Add CodeLeWM checkpoint artifact codelewm-scaled-20260520-9699b53

checkpoints/codelewm-scaled-20260520-9699b53/checkpoints/checkpoint.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09bf8d3880ec272a858dd9b19f2b29622a66a5ebbef6dbd1f8e4ebeb8b6392b8
+size 351495453

checkpoints/codelewm-scaled-20260520-9699b53/checkpoints/checkpoint.pt.manifest.json

@@ -0,0 +1,14 @@
+{
+  "checkpoint_path": "checkpoint.pt",
+  "checkpoint_sha256": "09bf8d3880ec272a858dd9b19f2b29622a66a5ebbef6dbd1f8e4ebeb8b6392b8",
+  "metadata": {
+    "action_view": "text",
+    "config_hash": "72822cb45ab87284e629399b567b0f6daa882d79e2f3454654b0835880a73408",
+    "latent_dim": 256,
+    "model_class": "TorchCodeTransitionModel",
+    "record_schema_version": "codelewm.transition.v1",
+    "schema_version": "codelewm.checkpoint.v1"
+  },
+  "migration_hook": null,
+  "schema_version": "codelewm.checkpoint.v1"
+}

checkpoints/codelewm-scaled-20260520-9699b53/config.json

@@ -0,0 +1,51 @@
+{
+  "data": {
+    "manifest": "/CodeLeWM/.artifacts/hf/codelewm-scaled-20260520-9699b53/pack/manifest.json",
+    "train": "/CodeLeWM/.artifacts/hf/codelewm-scaled-20260520-9699b53/pack/hdf5/train.hdf5",
+    "val": "/CodeLeWM/.artifacts/hf/codelewm-scaled-20260520-9699b53/pack/hdf5/val.hdf5"
+  },
+  "loader": {
+    "batch_size": 64,
+    "num_workers": 4,
+    "persistent_workers": true,
+    "pin_memory": true,
+    "shuffle": true
+  },
+  "loss": {
+    "enable_retrieval_loss": false,
+    "retrieval_temperature": 0.1,
+    "retrieval_weight": 0.0,
+    "sigreg_knots": 17,
+    "sigreg_num_proj": 1024,
+    "sigreg_weight": 0.09
+  },
+  "name": "codelewm_scaled_gpu_a10g",
+  "optimizer": {
+    "lr": 0.0001,
+    "type": "AdamW",
+    "weight_decay": 0.001
+  },
+  "output": {
+    "checkpoint_dir": "/CodeLeWM/.artifacts/hf/codelewm-scaled-20260520-9699b53/train/checkpoints",
+    "manifest_path": "/CodeLeWM/.artifacts/hf/codelewm-scaled-20260520-9699b53/train/training_manifest.json",
+    "metrics_path": "/CodeLeWM/.artifacts/hf/codelewm-scaled-20260520-9699b53/train/metrics.jsonl",
+    "run_dir": "/CodeLeWM/.artifacts/hf/codelewm-scaled-20260520-9699b53/train"
+  },
+  "schema_version": "codelewm.train_config.v1",
+  "seed": 240119,
+  "trainer": {
+    "accelerator": "gpu",
+    "devices": 1,
+    "gradient_clip_val": 1.0,
+    "max_steps": 60000,
+    "precision": "bf16-mixed"
+  },
+  "wm": {
+    "action_sequence_length": 256,
+    "action_view": "text",
+    "embed_dim": 256,
+    "history_size": 1,
+    "num_preds": 1,
+    "state_sequence_length": 1024
+  }
+}

