913446720594294,
+    "tokens_per_second": 562504.7457875693,
+    "tokens_processed": 610271232,
+    "total_compile_seconds": 155.68586288695224,
+    "train_compile_seconds": 65.31069452082738,
+    "train_loss": 3.641507625579834,
+    "train_seconds": 1084.9174812659621,
+    "training_data_epochs": 0.07724972079725992,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 4656,
+    "training_token_budget": 610271232,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 3.591915398836136,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1337,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

fuzzy-topk-sparsity-diagnostics-ladder/125m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1337/validation.csv

@@ -0,0 +1,2 @@
+step,tokens_processed,kind,domain,validation_tokens,validation_loss,perplexity,validation_seconds,canonical
+4656,610271232,fineweb,fineweb,1048576,3.591915398836136,36.30354514272369,0.517010869923979,true

fuzzy-topk-sparsity-diagnostics-ladder/125m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1338/diagnostics.riglog

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d61d77acd38a1d2ca4d21dc07a2f69fa9f659ac47b707966f8b494d4b536a190
+size 512740

fuzzy-topk-sparsity-diagnostics-ladder/125m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1338/fuzzy_sparsity.rigvec

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:74997501b57e7d4563b194e998c974476763b7882c7377c21a059d9fbc10bb3f
+size 841648004

fuzzy-topk-sparsity-diagnostics-ladder/125m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1338/fuzzy_sparsity_lossy.rigvec

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c7d84886e72b5c16e159bb5f9ce770a894b08859dc894b2d7ec1e9c750fb5f9b
+size 48660644

fuzzy-topk-sparsity-diagnostics-ladder/125m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1338/metrics.json

@@ -0,0 +1,256 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.riglog",
+    "fuzzy_sparsity": "fuzzy_sparsity.rigvec",
+    "training_curve": "training.riglog",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "context_preset": "8k",
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 640,
+      "heads": 10,
+      "layers": 11,
+      "mlp_activation": "fuzzy_topk_relu",
+      "mlp_mult": 16,
+      "mlp_top_k": 2560,
+      "normalization": "rms_norm",
+      "parameterization": "completep_fixed_tpp_v1",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "125m",
+      "vocab_size": 50304
+    },
+    "model_id": "fuzzy-topk-autoencoder-gpt-v1-family",
+    "sequence_length": 8192,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 3.581060439348221,
+      "perplexity": 35.91160273893408,
+      "scored_tokens": 1048576,
+      "seconds": 0.5136202480643988
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "3d77f3dec34a89456ba8284f3d7d149016d9913126698de06073a72e1ff1a550",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "context_preset": "8k",
+      "path": "dev.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "final_predictions": 1048576,
+          "probe_predictions": 0,
+          "val_every": 0,
+          "val_probe_batches": 0
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "document_masking": true,
+          "loss_backend": "tiled",
+          "sparse_mlp_backend": "choicewise",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10,
+          "sparsity_diagnostics_every": 10
+        },
+        "model": {
+          "d_model": 640,
+          "heads": 10,
+          "layers": 11,
+          "mlp_activation": "fuzzy_topk_relu",
+          "mlp_mult": 16,
+          "mlp_top_k": 2560,
+          "normalization": "rms_norm",
+          "parameterization": "completep_fixed_tpp_v1",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "125m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.00390625,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.00390625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 466,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_ratio": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 0.9166666666666666,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "ladder_data_multiplier": 1.0,
+          "name": "completep_fixed_tpp_v1",
+          "width_multiplier": 1.6666666666666667
+        },
+        "training": {
+          "batch_size": 16,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 8192,
+          "steps": 4656,
+          "target_tokens_per_parameter": 2.6913446720594294,
+          "tokens_per_parameter": 2.6913446720594294,
+          "train_tokens": 610271232
+        }
+      },
+      "schema_version": 6,
+      "sha256": "35ebd3ca279a100bf7c76299955bbcb664c4f1cc959e21f5c332270d82aad7a4"
+    },
+    "context_preset": "8k",
+    "document_masking": true,
+    "loss_backend": "tiled",
+    "sparse_mlp_backend": "choicewise",
+    "vocab_tile_size": 2048,
+    "weight_decay_policy": "weights_and_embeddings_only_v2"
+  },
+  "metrics": {
+    "achieved_tflops": 1083.86685823526,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.00390625,
+    "diagnostic_compile_seconds": 64.37414500606246,
+    "diagnostic_point_count": 467,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 1176245951036129280,
+    "eval_compile_seconds": 13.222909380914643,
+    "final_validation_seconds": 0.5136202480643988,
+    "flop_accounting": {
+      "by_site": {
+        "_choicewise_fuzzy_topk_mlp": 7086696038400,
+        "dot_general": 3033331138560,
+        "tpu_flash_causal_attention_bwd_dkv": 1889785610240,
+        "tpu_flash_causal_attention_bwd_dq": 1889785610240,
+        "tpu_flash_causal_attention_fwd": 1889785610240
+      },
+      "elementwise_per_sequence": 6690029490,
+      "matmul_per_sequence": 15789384007680,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 1927415040,
+    "mfu_estimate": 0.24633337687165,
+    "model_tier": "125m",
+    "parameter_count": 226753280,
+    "parameters": 226753280,
+    "schedule_steps": 4656,
+    "sparsity_diagnostic_compile_seconds": 11.524073926033452,
+    "sparsity_diagnostic_point_count": 467,
+    "sparsity_diagnostics_every": 10,
+    "stop_after_step": null,
+    "target_tokens_per_parameter": 2.6913446720594294,
+    "tokens_per_parameter": 2.6913446720594294,
+    "tokens_per_second": 562342.2229989759,
+    "tokens_processed": 610271232,
+    "total_compile_seconds": 154.58197558182292,
+    "train_compile_seconds": 65.46084726881236,
+    "train_loss": 3.6206488609313965,
+    "train_seconds": 1085.2310337740928,
+    "training_data_epochs": 0.07724972079725992,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 4656,
+    "training_token_budget": 610271232,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 3.581060439348221,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1338,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

fuzzy-topk-sparsity-diagnostics-ladder/125m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1338/result.json

