Add CodeLeWM checkpoint artifact codelewm-v0-2-action-swap-rerun-20260520-7c7cb0b
Browse files- checkpoints/codelewm-v0-2-action-swap-rerun-20260520-7c7cb0b/checkpoints/checkpoint.pt +3 -0
- checkpoints/codelewm-v0-2-action-swap-rerun-20260520-7c7cb0b/checkpoints/checkpoint.pt.manifest.json +14 -0
- checkpoints/codelewm-v0-2-action-swap-rerun-20260520-7c7cb0b/config.json +60 -0
- checkpoints/codelewm-v0-2-action-swap-rerun-20260520-7c7cb0b/manifest.json +130 -0
- checkpoints/codelewm-v0-2-action-swap-rerun-20260520-7c7cb0b/metrics.jsonl +1 -0
- checkpoints/codelewm-v0-2-action-swap-rerun-20260520-7c7cb0b/reports/metrics_report.json +46 -0
- checkpoints/codelewm-v0-2-action-swap-rerun-20260520-7c7cb0b/reports/torch_training_report.json +86 -0
- checkpoints/codelewm-v0-2-action-swap-rerun-20260520-7c7cb0b/training_manifest.json +103 -0
checkpoints/codelewm-v0-2-action-swap-rerun-20260520-7c7cb0b/checkpoints/checkpoint.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f2c5ba50ee0ec5e32ff5c3ceed848020e989ebdb1c98a917f17589ee523c6d7e
|
| 3 |
+
size 254388041
|
checkpoints/codelewm-v0-2-action-swap-rerun-20260520-7c7cb0b/checkpoints/checkpoint.pt.manifest.json
ADDED
|
@@ -0,0 +1,14 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"checkpoint_path": "checkpoint.pt",
|
| 3 |
+
"checkpoint_sha256": "f2c5ba50ee0ec5e32ff5c3ceed848020e989ebdb1c98a917f17589ee523c6d7e",
|
| 4 |
+
"metadata": {
|
| 5 |
+
"action_view": "text",
|
| 6 |
+
"config_hash": "5e42808433b8fc0d7eeff7ccd74f22453e2e3c95d2b6ea9e6057b1668043a793",
|
| 7 |
+
"latent_dim": 256,
|
| 8 |
+
"model_class": "TorchCodeTransitionModel",
|
| 9 |
+
"record_schema_version": "codelewm.transition.v1",
|
| 10 |
+
"schema_version": "codelewm.checkpoint.v1"
|
| 11 |
+
},
|
| 12 |
+
"migration_hook": null,
|
| 13 |
+
"schema_version": "codelewm.checkpoint.v1"
|
| 14 |
+
}
|
checkpoints/codelewm-v0-2-action-swap-rerun-20260520-7c7cb0b/config.json
ADDED
|
@@ -0,0 +1,60 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"data": {
|
| 3 |
+
"manifest": "/CodeLeWM/.artifacts/hf/codelewm-v0-2-action-swap-rerun-20260520-7c7cb0b/pack/manifest.json",
|
| 4 |
+
"train": "/CodeLeWM/.artifacts/hf/codelewm-v0-2-action-swap-rerun-20260520-7c7cb0b/pack/hdf5/train.hdf5",
|
| 5 |
+
"val": "/CodeLeWM/.artifacts/hf/codelewm-v0-2-action-swap-rerun-20260520-7c7cb0b/pack/hdf5/val.hdf5"
|
| 6 |
+
},
|
| 7 |
+
"loader": {
|
| 8 |
+
"batch_size": 64,
|
| 9 |
+
"num_workers": 4,
|
| 10 |
+
"persistent_workers": true,
|
| 11 |
+
"pin_memory": true,
|
| 12 |
+
"shuffle": true
|
| 13 |
+
},
|
| 14 |
+
"loss": {
|
| 15 |
+
"action_swap_contrastive_margin": 0.05,
|