checkpoints/codelewm-scaled-20260520-9699b53/manifest.json

@@ -0,0 +1,96 @@
+{
+  "artifact_id": "training_run-d9074199c0d58911",
+  "artifact_kind": "training_run",
+  "command": [
+    "codelewm",
+    "train",
+    "--config",
+    "/CodeLeWM/.artifacts/hf/codelewm-scaled-20260520-9699b53/configs/train.json",
+    "--executor",
+    "torch",
+    "--device",
+    "auto",
+    "--overwrite",
+    "--json",
+    "--log-jsonl",
+    "/CodeLeWM/.artifacts/hf/codelewm-scaled-20260520-9699b53/logs/train.jsonl"
+  ],
+  "config_sha256": "119507eb9909ba9c8dac134595266f5a941ea5c1d096ceeb61679da14d336dad",
+  "created_at": "2026-05-20T06:22:58Z",
+  "files": [
+    {
+      "bytes": 1556,
+      "path": "config.json",
+      "sha256": "889396e0262d4fadefa558ccab00934fac29ad58f307643a81612cca73d98488"
+    },
+    {
+      "bytes": 1040,
+      "path": "metrics.jsonl",
+      "sha256": "68a6b2d80774a62f31b87e9d6ee3c4b36978eddeb1805a849d47458b8871ffd9"
+    },
+    {
+      "bytes": 1150,
+      "path": "reports/metrics_report.json",
+      "sha256": "ea1a938c6a3b90f74c2eea9ab09723c004f24a4f7043fb3bdc7edfc95516f34f"
+    },
+    {
+      "bytes": 351495453,
+      "path": "checkpoints/checkpoint.pt",
+      "sha256": "09bf8d3880ec272a858dd9b19f2b29622a66a5ebbef6dbd1f8e4ebeb8b6392b8"
+    },
+    {
+      "bytes": 511,
+      "path": "checkpoints/checkpoint.pt.manifest.json",
+      "sha256": "b18a57a783d2c4693f7c99489ad4752ce6fa1ba8b64a61ccef688dfb599189fd"
+    },
+    {
+      "bytes": 1893,
+      "path": "reports/torch_training_report.json",
+      "sha256": "12d3ace9319d0c1b379d32986365fc2b20baad839c2f20ba2a86d249b4672004"
+    }
+  ],
+  "metadata": {
+    "dataset_manifest_path": ".artifacts/hf/codelewm-scaled-20260520-9699b53/pack/manifest.json",
+    "executor": {
+      "checkpoint_schema_version": "codelewm.checkpoint.v1",
+      "device": "cuda",
+      "executor": "torch",
+      "precision": "bf16-mixed",
+      "torch": "2.12.0+cu130",
+      "train_rows": 18019,
+      "val_rows": 1291
+    },
+    "final_metrics": {
+      "collapse/effective_rank": 5.437364589252875,
+      "collapse/effective_rank_ratio": 0.02123970542676904,
+      "collapse/embedding_count": 192.0,
+      "collapse/embedding_norm_mean": 14.346564877617572,
+      "collapse/latent_dim": 256.0,
+      "collapse/nearest_neighbor_entropy": 4.85942433071649,
+      "collapse/pairwise_cosine_mean": 0.008317808396257337,
+      "collapse/per_dim_variance_max": 3.289028716596054,
+      "collapse/per_dim_variance_median": 0.6952048902794287,
+      "collapse/per_dim_variance_min": 0.0540634111270202,
+      "loss/prediction_mse": 0.005166558548808098,
+      "loss/sigreg": 1.203125,
+      "loss/sigreg_weighted": 0.1083984375,
+      "loss/total": 0.11356499791145325,
+      "train/examples": 3833852.0,
+      "train/examples_per_second": 1103.204300183749,
+      "train/gradient_norm": 1.8980227708816528,
+      "val/loss/prediction_mse": 0.17075516851175399,
+      "val/loss/sigreg": 2.048149585723877,
+      "val/loss/sigreg_weighted": 0.1843334688317208,
+      "val/loss/total": 0.3550886376982644
+    },
+    "run_id": "codelewm_scaled_gpu_a10g",
+    "schema_version": "codelewm.training_run.v1",
+    "seed": 240119,
+    "step_count": 60000
+  },
+  "parent_artifacts": [
+    "dataset-ef8ad3f4f48dea9e"
+  ],
+  "schema_version": "codelewm.artifact_manifest.v1",
+  "source_git_sha": "9699b5309e43a3278f272663ef60cda23040d92a"
+}