@@ -0,0 +1,256 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.riglog",
+    "fuzzy_sparsity": "fuzzy_sparsity.rigvec",
+    "training_curve": "training.riglog",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "context_preset": "8k",
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 640,
+      "heads": 10,
+      "layers": 11,
+      "mlp_activation": "fuzzy_topk_relu",
+      "mlp_mult": 16,
+      "mlp_top_k": 2560,
+      "normalization": "rms_norm",
+      "parameterization": "completep_fixed_tpp_v1",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "125m",
+      "vocab_size": 50304
+    },
+    "model_id": "fuzzy-topk-autoencoder-gpt-v1-family",
+    "sequence_length": 8192,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 3.581060439348221,
+      "perplexity": 35.91160273893408,
+      "scored_tokens": 1048576,
+      "seconds": 0.5136202480643988
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "3d77f3dec34a89456ba8284f3d7d149016d9913126698de06073a72e1ff1a550",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "context_preset": "8k",
+      "path": "dev.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "final_predictions": 1048576,
+          "probe_predictions": 0,
+          "val_every": 0,
+          "val_probe_batches": 0
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "document_masking": true,
+          "loss_backend": "tiled",
+          "sparse_mlp_backend": "choicewise",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10,
+          "sparsity_diagnostics_every": 10
+        },
+        "model": {
+          "d_model": 640,
+          "heads": 10,
+          "layers": 11,
+          "mlp_activation": "fuzzy_topk_relu",
+          "mlp_mult": 16,
+          "mlp_top_k": 2560,
+          "normalization": "rms_norm",
+          "parameterization": "completep_fixed_tpp_v1",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "125m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.00390625,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.00390625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 466,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_ratio": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 0.9166666666666666,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "ladder_data_multiplier": 1.0,
+          "name": "completep_fixed_tpp_v1",
+          "width_multiplier": 1.6666666666666667
+        },
+        "training": {
+          "batch_size": 16,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 8192,
+          "steps": 4656,
+          "target_tokens_per_parameter": 2.6913446720594294,
+          "tokens_per_parameter": 2.6913446720594294,
+          "train_tokens": 610271232
+        }
+      },
+      "schema_version": 6,
+      "sha256": "35ebd3ca279a100bf7c76299955bbcb664c4f1cc959e21f5c332270d82aad7a4"
+    },
+    "context_preset": "8k",
+    "document_masking": true,
+    "loss_backend": "tiled",
+    "sparse_mlp_backend": "choicewise",
+    "vocab_tile_size": 2048,
+    "weight_decay_policy": "weights_and_embeddings_only_v2"
+  },
+  "metrics": {
+    "achieved_tflops": 1083.86685823526,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.00390625,
+    "diagnostic_compile_seconds": 64.37414500606246,
+    "diagnostic_point_count": 467,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 1176245951036129280,
+    "eval_compile_seconds": 13.222909380914643,
+    "final_validation_seconds": 0.5136202480643988,
+    "flop_accounting": {
+      "by_site": {
+        "_choicewise_fuzzy_topk_mlp": 7086696038400,
+        "dot_general": 3033331138560,
+        "tpu_flash_causal_attention_bwd_dkv": 1889785610240,
+        "tpu_flash_causal_attention_bwd_dq": 1889785610240,
+        "tpu_flash_causal_attention_fwd": 1889785610240
+      },
+      "elementwise_per_sequence": 6690029490,
+      "matmul_per_sequence": 15789384007680,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 1927415040,
+    "mfu_estimate": 0.24633337687165,
+    "model_tier": "125m",
+    "parameter_count": 226753280,
+    "parameters": 226753280,
+    "schedule_steps": 4656,
+    "sparsity_diagnostic_compile_seconds": 11.524073926033452,
+    "sparsity_diagnostic_point_count": 467,
+    "sparsity_diagnostics_every": 10,
+    "stop_after_step": null,
+    "target_tokens_per_parameter": 2.6913446720594294,
+    "tokens_per_parameter": 2.6913446720594294,
+    "tokens_per_second": 562342.2229989759,
+    "tokens_processed": 610271232,
+    "total_compile_seconds": 154.58197558182292,
+    "train_compile_seconds": 65.46084726881236,
+    "train_loss": 3.6206488609313965,
+    "train_seconds": 1085.2310337740928,
+    "training_data_epochs": 0.07724972079725992,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 4656,
+    "training_token_budget": 610271232,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 3.581060439348221,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1338,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

fuzzy-topk-sparsity-diagnostics-ladder/125m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1338/validation.csv

@@ -0,0 +1,2 @@
+step,tokens_processed,kind,domain,validation_tokens,validation_loss,perplexity,validation_seconds,canonical
+4656,610271232,fineweb,fineweb,1048576,3.581060439348221,35.91160273893408,0.5136202480643988,true

fuzzy-topk-sparsity-diagnostics-ladder/125m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1339/diagnostics.riglog

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:62b57e17b6624f065ac0b0fb5dcb7a7fc7c2b9d3670f2faec885a0538fcbbe09
+size 512740

fuzzy-topk-sparsity-diagnostics-ladder/125m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1339/fuzzy_sparsity.rigvec

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:af03bbd8c3aa08f7f406368d502c226bf71b24f9be1d43eff74af7c7dc5a081f
+size 841648004

fuzzy-topk-sparsity-diagnostics-ladder/125m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1339/fuzzy_sparsity_lossy.rigvec

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f1f249d2f9c878bacc42332ab759f4f308f1c1ffbe872a0543ccaa3fd88b16e
+size 48660644

fuzzy-topk-sparsity-diagnostics-ladder/125m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1339/metrics.json

@@ -0,0 +1,256 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.riglog",
+    "fuzzy_sparsity": "fuzzy_sparsity.rigvec",
+    "training_curve": "training.riglog",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "context_preset": "8k",
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 640,
+      "heads": 10,
+      "layers": 11,
+      "mlp_activation": "fuzzy_topk_relu",
+      "mlp_mult": 16,
+      "mlp_top_k": 2560,
+      "normalization": "rms_norm",
+      "parameterization": "completep_fixed_tpp_v1",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "125m",
+      "vocab_size": 50304
+    },
+    "model_id": "fuzzy-topk-autoencoder-gpt-v1-family",
+    "sequence_length": 8192,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 3.584172636270523,
+      "perplexity": 36.023540814770165,
+      "scored_tokens": 1048576,
+      "seconds": 0.5274990880861878
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "3d77f3dec34a89456ba8284f3d7d149016d9913126698de06073a72e1ff1a550",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "context_preset": "8k",
+      "path": "dev.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "final_predictions": 1048576,
+          "probe_predictions": 0,
+          "val_every": 0,
+          "val_probe_batches": 0
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "document_masking": true,
+          "loss_backend": "tiled",
+          "sparse_mlp_backend": "choicewise",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10,
+          "sparsity_diagnostics_every": 10
+        },
+        "model": {
+          "d_model": 640,
+          "heads": 10,
+          "layers": 11,
+          "mlp_activation": "fuzzy_topk_relu",
+          "mlp_mult": 16,
+          "mlp_top_k": 2560,
+          "normalization": "rms_norm",
+          "parameterization": "completep_fixed_tpp_v1",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "125m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.00390625,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.00390625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 466,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_ratio": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 0.9166666666666666,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "ladder_data_multiplier": 1.0,
+          "name": "completep_fixed_tpp_v1",
+          "width_multiplier": 1.6666666666666667
+        },
+        "training": {
+          "batch_size": 16,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 8192,
+          "steps": 4656,
+          "target_tokens_per_parameter": 2.6913446720594294,
+          "tokens_per_parameter": 2.6913446720594294,
+          "train_tokens": 610271232
+        }
+      },
+      "schema_version": 6,
+      "sha256": "35ebd3ca279a100bf7c76299955bbcb664c4f1cc959e21f5c332270d82aad7a4"
+    },
+    "context_preset": "8k",
+    "document_masking": true,
+    "loss_backend": "tiled",
+    "sparse_mlp_backend": "choicewise",
+    "vocab_tile_size": 2048,
+    "weight_decay_policy": "weights_and_embeddings_only_v2"
+  },
+  "metrics": {
+    "achieved_tflops": 1083.9200870397517,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.00390625,
+    "diagnostic_compile_seconds": 63.24943733587861,
+    "diagnostic_point_count": 467,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 1176245951036129280,
+    "eval_compile_seconds": 12.438345218077302,
+    "final_validation_seconds": 0.5274990880861878,
+    "flop_accounting": {
+      "by_site": {
+        "_choicewise_fuzzy_topk_mlp": 7086696038400,
+        "dot_general": 3033331138560,
+        "tpu_flash_causal_attention_bwd_dkv": 1889785610240,
+        "tpu_flash_causal_attention_bwd_dq": 1889785610240,
+        "tpu_flash_causal_attention_fwd": 1889785610240
+      },
+      "elementwise_per_sequence": 6690029490,
+      "matmul_per_sequence": 15789384007680,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 1927415040,
+    "mfu_estimate": 0.2463454743272163,
+    "model_tier": "125m",
+    "parameter_count": 226753280,
+    "parameters": 226753280,
+    "schedule_steps": 4656,
+    "sparsity_diagnostic_compile_seconds": 11.386971961008385,
+    "sparsity_diagnostic_point_count": 467,
+    "sparsity_diagnostics_every": 10,
+    "stop_after_step": null,
+    "target_tokens_per_parameter": 2.6913446720594294,
+    "tokens_per_parameter": 2.6913446720594294,
+    "tokens_per_second": 562369.8396790302,
+    "tokens_processed": 610271232,
+    "total_compile_seconds": 135.26415866799653,
+    "train_compile_seconds": 48.18940415303223,
+    "train_loss": 3.5505714416503906,
+    "train_seconds": 1085.1777405920438,
+    "training_data_epochs": 0.07724972079725992,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 4656,
+    "training_token_budget": 610271232,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 3.584172636270523,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1339,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

fuzzy-topk-sparsity-diagnostics-ladder/125m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1339/result.json