| 16 |
+
"action_swap_contrastive_weight": 0.2,
|
| 17 |
+
"action_use_margin": 0.02,
|
| 18 |
+
"action_use_margin_weight": 0.25,
|
| 19 |
+
"enable_action_swap_contrastive": true,
|
| 20 |
+
"enable_action_use_margin": true,
|
| 21 |
+
"enable_inverse_action_reconstruction": true,
|
| 22 |
+
"enable_retrieval_loss": false,
|
| 23 |
+
"inverse_action_reconstruction_weight": 0.1,
|
| 24 |
+
"retrieval_temperature": 0.1,
|
| 25 |
+
"retrieval_weight": 0.0,
|
| 26 |
+
"sigreg_knots": 17,
|
| 27 |
+
"sigreg_num_proj": 1024,
|
| 28 |
+
"sigreg_weight": 0.09
|
| 29 |
+
},
|
| 30 |
+
"name": "codelewm_scaled_v0_2_action_swap_inverse_gpu_a10g",
|
| 31 |
+
"optimizer": {
|
| 32 |
+
"lr": 0.0001,
|
| 33 |
+
"type": "AdamW",
|
| 34 |
+
"weight_decay": 0.001
|
| 35 |
+
},
|
| 36 |
+
"output": {
|
| 37 |
+
"checkpoint_dir": "/CodeLeWM/.artifacts/hf/codelewm-v0-2-action-swap-rerun-20260520-7c7cb0b/train/checkpoints",
|
| 38 |
+
"manifest_path": "/CodeLeWM/.artifacts/hf/codelewm-v0-2-action-swap-rerun-20260520-7c7cb0b/train/training_manifest.json",
|
| 39 |
+
"metrics_path": "/CodeLeWM/.artifacts/hf/codelewm-v0-2-action-swap-rerun-20260520-7c7cb0b/train/metrics.jsonl",
|
| 40 |
+
"run_dir": "/CodeLeWM/.artifacts/hf/codelewm-v0-2-action-swap-rerun-20260520-7c7cb0b/train"
|
| 41 |
+
},
|
| 42 |
+
"schema_version": "codelewm.train_config.v1",
|
| 43 |
+
"seed": 240119,
|
| 44 |
+
"trainer": {
|
| 45 |
+
"accelerator": "gpu",
|
| 46 |
+
"devices": 1,
|
| 47 |
+
"gradient_clip_val": 1.0,
|
| 48 |
+
"max_steps": 60000,
|
| 49 |
+
"precision": "bf16-mixed"
|
| 50 |
+
},
|
| 51 |
+
"wm": {
|
| 52 |
+
"action_fusion": "gated_residual",
|
| 53 |
+
"action_sequence_length": 256,
|
| 54 |
+
"action_view": "text",
|
| 55 |
+
"embed_dim": 256,
|
| 56 |
+
"history_size": 1,
|
| 57 |
+
"num_preds": 1,
|
| 58 |
+
"state_sequence_length": 1024
|
| 59 |
+
}
|
| 60 |
+
}
|
checkpoints/codelewm-v0-2-action-swap-rerun-20260520-7c7cb0b/manifest.json
ADDED
|
@@ -0,0 +1,130 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"artifact_id": "training_run-0a41863d1da33737",
|
| 3 |
+
"artifact_kind": "training_run",
|
| 4 |
+
"command": [
|
| 5 |
+
"codelewm",
|
| 6 |
+
"train",
|
| 7 |
+
"--config",
|
| 8 |
+
"/CodeLeWM/.artifacts/hf/codelewm-v0-2-action-swap-rerun-20260520-7c7cb0b/configs/train.json",
|
| 9 |
+
"--executor",
|
| 10 |
+
"torch",
|
| 11 |
+
"--device",
|
| 12 |
+
"auto",
|
| 13 |
+
"--overwrite",
|
| 14 |
+
"--json",
|
| 15 |
+
"--log-jsonl",
|
| 16 |
+
"/CodeLeWM/.artifacts/hf/codelewm-v0-2-action-swap-rerun-20260520-7c7cb0b/logs/train.jsonl"
|
| 17 |
+
],
|
| 18 |
+