checkpoints/codelewm-scaled-20260520-9699b53/metrics.jsonl

@@ -0,0 +1 @@
+{"metrics": {"collapse/effective_rank": 5.437364589252875, "collapse/effective_rank_ratio": 0.02123970542676904, "collapse/embedding_count": 192.0, "collapse/embedding_norm_mean": 14.346564877617572, "collapse/latent_dim": 256.0, "collapse/nearest_neighbor_entropy": 4.85942433071649, "collapse/pairwise_cosine_mean": 0.008317808396257337, "collapse/per_dim_variance_max": 3.289028716596054, "collapse/per_dim_variance_median": 0.6952048902794287, "collapse/per_dim_variance_min": 0.0540634111270202, "loss/prediction_mse": 0.005166558548808098, "loss/sigreg": 1.203125, "loss/sigreg_weighted": 0.1083984375, "loss/total": 0.11356499791145325, "train/examples": 3833852.0, "train/examples_per_second": 1103.204300183749, "train/gradient_norm": 1.8980227708816528, "val/loss/prediction_mse": 0.17075516851175399, "val/loss/sigreg": 2.048149585723877, "val/loss/sigreg_weighted": 0.1843334688317208, "val/loss/total": 0.3550886376982644}, "run_id": "codelewm_scaled_gpu_a10g", "schema_version": "codelewm.training_metrics.v1", "step": 60000}

checkpoints/codelewm-scaled-20260520-9699b53/reports/metrics_report.json

@@ -0,0 +1,28 @@
+{
+  "final_metrics": {
+    "collapse/effective_rank": 5.437364589252875,
+    "collapse/effective_rank_ratio": 0.02123970542676904,
+    "collapse/embedding_count": 192.0,
+    "collapse/embedding_norm_mean": 14.346564877617572,
+    "collapse/latent_dim": 256.0,
+    "collapse/nearest_neighbor_entropy": 4.85942433071649,
+    "collapse/pairwise_cosine_mean": 0.008317808396257337,
+    "collapse/per_dim_variance_max": 3.289028716596054,
+    "collapse/per_dim_variance_median": 0.6952048902794287,
+    "collapse/per_dim_variance_min": 0.0540634111270202,
+    "loss/prediction_mse": 0.005166558548808098,
+    "loss/sigreg": 1.203125,
+    "loss/sigreg_weighted": 0.1083984375,
+    "loss/total": 0.11356499791145325,
+    "train/examples": 3833852.0,
+    "train/examples_per_second": 1103.204300183749,
+    "train/gradient_norm": 1.8980227708816528,
+    "val/loss/prediction_mse": 0.17075516851175399,
+    "val/loss/sigreg": 2.048149585723877,
+    "val/loss/sigreg_weighted": 0.1843334688317208,
+    "val/loss/total": 0.3550886376982644
+  },
+  "run_id": "codelewm_scaled_gpu_a10g",
+  "schema_version": "codelewm.training_metrics.v1",
+  "step_count": 60000
+}

checkpoints/codelewm-scaled-20260520-9699b53/reports/torch_training_report.json

@@ -0,0 +1,52 @@
+{
+  "collapse_report": {
+    "effective_rank": 5.437364589252875,
+    "effective_rank_ratio": 0.02123970542676904,
+    "embedding_count": 192,
+    "embedding_norm_mean": 14.346564877617572,
+    "latent_dim": 256,
+    "nearest_neighbor_entropy": 4.85942433071649,
+    "pairwise_cosine_mean": 0.008317808396257337,
+    "per_dim_variance_max": 3.289028716596054,
+    "per_dim_variance_median": 0.6952048902794287,
+    "per_dim_variance_min": 0.0540634111270202,
+    "schema_version": "codelewm.eval.collapse_report.v1"
+  },
+  "dataset": {
+    "action_view": "text",
+    "train_rows": 18019,
+    "val_rows": 1291
+  },
+  "metrics": {
+    "collapse/effective_rank": 5.437364589252875,
+    "collapse/effective_rank_ratio": 0.02123970542676904,
+    "collapse/embedding_count": 192.0,
+    "collapse/embedding_norm_mean": 14.346564877617572,
+    "collapse/latent_dim": 256.0,
+    "collapse/nearest_neighbor_entropy": 4.85942433071649,
+    "collapse/pairwise_cosine_mean": 0.008317808396257337,
+    "collapse/per_dim_variance_max": 3.289028716596054,
+    "collapse/per_dim_variance_median": 0.6952048902794287,
+    "collapse/per_dim_variance_min": 0.0540634111270202,
+    "loss/prediction_mse": 0.005166558548808098,
+    "loss/sigreg": 1.203125,
+    "loss/sigreg_weighted": 0.1083984375,
+    "loss/total": 0.11356499791145325,
+    "train/examples": 3833852.0,
+    "train/examples_per_second": 1103.204300183749,
+    "train/gradient_norm": 1.8980227708816528,
+    "val/loss/prediction_mse": 0.17075516851175399,
+    "val/loss/sigreg": 2.048149585723877,
+    "val/loss/sigreg_weighted": 0.1843334688317208,
+    "val/loss/total": 0.3550886376982644
+  },
+  "run_id": "codelewm_scaled_gpu_a10g",
+  "runtime": {
+    "device": "cuda",
+    "dtype": "torch.bfloat16",
+    "precision": "bf16-mixed",
+    "torch": "2.12.0+cu130"
+  },
+  "schema_version": "codelewm.torch_training_report.v1",
+  "step_count": 60000
+}