@@ -0,0 +1,256 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.riglog",
+    "fuzzy_sparsity": "fuzzy_sparsity.rigvec",
+    "training_curve": "training.riglog",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "context_preset": "8k",
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 640,
+      "heads": 10,
+      "layers": 11,
+      "mlp_activation": "fuzzy_topk_relu",
+      "mlp_mult": 16,
+      "mlp_top_k": 2560,
+      "normalization": "rms_norm",
+      "parameterization": "completep_fixed_tpp_v1",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "125m",
+      "vocab_size": 50304
+    },
+    "model_id": "fuzzy-topk-autoencoder-gpt-v1-family",
+    "sequence_length": 8192,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 3.584172636270523,
+      "perplexity": 36.023540814770165,
+      "scored_tokens": 1048576,
+      "seconds": 0.5274990880861878
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "3d77f3dec34a89456ba8284f3d7d149016d9913126698de06073a72e1ff1a550",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "context_preset": "8k",
+      "path": "dev.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "final_predictions": 1048576,
+          "probe_predictions": 0,
+          "val_every": 0,
+          "val_probe_batches": 0
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "document_masking": true,
+          "loss_backend": "tiled",
+          "sparse_mlp_backend": "choicewise",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10,
+          "sparsity_diagnostics_every": 10
+        },
+        "model": {
+          "d_model": 640,
+          "heads": 10,
+          "layers": 11,
+          "mlp_activation": "fuzzy_topk_relu",
+          "mlp_mult": 16,
+          "mlp_top_k": 2560,
+          "normalization": "rms_norm",
+          "parameterization": "completep_fixed_tpp_v1",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "125m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.00390625,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.00390625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 466,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_ratio": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 0.9166666666666666,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "ladder_data_multiplier": 1.0,
+          "name": "completep_fixed_tpp_v1",
+          "width_multiplier": 1.6666666666666667
+        },
+        "training": {
+          "batch_size": 16,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 8192,
+          "steps": 4656,
+          "target_tokens_per_parameter": 2.6913446720594294,
+          "tokens_per_parameter": 2.6913446720594294,
+          "train_tokens": 610271232
+        }
+      },
+      "schema_version": 6,
+      "sha256": "35ebd3ca279a100bf7c76299955bbcb664c4f1cc959e21f5c332270d82aad7a4"
+    },
+    "context_preset": "8k",
+    "document_masking": true,
+    "loss_backend": "tiled",
+    "sparse_mlp_backend": "choicewise",
+    "vocab_tile_size": 2048,
+    "weight_decay_policy": "weights_and_embeddings_only_v2"
+  },
+  "metrics": {
+    "achieved_tflops": 1083.9200870397517,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.00390625,
+    "diagnostic_compile_seconds": 63.24943733587861,
+    "diagnostic_point_count": 467,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 1176245951036129280,
+    "eval_compile_seconds": 12.438345218077302,
+    "final_validation_seconds": 0.5274990880861878,
+    "flop_accounting": {
+      "by_site": {
+        "_choicewise_fuzzy_topk_mlp": 7086696038400,
+        "dot_general": 3033331138560,
+        "tpu_flash_causal_attention_bwd_dkv": 1889785610240,
+        "tpu_flash_causal_attention_bwd_dq": 1889785610240,
+        "tpu_flash_causal_attention_fwd": 1889785610240
+      },
+      "elementwise_per_sequence": 6690029490,
+      "matmul_per_sequence": 15789384007680,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 1927415040,
+    "mfu_estimate": 0.2463454743272163,
+    "model_tier": "125m",
+    "parameter_count": 226753280,
+    "parameters": 226753280,
+    "schedule_steps": 4656,
+    "sparsity_diagnostic_compile_seconds": 11.386971961008385,
+    "sparsity_diagnostic_point_count": 467,
+    "sparsity_diagnostics_every": 10,
+    "stop_after_step": null,
+    "target_tokens_per_parameter": 2.6913446720594294,
+    "tokens_per_parameter": 2.6913446720594294,
+    "tokens_per_second": 562369.8396790302,
+    "tokens_processed": 610271232,
+    "total_compile_seconds": 135.26415866799653,
+    "train_compile_seconds": 48.18940415303223,
+    "train_loss": 3.5505714416503906,
+    "train_seconds": 1085.1777405920438,
+    "training_data_epochs": 0.07724972079725992,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 4656,
+    "training_token_budget": 610271232,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 3.584172636270523,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1339,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

fuzzy-topk-sparsity-diagnostics-ladder/125m-fuzzy-h16d-k4d-l11-3tpp-bs16-lr2e-8-s1339/validation.csv

@@ -0,0 +1,2 @@
+step,tokens_processed,kind,domain,validation_tokens,validation_loss,perplexity,validation_seconds,canonical
+4656,610271232,fineweb,fineweb,1048576,3.584172636270523,36.023540814770165,0.5274990880861878,true

fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1337/diagnostics.riglog

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f6c350e8a3a3b73522c617c30bad2dfcd13c42ee48ec6fe6150300deba4ba0da
+size 1232408

fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1337/fuzzy_sparsity.rigvec

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9a300d96d0c1a4bc069975ce48f5fbca9c3e0e94fd86110360cc23e05a4e516d
+size 3025014512

fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1337/fuzzy_sparsity_lossy.rigvec

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f88d3aba739b15e5f9481deee7081550aa183bc6683f01192c4cc72eacf5119f
+size 89915560

fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1337/metrics.json

@@ -0,0 +1,256 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.riglog",
+    "fuzzy_sparsity": "fuzzy_sparsity.rigvec",
+    "training_curve": "training.riglog",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "context_preset": "8k",
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 896,
+      "heads": 14,
+      "layers": 14,
+      "mlp_activation": "fuzzy_topk_relu",
+      "mlp_mult": 16,
+      "mlp_top_k": 3584,
+      "normalization": "rms_norm",
+      "parameterization": "completep_fixed_tpp_v1",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "250m",
+      "vocab_size": 50304
+    },
+    "model_id": "fuzzy-topk-autoencoder-gpt-v1-family",
+    "sequence_length": 8192,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 3.314135879278183,
+      "perplexity": 27.498621586202635,
+      "scored_tokens": 1048576,
+      "seconds": 0.9843828401062638
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "e06781c1c2978705df2ad6d078d005991aaaf8908efa50bfd764dd157326f324",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "context_preset": "8k",
+      "path": "dev.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "final_predictions": 1048576,
+          "probe_predictions": 0,
+          "val_every": 0,
+          "val_probe_batches": 0
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "document_masking": true,
+          "loss_backend": "tiled",
+          "sparse_mlp_backend": "choicewise",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10,
+          "sparsity_diagnostics_every": 10
+        },
+        "model": {
+          "d_model": 896,
+          "heads": 14,
+          "layers": 14,
+          "mlp_activation": "fuzzy_topk_relu",
+          "mlp_mult": 16,
+          "mlp_top_k": 3584,
+          "normalization": "rms_norm",
+          "parameterization": "completep_fixed_tpp_v1",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "250m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.00390625,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.00390625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 940,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_ratio": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 1.1666666666666667,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "ladder_data_multiplier": 1.0,
+          "name": "completep_fixed_tpp_v1",
+          "width_multiplier": 2.3333333333333335
+        },
+        "training": {
+          "batch_size": 16,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 8192,
+          "steps": 9402,
+          "target_tokens_per_parameter": 2.489304879893371,
+          "tokens_per_parameter": 2.489304879893371,
+          "train_tokens": 1232338944
+        }
+      },
+      "schema_version": 6,
+      "sha256": "35ebd3ca279a100bf7c76299955bbcb664c4f1cc959e21f5c332270d82aad7a4"
+    },
+    "context_preset": "8k",
+    "document_masking": true,
+    "loss_backend": "tiled",
+    "sparse_mlp_backend": "choicewise",
+    "vocab_tile_size": 2048,
+    "weight_decay_policy": "weights_and_embeddings_only_v2"
+  },
+  "metrics": {
+    "achieved_tflops": 1122.39495655022,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.00390625,
+    "diagnostic_compile_seconds": 76.33016821881756,
+    "diagnostic_point_count": 942,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 4963669455340044288,
+    "eval_compile_seconds": 12.449057026999071,
+    "final_validation_seconds": 0.9843828401062638,
+    "flop_accounting": {
+      "by_site": {
+        "_choicewise_fuzzy_topk_mlp": 17678085390336,
+        "dot_general": 5216252461056,
+        "tpu_flash_causal_attention_bwd_dkv": 3367254360064,
+        "tpu_flash_causal_attention_bwd_dq": 3367254360064,
+        "tpu_flash_causal_attention_fwd": 3367254360064
+      },
+      "elementwise_per_sequence": 8887879538,
+      "matmul_per_sequence": 32996100931584,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 4027844352,
+    "mfu_estimate": 0.25508976285232277,
+    "model_tier": "250m",
+    "parameter_count": 495053440,
+    "parameters": 495053440,
+    "schedule_steps": 9402,
+    "sparsity_diagnostic_compile_seconds": 14.994110438041389,
+    "sparsity_diagnostic_point_count": 942,
+    "sparsity_diagnostics_every": 10,
+    "stop_after_step": null,
+    "target_tokens_per_parameter": 2.489304879893371,
+    "tokens_per_parameter": 2.489304879893371,
+    "tokens_per_second": 278658.9695286766,
+    "tokens_processed": 1232338944,
+    "total_compile_seconds": 164.92245301185176,
+    "train_compile_seconds": 61.149117327993736,
+    "train_loss": 3.3761191368103027,
+    "train_seconds": 4422.391090027988,
+    "training_data_epochs": 0.15599267073364215,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 9402,
+    "training_token_budget": 1232338944,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 3.314135879278183,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1337,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1337/result.json