checkpoints/codelewm-scaled-20260520-9699b53/training_manifest.json

@@ -0,0 +1,69 @@
+{
+  "artifact_manifest_id": "training_run-d9074199c0d58911",
+  "artifact_manifest_path": "manifest.json",
+  "checkpoint_files": [
+    {
+      "bytes": 351495453,
+      "path": "checkpoints/checkpoint.pt",
+      "sha256": "09bf8d3880ec272a858dd9b19f2b29622a66a5ebbef6dbd1f8e4ebeb8b6392b8"
+    },
+    {
+      "bytes": 511,
+      "path": "checkpoints/checkpoint.pt.manifest.json",
+      "sha256": "b18a57a783d2c4693f7c99489ad4752ce6fa1ba8b64a61ccef688dfb599189fd"
+    }
+  ],
+  "config_path": "config.json",
+  "config_sha256": "119507eb9909ba9c8dac134595266f5a941ea5c1d096ceeb61679da14d336dad",
+  "dataset_manifest_path": ".artifacts/hf/codelewm-scaled-20260520-9699b53/pack/manifest.json",
+  "final_metrics": {
+    "collapse/effective_rank": 5.437364589252875,
+    "collapse/effective_rank_ratio": 0.02123970542676904,
+    "collapse/embedding_count": 192.0,
+    "collapse/embedding_norm_mean": 14.346564877617572,
+    "collapse/latent_dim": 256.0,
+    "collapse/nearest_neighbor_entropy": 4.85942433071649,
+    "collapse/pairwise_cosine_mean": 0.008317808396257337,
+    "collapse/per_dim_variance_max": 3.289028716596054,
+    "collapse/per_dim_variance_median": 0.6952048902794287,
+    "collapse/per_dim_variance_min": 0.0540634111270202,
+    "loss/prediction_mse": 0.005166558548808098,
+    "loss/sigreg": 1.203125,
+    "loss/sigreg_weighted": 0.1083984375,
+    "loss/total": 0.11356499791145325,
+    "train/examples": 3833852.0,
+    "train/examples_per_second": 1103.204300183749,
+    "train/gradient_norm": 1.8980227708816528,
+    "val/loss/prediction_mse": 0.17075516851175399,
+    "val/loss/sigreg": 2.048149585723877,
+    "val/loss/sigreg_weighted": 0.1843334688317208,
+    "val/loss/total": 0.3550886376982644
+  },
+  "metadata": {
+    "executor": {
+      "checkpoint_schema_version": "codelewm.checkpoint.v1",
+      "device": "cuda",
+      "executor": "torch",
+      "precision": "bf16-mixed",
+      "torch": "2.12.0+cu130",
+      "train_rows": 18019,
+      "val_rows": 1291
+    }
+  },
+  "metrics_path": "metrics.jsonl",
+  "metrics_report_path": "reports/metrics_report.json",
+  "parent_artifacts": [
+    "dataset-ef8ad3f4f48dea9e"
+  ],
+  "report_files": [
+    {
+      "bytes": 1893,
+      "path": "reports/torch_training_report.json",
+      "sha256": "12d3ace9319d0c1b379d32986365fc2b20baad839c2f20ba2a86d249b4672004"
+    }
+  ],
+  "run_id": "codelewm_scaled_gpu_a10g",
+  "schema_version": "codelewm.training_run.v1",
+  "seed": 240119,
+  "step_count": 60000
+}