@@ -0,0 +1,256 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.riglog",
+    "fuzzy_sparsity": "fuzzy_sparsity.rigvec",
+    "training_curve": "training.riglog",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "context_preset": "8k",
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 896,
+      "heads": 14,
+      "layers": 14,
+      "mlp_activation": "fuzzy_topk_relu",
+      "mlp_mult": 16,
+      "mlp_top_k": 3584,
+      "normalization": "rms_norm",
+      "parameterization": "completep_fixed_tpp_v1",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "250m",
+      "vocab_size": 50304
+    },
+    "model_id": "fuzzy-topk-autoencoder-gpt-v1-family",
+    "sequence_length": 8192,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 3.314135879278183,
+      "perplexity": 27.498621586202635,
+      "scored_tokens": 1048576,
+      "seconds": 0.9843828401062638
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "e06781c1c2978705df2ad6d078d005991aaaf8908efa50bfd764dd157326f324",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "context_preset": "8k",
+      "path": "dev.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "final_predictions": 1048576,
+          "probe_predictions": 0,
+          "val_every": 0,
+          "val_probe_batches": 0
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "document_masking": true,
+          "loss_backend": "tiled",
+          "sparse_mlp_backend": "choicewise",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10,
+          "sparsity_diagnostics_every": 10
+        },
+        "model": {
+          "d_model": 896,
+          "heads": 14,
+          "layers": 14,
+          "mlp_activation": "fuzzy_topk_relu",
+          "mlp_mult": 16,
+          "mlp_top_k": 3584,
+          "normalization": "rms_norm",
+          "parameterization": "completep_fixed_tpp_v1",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "250m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.00390625,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.00390625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 940,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_ratio": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 1.1666666666666667,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "ladder_data_multiplier": 1.0,
+          "name": "completep_fixed_tpp_v1",
+          "width_multiplier": 2.3333333333333335
+        },
+        "training": {
+          "batch_size": 16,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 8192,
+          "steps": 9402,
+          "target_tokens_per_parameter": 2.489304879893371,
+          "tokens_per_parameter": 2.489304879893371,
+          "train_tokens": 1232338944
+        }
+      },
+      "schema_version": 6,
+      "sha256": "35ebd3ca279a100bf7c76299955bbcb664c4f1cc959e21f5c332270d82aad7a4"
+    },
+    "context_preset": "8k",
+    "document_masking": true,
+    "loss_backend": "tiled",
+    "sparse_mlp_backend": "choicewise",
+    "vocab_tile_size": 2048,
+    "weight_decay_policy": "weights_and_embeddings_only_v2"
+  },
+  "metrics": {
+    "achieved_tflops": 1122.39495655022,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.00390625,
+    "diagnostic_compile_seconds": 76.33016821881756,
+    "diagnostic_point_count": 942,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 4963669455340044288,
+    "eval_compile_seconds": 12.449057026999071,
+    "final_validation_seconds": 0.9843828401062638,
+    "flop_accounting": {
+      "by_site": {
+        "_choicewise_fuzzy_topk_mlp": 17678085390336,
+        "dot_general": 5216252461056,
+        "tpu_flash_causal_attention_bwd_dkv": 3367254360064,
+        "tpu_flash_causal_attention_bwd_dq": 3367254360064,
+        "tpu_flash_causal_attention_fwd": 3367254360064
+      },
+      "elementwise_per_sequence": 8887879538,
+      "matmul_per_sequence": 32996100931584,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 4027844352,
+    "mfu_estimate": 0.25508976285232277,
+    "model_tier": "250m",
+    "parameter_count": 495053440,
+    "parameters": 495053440,
+    "schedule_steps": 9402,
+    "sparsity_diagnostic_compile_seconds": 14.994110438041389,
+    "sparsity_diagnostic_point_count": 942,
+    "sparsity_diagnostics_every": 10,
+    "stop_after_step": null,
+    "target_tokens_per_parameter": 2.489304879893371,
+    "tokens_per_parameter": 2.489304879893371,
+    "tokens_per_second": 278658.9695286766,
+    "tokens_processed": 1232338944,
+    "total_compile_seconds": 164.92245301185176,
+    "train_compile_seconds": 61.149117327993736,
+    "train_loss": 3.3761191368103027,
+    "train_seconds": 4422.391090027988,
+    "training_data_epochs": 0.15599267073364215,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 9402,
+    "training_token_budget": 1232338944,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 3.314135879278183,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1337,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1337/training.riglog

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:34e549a8d940eb2afcd495773e93aa9a2071943cca85bce4736980d9898e670d
+size 150536

fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1337/validation.csv

@@ -0,0 +1,2 @@
+step,tokens_processed,kind,domain,validation_tokens,validation_loss,perplexity,validation_seconds,canonical
+9402,1232338944,fineweb,fineweb,1048576,3.314135879278183,27.498621586202635,0.9843828401062638,true

fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1338/diagnostics.riglog

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0e90ea4b61a38334d46f127cc46689466bc48505961e625f0d3afe109c91430e
+size 1232408

fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1338/fuzzy_sparsity.rigvec

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:228bb4a2438541a9a3cbc86ed3d3d9383279cf361af9c4c7b3566f20669427c7
+size 3025014512

fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1338/fuzzy_sparsity_lossy.rigvec

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2c6eb8512c8cb5867d9390626a5d99a60d515bdb5ac4518d8b365f9e3d6ce9c7
+size 89915560

fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1338/metrics.json

@@ -0,0 +1,256 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.riglog",
+    "fuzzy_sparsity": "fuzzy_sparsity.rigvec",
+    "training_curve": "training.riglog",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "context_preset": "8k",
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 896,
+      "heads": 14,
+      "layers": 14,
+      "mlp_activation": "fuzzy_topk_relu",
+      "mlp_mult": 16,
+      "mlp_top_k": 3584,
+      "normalization": "rms_norm",
+      "parameterization": "completep_fixed_tpp_v1",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "250m",
+      "vocab_size": 50304
+    },
+    "model_id": "fuzzy-topk-autoencoder-gpt-v1-family",
+    "sequence_length": 8192,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 3.3040049374103546,
+      "perplexity": 27.221441070036004,
+      "scored_tokens": 1048576,
+      "seconds": 0.9745558970607817
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "e06781c1c2978705df2ad6d078d005991aaaf8908efa50bfd764dd157326f324",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "context_preset": "8k",
+      "path": "dev.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "final_predictions": 1048576,
+          "probe_predictions": 0,
+          "val_every": 0,
+          "val_probe_batches": 0
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "document_masking": true,
+          "loss_backend": "tiled",
+          "sparse_mlp_backend": "choicewise",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10,
+          "sparsity_diagnostics_every": 10
+        },
+        "model": {
+          "d_model": 896,
+          "heads": 14,
+          "layers": 14,
+          "mlp_activation": "fuzzy_topk_relu",
+          "mlp_mult": 16,
+          "mlp_top_k": 3584,
+          "normalization": "rms_norm",
+          "parameterization": "completep_fixed_tpp_v1",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "250m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.00390625,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.00390625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 940,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_ratio": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 1.1666666666666667,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "ladder_data_multiplier": 1.0,
+          "name": "completep_fixed_tpp_v1",
+          "width_multiplier": 2.3333333333333335
+        },
+        "training": {
+          "batch_size": 16,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 8192,
+          "steps": 9402,
+          "target_tokens_per_parameter": 2.489304879893371,
+          "tokens_per_parameter": 2.489304879893371,
+          "train_tokens": 1232338944
+        }
+      },
+      "schema_version": 6,
+      "sha256": "35ebd3ca279a100bf7c76299955bbcb664c4f1cc959e21f5c332270d82aad7a4"
+    },
+    "context_preset": "8k",
+    "document_masking": true,
+    "loss_backend": "tiled",
+    "sparse_mlp_backend": "choicewise",
+    "vocab_tile_size": 2048,
+    "weight_decay_policy": "weights_and_embeddings_only_v2"
+  },
+  "metrics": {
+    "achieved_tflops": 1122.3941557798357,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.00390625,
+    "diagnostic_compile_seconds": 81.67842828086577,
+    "diagnostic_point_count": 942,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 4963669455340044288,
+    "eval_compile_seconds": 12.553831937955692,
+    "final_validation_seconds": 0.9745558970607817,
+    "flop_accounting": {
+      "by_site": {
+        "_choicewise_fuzzy_topk_mlp": 17678085390336,
+        "dot_general": 5216252461056,
+        "tpu_flash_causal_attention_bwd_dkv": 3367254360064,
+        "tpu_flash_causal_attention_bwd_dq": 3367254360064,
+        "tpu_flash_causal_attention_fwd": 3367254360064
+      },
+      "elementwise_per_sequence": 8887879538,
+      "matmul_per_sequence": 32996100931584,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 4027844352,
+    "mfu_estimate": 0.2550895808590536,
+    "model_tier": "250m",
+    "parameter_count": 495053440,
+    "parameters": 495053440,
+    "schedule_steps": 9402,
+    "sparsity_diagnostic_compile_seconds": 15.16223825304769,
+    "sparsity_diagnostic_point_count": 942,
+    "sparsity_diagnostics_every": 10,
+    "stop_after_step": null,
+    "target_tokens_per_parameter": 2.489304879893371,
+    "tokens_per_parameter": 2.489304879893371,
+    "tokens_per_second": 278658.77072000515,
+    "tokens_processed": 1232338944,
+    "total_compile_seconds": 166.9548121576663,
+    "train_compile_seconds": 57.560313685797155,
+    "train_loss": 3.1848177909851074,
+    "train_seconds": 4422.39424517611,
+    "training_data_epochs": 0.15599267073364215,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 9402,
+    "training_token_budget": 1232338944,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 3.3040049374103546,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1338,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1338/result.json

@@ -0,0 +1,256 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.riglog",
+    "fuzzy_sparsity": "fuzzy_sparsity.rigvec",
+    "training_curve": "training.riglog",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "context_preset": "8k",
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 896,
+      "heads": 14,
+      "layers": 14,
+      "mlp_activation": "fuzzy_topk_relu",
+      "mlp_mult": 16,
+      "mlp_top_k": 3584,
+      "normalization": "rms_norm",
+      "parameterization": "completep_fixed_tpp_v1",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "250m",
+      "vocab_size": 50304
+    },
+    "model_id": "fuzzy-topk-autoencoder-gpt-v1-family",
+    "sequence_length": 8192,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 3.3040049374103546,
+      "perplexity": 27.221441070036004,
+      "scored_tokens": 1048576,
+      "seconds": 0.9745558970607817
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "e06781c1c2978705df2ad6d078d005991aaaf8908efa50bfd764dd157326f324",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "context_preset": "8k",
+      "path": "dev.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "final_predictions": 1048576,
+          "probe_predictions": 0,
+          "val_every": 0,
+          "val_probe_batches": 0
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "document_masking": true,
+          "loss_backend": "tiled",
+          "sparse_mlp_backend": "choicewise",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10,
+          "sparsity_diagnostics_every": 10
+        },
+        "model": {
+          "d_model": 896,
+          "heads": 14,
+          "layers": 14,
+          "mlp_activation": "fuzzy_topk_relu",
+          "mlp_mult": 16,
+          "mlp_top_k": 3584,
+          "normalization": "rms_norm",
+          "parameterization": "completep_fixed_tpp_v1",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "250m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.00390625,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.00390625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 940,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_ratio": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 1.1666666666666667,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "ladder_data_multiplier": 1.0,
+          "name": "completep_fixed_tpp_v1",
+          "width_multiplier": 2.3333333333333335
+        },
+        "training": {
+          "batch_size": 16,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 8192,
+          "steps": 9402,
+          "target_tokens_per_parameter": 2.489304879893371,
+          "tokens_per_parameter": 2.489304879893371,
+          "train_tokens": 1232338944
+        }
+      },
+      "schema_version": 6,
+      "sha256": "35ebd3ca279a100bf7c76299955bbcb664c4f1cc959e21f5c332270d82aad7a4"
+    },
+    "context_preset": "8k",
+    "document_masking": true,
+    "loss_backend": "tiled",
+    "sparse_mlp_backend": "choicewise",
+    "vocab_tile_size": 2048,
+    "weight_decay_policy": "weights_and_embeddings_only_v2"
+  },
+  "metrics": {
+    "achieved_tflops": 1122.3941557798357,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.00390625,
+    "diagnostic_compile_seconds": 81.67842828086577,
+    "diagnostic_point_count": 942,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 4963669455340044288,
+    "eval_compile_seconds": 12.553831937955692,
+    "final_validation_seconds": 0.9745558970607817,
+    "flop_accounting": {
+      "by_site": {
+        "_choicewise_fuzzy_topk_mlp": 17678085390336,
+        "dot_general": 5216252461056,
+        "tpu_flash_causal_attention_bwd_dkv": 3367254360064,
+        "tpu_flash_causal_attention_bwd_dq": 3367254360064,
+        "tpu_flash_causal_attention_fwd": 3367254360064
+      },
+      "elementwise_per_sequence": 8887879538,
+      "matmul_per_sequence": 32996100931584,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 4027844352,
+    "mfu_estimate": 0.2550895808590536,
+    "model_tier": "250m",
+    "parameter_count": 495053440,
+    "parameters": 495053440,
+    "schedule_steps": 9402,
+    "sparsity_diagnostic_compile_seconds": 15.16223825304769,
+    "sparsity_diagnostic_point_count": 942,
+    "sparsity_diagnostics_every": 10,
+    "stop_after_step": null,
+    "target_tokens_per_parameter": 2.489304879893371,
+    "tokens_per_parameter": 2.489304879893371,
+    "tokens_per_second": 278658.77072000515,
+    "tokens_processed": 1232338944,
+    "total_compile_seconds": 166.9548121576663,
+    "train_compile_seconds": 57.560313685797155,
+    "train_loss": 3.1848177909851074,
+    "train_seconds": 4422.39424517611,
+    "training_data_epochs": 0.15599267073364215,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 9402,
+    "training_token_budget": 1232338944,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 3.3040049374103546,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1338,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1338/training.riglog

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:70c0ffdd65ec23cc21e1f21f5d919c167b7c30242c2e0c5e7ee4926fc9a64831
+size 150536

fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1338/validation.csv

@@ -0,0 +1,2 @@
+step,tokens_processed,kind,domain,validation_tokens,validation_loss,perplexity,validation_seconds,canonical
+9402,1232338944,fineweb,fineweb,1048576,3.3040049374103546,27.221441070036004,0.9745558970607817,true

fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1339/diagnostics.riglog

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0d5ec373ee087cda4d56c63d826dc29b4a042458ef3f1e72a5e83a8e2b29ee8a
+size 1232408

fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1339/fuzzy_sparsity.rigvec

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:822c4975de87808dd44684e2cc6f52650e7da926b76624a9bce0284bc87ddaed
+size 3025014512

fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1339/fuzzy_sparsity_lossy.rigvec

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0f15f58782ff3ee42fbf1208790f45f2a95dc054ad36ab583aafd4b28a35e2c4
+size 89915560

fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1339/metrics.json

@@ -0,0 +1,256 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.riglog",
+    "fuzzy_sparsity": "fuzzy_sparsity.rigvec",
+    "training_curve": "training.riglog",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "context_preset": "8k",
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 896,
+      "heads": 14,
+      "layers": 14,
+      "mlp_activation": "fuzzy_topk_relu",
+      "mlp_mult": 16,
+      "mlp_top_k": 3584,
+      "normalization": "rms_norm",
+      "parameterization": "completep_fixed_tpp_v1",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "250m",
+      "vocab_size": 50304
+    },
+    "model_id": "fuzzy-topk-autoencoder-gpt-v1-family",
+    "sequence_length": 8192,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 3.306984007358551,
+      "perplexity": 27.302656560322916,
+      "scored_tokens": 1048576,
+      "seconds": 0.9671843170654029
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "e06781c1c2978705df2ad6d078d005991aaaf8908efa50bfd764dd157326f324",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "context_preset": "8k",
+      "path": "dev.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "final_predictions": 1048576,
+          "probe_predictions": 0,
+          "val_every": 0,
+          "val_probe_batches": 0
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "document_masking": true,
+          "loss_backend": "tiled",
+          "sparse_mlp_backend": "choicewise",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10,
+          "sparsity_diagnostics_every": 10
+        },
+        "model": {
+          "d_model": 896,
+          "heads": 14,
+          "layers": 14,
+          "mlp_activation": "fuzzy_topk_relu",
+          "mlp_mult": 16,
+          "mlp_top_k": 3584,
+          "normalization": "rms_norm",
+          "parameterization": "completep_fixed_tpp_v1",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "250m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.00390625,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.00390625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 940,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_ratio": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 1.1666666666666667,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "ladder_data_multiplier": 1.0,
+          "name": "completep_fixed_tpp_v1",
+          "width_multiplier": 2.3333333333333335
+        },
+        "training": {
+          "batch_size": 16,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 8192,
+          "steps": 9402,
+          "target_tokens_per_parameter": 2.489304879893371,
+          "tokens_per_parameter": 2.489304879893371,
+          "train_tokens": 1232338944
+        }
+      },
+      "schema_version": 6,
+      "sha256": "35ebd3ca279a100bf7c76299955bbcb664c4f1cc959e21f5c332270d82aad7a4"
+    },
+    "context_preset": "8k",
+    "document_masking": true,
+    "loss_backend": "tiled",
+    "sparse_mlp_backend": "choicewise",
+    "vocab_tile_size": 2048,
+    "weight_decay_policy": "weights_and_embeddings_only_v2"
+  },
+  "metrics": {
+    "achieved_tflops": 1122.6806043428,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.00390625,
+    "diagnostic_compile_seconds": 83.23276506806724,
+    "diagnostic_point_count": 942,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 4963669455340044288,
+    "eval_compile_seconds": 12.998791442951187,
+    "final_validation_seconds": 0.9671843170654029,
+    "flop_accounting": {
+      "by_site": {
+        "_choicewise_fuzzy_topk_mlp": 17678085390336,
+        "dot_general": 5216252461056,
+        "tpu_flash_causal_attention_bwd_dkv": 3367254360064,
+        "tpu_flash_causal_attention_bwd_dq": 3367254360064,
+        "tpu_flash_causal_attention_fwd": 3367254360064
+      },
+      "elementwise_per_sequence": 8887879538,
+      "matmul_per_sequence": 32996100931584,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 4027844352,
+    "mfu_estimate": 0.25515468280518183,
+    "model_tier": "250m",
+    "parameter_count": 495053440,
+    "parameters": 495053440,
+    "schedule_steps": 9402,
+    "sparsity_diagnostic_compile_seconds": 15.126094905892387,
+    "sparsity_diagnostic_point_count": 942,
+    "sparsity_diagnostics_every": 10,
+    "stop_after_step": null,
+    "target_tokens_per_parameter": 2.489304879893371,
+    "tokens_per_parameter": 2.489304879893371,
+    "tokens_per_second": 278729.88780843536,
+    "tokens_processed": 1232338944,
+    "total_compile_seconds": 181.17454498284496,
+    "train_compile_seconds": 69.81689356593415,
+    "train_loss": 3.2581043243408203,
+    "train_seconds": 4421.265884650871,
+    "training_data_epochs": 0.15599267073364215,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 9402,
+    "training_token_budget": 1232338944,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 3.306984007358551,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1339,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1339/result.json

@@ -0,0 +1,256 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.riglog",
+    "fuzzy_sparsity": "fuzzy_sparsity.rigvec",
+    "training_curve": "training.riglog",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "context_preset": "8k",
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 896,
+      "heads": 14,
+      "layers": 14,
+      "mlp_activation": "fuzzy_topk_relu",
+      "mlp_mult": 16,
+      "mlp_top_k": 3584,
+      "normalization": "rms_norm",
+      "parameterization": "completep_fixed_tpp_v1",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "250m",
+      "vocab_size": 50304
+    },
+    "model_id": "fuzzy-topk-autoencoder-gpt-v1-family",
+    "sequence_length": 8192,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 3.306984007358551,
+      "perplexity": 27.302656560322916,
+      "scored_tokens": 1048576,
+      "seconds": 0.9671843170654029
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "e06781c1c2978705df2ad6d078d005991aaaf8908efa50bfd764dd157326f324",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "context_preset": "8k",
+      "path": "dev.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "final_predictions": 1048576,
+          "probe_predictions": 0,
+          "val_every": 0,
+          "val_probe_batches": 0
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "document_masking": true,
+          "loss_backend": "tiled",
+          "sparse_mlp_backend": "choicewise",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10,
+          "sparsity_diagnostics_every": 10
+        },
+        "model": {
+          "d_model": 896,
+          "heads": 14,
+          "layers": 14,
+          "mlp_activation": "fuzzy_topk_relu",
+          "mlp_mult": 16,
+          "mlp_top_k": 3584,
+          "normalization": "rms_norm",
+          "parameterization": "completep_fixed_tpp_v1",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "250m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.00390625,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.00390625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 940,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_ratio": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 1.1666666666666667,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "ladder_data_multiplier": 1.0,
+          "name": "completep_fixed_tpp_v1",
+          "width_multiplier": 2.3333333333333335
+        },
+        "training": {
+          "batch_size": 16,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 8192,
+          "steps": 9402,
+          "target_tokens_per_parameter": 2.489304879893371,
+          "tokens_per_parameter": 2.489304879893371,
+          "train_tokens": 1232338944
+        }
+      },
+      "schema_version": 6,
+      "sha256": "35ebd3ca279a100bf7c76299955bbcb664c4f1cc959e21f5c332270d82aad7a4"
+    },
+    "context_preset": "8k",
+    "document_masking": true,
+    "loss_backend": "tiled",
+    "sparse_mlp_backend": "choicewise",
+    "vocab_tile_size": 2048,
+    "weight_decay_policy": "weights_and_embeddings_only_v2"
+  },
+  "metrics": {
+    "achieved_tflops": 1122.6806043428,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.00390625,
+    "diagnostic_compile_seconds": 83.23276506806724,
+    "diagnostic_point_count": 942,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 4963669455340044288,
+    "eval_compile_seconds": 12.998791442951187,
+    "final_validation_seconds": 0.9671843170654029,
+    "flop_accounting": {
+      "by_site": {
+        "_choicewise_fuzzy_topk_mlp": 17678085390336,
+        "dot_general": 5216252461056,
+        "tpu_flash_causal_attention_bwd_dkv": 3367254360064,
+        "tpu_flash_causal_attention_bwd_dq": 3367254360064,
+        "tpu_flash_causal_attention_fwd": 3367254360064
+      },
+      "elementwise_per_sequence": 8887879538,
+      "matmul_per_sequence": 32996100931584,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 4027844352,
+    "mfu_estimate": 0.25515468280518183,
+    "model_tier": "250m",
+    "parameter_count": 495053440,
+    "parameters": 495053440,
+    "schedule_steps": 9402,
+    "sparsity_diagnostic_compile_seconds": 15.126094905892387,
+    "sparsity_diagnostic_point_count": 942,
+    "sparsity_diagnostics_every": 10,
+    "stop_after_step": null,
+    "target_tokens_per_parameter": 2.489304879893371,
+    "tokens_per_parameter": 2.489304879893371,
+    "tokens_per_second": 278729.88780843536,
+    "tokens_processed": 1232338944,
+    "total_compile_seconds": 181.17454498284496,
+    "train_compile_seconds": 69.81689356593415,
+    "train_loss": 3.2581043243408203,
+    "train_seconds": 4421.265884650871,
+    "training_data_epochs": 0.15599267073364215,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 9402,
+    "training_token_budget": 1232338944,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 3.306984007358551,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1339,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1339/training.riglog

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b6062f911a5a81665bc9fdd1259b13bc63f66d91a9523161437631f1a195ffc2
+size 150536

fuzzy-topk-sparsity-diagnostics-ladder/250m-fuzzy-h16d-k4d-l14-2tpp-bs16-lr2e-8-s1339/validation.csv

@@ -0,0 +1,2 @@
+step,tokens_processed,kind,domain,validation_tokens,validation_loss,perplexity,validation_seconds,canonical
+9402,1232338944,fineweb,fineweb,1048576,3.306984007358551,27.302656560322916,0.9671843170654029,true

fuzzy-topk-sparsity-diagnostics-ladder/500m-fuzzy-h16d-k4d-l16-2tpp-bs16-lr2e-8-s1337/diagnostics.riglog

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ccd0cb279b9405d22db1b4a5a9c328a93070e5670ff52711a9a693016fd2e5ea
+size 2836024

fuzzy-topk-sparsity-diagnostics-ladder/500m-fuzzy-h16d-k4d-l16-2tpp-bs16-lr2e-8-s1337/fuzzy_sparsity.rigvec

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e982261923a00d3cd5962a7adcff671c2600400cff132fdeb3d6660383046ea
+size 10265566928

fuzzy-topk-sparsity-diagnostics-ladder/500m-fuzzy-h16d-k4d-l16-2tpp-bs16-lr2e-8-s1337/fuzzy_sparsity_lossy.rigvec

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a0c72cb5b650b5426e55a6b38c53a288611413020d62e00aefb8f5256fad31a7
+size 152043692

fuzzy-topk-sparsity-diagnostics-ladder/500m-fuzzy-h16d-k4d-l16-2tpp-bs16-lr2e-8-s1337/metrics.json

@@ -0,0 +1,256 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.riglog",
+    "fuzzy_sparsity": "fuzzy_sparsity.rigvec",
+    "training_curve": "training.riglog",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "context_preset": "8k",
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 1280,
+      "heads": 20,
+      "layers": 16,
+      "mlp_activation": "fuzzy_topk_relu",
+      "mlp_mult": 16,
+      "mlp_top_k": 5120,
+      "normalization": "rms_norm",
+      "parameterization": "completep_fixed_tpp_v1",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "500m",
+      "vocab_size": 50304
+    },
+    "model_id": "fuzzy-topk-autoencoder-gpt-v1-family",
+    "sequence_length": 8192,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 3.093101680278778,
+      "perplexity": 22.04534966885452,
+      "scored_tokens": 1048576,
+      "seconds": 1.7293408529367298
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "84fe3fd56290695de85e933b91350cf7559555228c45b8400b40d51e9e691c20",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "context_preset": "8k",
+      "path": "dev.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "final_predictions": 1048576,
+          "probe_predictions": 0,
+          "val_every": 0,
+          "val_probe_batches": 0
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "document_masking": true,
+          "loss_backend": "tiled",
+          "sparse_mlp_backend": "choicewise",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10,
+          "sparsity_diagnostics_every": 10
+        },
+        "model": {
+          "d_model": 1280,
+          "heads": 20,
+          "layers": 16,
+          "mlp_activation": "fuzzy_topk_relu",
+          "mlp_mult": 16,
+          "mlp_top_k": 5120,
+          "normalization": "rms_norm",
+          "parameterization": "completep_fixed_tpp_v1",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "500m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.00390625,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.00390625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 1956,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_ratio": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 1.3333333333333333,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "ladder_data_multiplier": 1.0,
+          "name": "completep_fixed_tpp_v1",
+          "width_multiplier": 3.3333333333333335
+        },
+        "training": {
+          "batch_size": 16,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 8192,
+          "steps": 19561,
+          "target_tokens_per_parameter": 2.389537337594556,
+          "tokens_per_parameter": 2.389537337594556,
+          "train_tokens": 2563899392
+        }
+      },
+      "schema_version": 6,
+      "sha256": "35ebd3ca279a100bf7c76299955bbcb664c4f1cc959e21f5c332270d82aad7a4"
+    },
+    "context_preset": "8k",
+    "document_masking": true,
+    "loss_backend": "tiled",
+    "sparse_mlp_backend": "choicewise",
+    "vocab_tile_size": 2048,
+    "weight_decay_policy": "weights_and_embeddings_only_v2"
+  },
+  "metrics": {
+    "achieved_tflops": 1164.0399494191674,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.00390625,
+    "diagnostic_compile_seconds": 92.70696510304697,
+    "diagnostic_point_count": 1958,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 21023633708112936960,
+    "eval_compile_seconds": 16.746170554775745,
+    "final_validation_seconds": 1.7293408529367298,
+    "flop_accounting": {
+      "by_site": {
+        "_choicewise_fuzzy_topk_mlp": 41231686041600,
+        "dot_general": 9448949022720,
+        "tpu_flash_causal_attention_bwd_dkv": 5497558138880,
+        "tpu_flash_causal_attention_bwd_dq": 5497558138880,
+        "tpu_flash_causal_attention_fwd": 5497558138880
+      },
+      "elementwise_per_sequence": 12057100530,
+      "matmul_per_sequence": 67173309480960,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 8199866880,
+    "mfu_estimate": 0.2645545339589017,
+    "model_tier": "500m",
+    "parameter_count": 1072968960,
+    "parameters": 1072968960,
+    "schedule_steps": 19561,
+    "sparsity_diagnostic_compile_seconds": 19.950090633006766,
+    "sparsity_diagnostic_point_count": 1958,
+    "sparsity_diagnostics_every": 10,
+    "stop_after_step": null,
+    "target_tokens_per_parameter": 2.389537337594556,
+    "tokens_per_parameter": 2.389537337594556,
+    "tokens_per_second": 141958.39596595592,
+    "tokens_processed": 2563899392,
+    "total_compile_seconds": 203.57062846887857,
+    "train_compile_seconds": 74.16740217804909,
+    "train_loss": 3.0271120071411133,
+    "train_seconds": 18060.921120966086,
+    "training_data_epochs": 0.32454505767079067,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 19561,
+    "training_token_budget": 2563899392,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 3.093101680278778,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1337,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

fuzzy-topk-sparsity-diagnostics-ladder/500m-fuzzy-h16d-k4d-l16-2tpp-bs16-lr2e-8-s1337/result.json

@@ -0,0 +1,256 @@
+{
+  "artifacts": {
+    "diagnostics": "diagnostics.riglog",
+    "fuzzy_sparsity": "fuzzy_sparsity.rigvec",
+    "training_curve": "training.riglog",
+    "validation_curve": "validation.csv"
+  },
+  "checkpoint": null,
+  "contract": {
+    "context_preset": "8k",
+    "dataset_id": "fineweb-8b-gpt2",
+    "model": {
+      "d_model": 1280,
+      "heads": 20,
+      "layers": 16,
+      "mlp_activation": "fuzzy_topk_relu",
+      "mlp_mult": 16,
+      "mlp_top_k": 5120,
+      "normalization": "rms_norm",
+      "parameterization": "completep_fixed_tpp_v1",
+      "position_encoding": "rope_base_10000",
+      "semantic_vocab_size": 50304,
+      "tied_embeddings": false,
+      "tier": "500m",
+      "vocab_size": 50304
+    },
+    "model_id": "fuzzy-topk-autoencoder-gpt-v1-family",
+    "sequence_length": 8192,
+    "tokenizer_id": "gpt2"
+  },
+  "evaluations": {
+    "fineweb": {
+      "canonical": true,
+      "loss": 3.093101680278778,
+      "perplexity": 22.04534966885452,
+      "scored_tokens": 1048576,
+      "seconds": 1.7293408529367298
+    }
+  },
+  "implementation": {
+    "attention_backend": "tpu_flash",
+    "attention_tuning": {
+      "key_digest": "84fe3fd56290695de85e933b91350cf7559555228c45b8400b40d51e9e691c20",
+      "resolution_source": "heuristic",
+      "tiles": {
+        "block_kv": 512,
+        "block_kv_compute": 256,
+        "block_kv_dkv": 512,
+        "block_kv_dkv_compute": 256,
+        "block_kv_dq": 512,
+        "block_kv_dq_compute": 256,
+        "block_q": 512,
+        "block_q_dkv": 512,
+        "block_q_dkv_compute": 256,
+        "block_q_dq": 256
+      },
+      "tune_seconds": 0.0
+    },
+    "configuration": {
+      "context_preset": "8k",
+      "path": "dev.yaml",
+      "profile": "dev",
+      "resolved": {
+        "evaluation": {
+          "eval_batches": 8,
+          "final_predictions": 1048576,
+          "probe_predictions": 0,
+          "val_every": 0,
+          "val_probe_batches": 0
+        },
+        "kernels": {
+          "attention_backend": "tpu_flash",
+          "document_masking": true,
+          "loss_backend": "tiled",
+          "sparse_mlp_backend": "choicewise",
+          "vocab_tile_size": 2048
+        },
+        "logging": {
+          "diagnostics_every": 10,
+          "log_every": 10,
+          "sparsity_diagnostics_every": 10
+        },
+        "model": {
+          "d_model": 1280,
+          "heads": 20,
+          "layers": 16,
+          "mlp_activation": "fuzzy_topk_relu",
+          "mlp_mult": 16,
+          "mlp_top_k": 5120,
+          "normalization": "rms_norm",
+          "parameterization": "completep_fixed_tpp_v1",
+          "position_encoding": "rope_base_10000",
+          "semantic_vocab_size": 50304,
+          "tied_embeddings": false,
+          "tier": "500m",
+          "vocab_size": 50304
+        },
+        "optimizer": {
+          "adam_epsilon": 1e-08,
+          "beta1": 0.9,
+          "beta2": 0.95,
+          "effective": {
+            "adam_epsilon_horizon_multiplier": 1.0,
+            "beta1": 0.9,
+            "beta2": 0.95,
+            "global_peak_learning_rate": 0.00390625,
+            "weight_decay_horizon_multiplier": 1.0
+          },
+          "grad_clip": 0.0,
+          "learning_rate": 0.00390625,
+          "min_lr_ratio": 0.1,
+          "warmup_steps": 1956,
+          "weight_decay": 0.1
+        },
+        "parameterization": {
+          "attention_scale": "inverse_head_dim",
+          "base_depth": 12,
+          "base_width": 384,
+          "batch_ratio": 1.0,
+          "depth_alpha": 1.0,
+          "depth_multiplier": 1.3333333333333333,
+          "embeddings": "untied",
+          "init_std": 0.02,
+          "ladder_data_multiplier": 1.0,
+          "name": "completep_fixed_tpp_v1",
+          "width_multiplier": 3.3333333333333335
+        },
+        "training": {
+          "batch_size": 16,
+          "dtype": "bfloat16",
+          "sampling": "shuffled_epochs",
+          "seq_len": 8192,
+          "steps": 19561,
+          "target_tokens_per_parameter": 2.389537337594556,
+          "tokens_per_parameter": 2.389537337594556,
+          "train_tokens": 2563899392
+        }
+      },
+      "schema_version": 6,
+      "sha256": "35ebd3ca279a100bf7c76299955bbcb664c4f1cc959e21f5c332270d82aad7a4"
+    },
+    "context_preset": "8k",
+    "document_masking": true,
+    "loss_backend": "tiled",
+    "sparse_mlp_backend": "choicewise",
+    "vocab_tile_size": 2048,
+    "weight_decay_policy": "weights_and_embeddings_only_v2"
+  },
+  "metrics": {
+    "achieved_tflops": 1164.0399494191674,
+    "attention_tune_seconds": 0.0,
+    "base_learning_rate": 0.00390625,
+    "diagnostic_compile_seconds": 92.70696510304697,
+    "diagnostic_point_count": 1958,
+    "diagnostics_every": 10,
+    "estimated_total_flops": 21023633708112936960,
+    "eval_compile_seconds": 16.746170554775745,
+    "final_validation_seconds": 1.7293408529367298,
+    "flop_accounting": {
+      "by_site": {
+        "_choicewise_fuzzy_topk_mlp": 41231686041600,
+        "dot_general": 9448949022720,
+        "tpu_flash_causal_attention_bwd_dkv": 5497558138880,
+        "tpu_flash_causal_attention_bwd_dq": 5497558138880,
+        "tpu_flash_causal_attention_fwd": 5497558138880
+      },
+      "elementwise_per_sequence": 12057100530,
+      "matmul_per_sequence": 67173309480960,
+      "method": "traced-jaxpr",
+      "warnings": []
+    },
+    "flops_per_token": 8199866880,
+    "mfu_estimate": 0.2645545339589017,
+    "model_tier": "500m",
+    "parameter_count": 1072968960,
+    "parameters": 1072968960,
+    "schedule_steps": 19561,
+    "sparsity_diagnostic_compile_seconds": 19.950090633006766,
+    "sparsity_diagnostic_point_count": 1958,
+    "sparsity_diagnostics_every": 10,
+    "stop_after_step": null,
+    "target_tokens_per_parameter": 2.389537337594556,
+    "tokens_per_parameter": 2.389537337594556,
+    "tokens_per_second": 141958.39596595592,
+    "tokens_processed": 2563899392,
+    "total_compile_seconds": 203.57062846887857,
+    "train_compile_seconds": 74.16740217804909,
+    "train_loss": 3.0271120071411133,
+    "train_seconds": 18060.921120966086,
+    "training_data_epochs": 0.32454505767079067,
+    "training_data_sharding": "rank_disjoint_shuffled_windows",
+    "training_sampling": "shuffled_epochs",
+    "training_steps": 19561,
+    "training_token_budget": 2563899392,
+    "training_usable_tokens_per_epoch": 7899979776,
+    "validation_loss": 3.093101680278778,
+    "validation_probe_count": 0,
+    "validation_probe_seconds": 0.0,
+    "validation_tokens": 1048576
+  },
+  "profile": "dev",
+  "schema_version": 1,
+  "seed": 1337,
+  "status": "ok",
+  "system": {
+    "controller_process_index": 1,
+    "device_count": 16,
+    "device_ids": [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15
+    ],
+    "device_kinds": [
+      "TPU v4"
+    ],
+    "jax_version": "0.11.0",
+    "jaxlib_version": "0.11.0",
+    "libtpu_version": "0.0.44.1",
+    "local_device_count": 4,
+    "platform": "tpu",
+    "process_count": 4,
+    "process_indices": [
+      0,
+      0,
+      0,
+      0,
+      1,
+      1,
+      1,
+      1,
+      2,
+      2,
+      2,
+      2,
+      3,
+      3,
+      3,
+      3
+    ],
+    "python_version": "3.12.13"
+  },
+  "track": "open"
+}

fuzzy-topk-sparsity-diagnostics-ladder/500m-fuzzy-h16d-k4d-l16-2tpp-bs16-lr2e-8-s1337/training.riglog

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4e6ede524e9c5b6ee96944b98f792696f086025eadf0ca64da88cebd1a51cb6a
+size 313080