"config_sha256": "6bf543fe67dc43831a0c450b746986641f31cdf33179d8cb669fc05968ff967b",
|
| 19 |
+
"created_at": "2026-05-20T18:04:49Z",
|
| 20 |
+
"files": [
|
| 21 |
+
{
|
| 22 |
+
"bytes": 2069,
|
| 23 |
+
"path": "config.json",
|
| 24 |
+
"sha256": "beb59b73ecd13e78852635c9f8eb3f37bd2aceb36ff563d9599f27d8dd5abec4"
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"bytes": 2144,
|
| 28 |
+
"path": "metrics.jsonl",
|
| 29 |
+
"sha256": "4858f91df9b0b5e51dc97e11583caab637048baa9b51e0ecb7d2f9288f7da91f"
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"bytes": 2326,
|
| 33 |
+
"path": "reports/metrics_report.json",
|
| 34 |
+
"sha256": "794990f624ccde341d9c630df0c623161545915ff46a31fb237b99808f6873fd"
|
| 35 |
+
},
|
| 36 |
+
{
|
| 37 |
+
"bytes": 254388041,
|
| 38 |
+
"path": "checkpoints/checkpoint.pt",
|
| 39 |
+
"sha256": "f2c5ba50ee0ec5e32ff5c3ceed848020e989ebdb1c98a917f17589ee523c6d7e"
|
| 40 |
+
},
|
| 41 |
+
{
|
| 42 |
+
"bytes": 511,
|
| 43 |
+
"path": "checkpoints/checkpoint.pt.manifest.json",
|
| 44 |
+
"sha256": "98b3be6594bcb3d6e59999b47017ea146083d0d7ce54d675bfdc707d09c0b5bf"
|
| 45 |
+
},
|
| 46 |
+
{
|
| 47 |
+
"bytes": 3608,
|
| 48 |
+
"path": "reports/torch_training_report.json",
|
| 49 |
+
"sha256": "7a669bfeef22ff0e3420fc4b7453fdb436b76472bfa633b0ebef28cc9d616165"
|
| 50 |
+
}
|
| 51 |
+
],
|
| 52 |
+
"metadata": {
|
| 53 |
+
"dataset_manifest_path": ".artifacts/hf/codelewm-v0-2-action-swap-rerun-20260520-7c7cb0b/pack/manifest.json",
|
| 54 |
+
"executor": {
|
| 55 |
+
"checkpoint_schema_version": "codelewm.checkpoint.v1",
|
| 56 |
+
"device": "cuda",
|
| 57 |
+
"executor": "torch",
|
| 58 |
+
"objective": {
|
| 59 |
+
"action_swap_contrastive_margin": 0.05,
|
| 60 |
+
"action_swap_contrastive_weight": 0.2,
|
| 61 |
+
"action_use_margin": 0.02,
|
| 62 |
+
"action_use_margin_weight": 0.25,
|
| 63 |
+
"enable_action_swap_contrastive": true,
|
| 64 |
+
"enable_action_use_margin": true,
|
| 65 |
+
"enable_inverse_action_reconstruction": true,
|
| 66 |
+
"enable_retrieval_loss": false,
|
| 67 |
+
"inverse_action_reconstruction_weight": 0.1,
|
| 68 |
+
"retrieval_temperature": 0.1,
|
| 69 |
+
"retrieval_weight": 0.0,
|
| 70 |
+
"sigreg_knots": 17,
|
| 71 |
+
"sigreg_num_proj": 1024,
|
| 72 |
+
"sigreg_weight": 0.09
|
| 73 |
+
},
|
| 74 |
+
"precision": "bf16-mixed",
|
| 75 |
+
"torch": "2.12.0+cu130",
|
| 76 |
+
"train_rows": 18019,
|
| 77 |
+
"val_rows": 1291
|
| 78 |
+
},
|
| 79 |
+
"final_metrics": {
|
| 80 |
+
"action_diagnostics/positive_distance": 0.004852294921875,
|
| 81 |
+
"action_diagnostics/swap_distance_gap": 0.26171875,
|
| 82 |
+
"action_diagnostics/swapped_distance": 0.265625,
|
| 83 |
+
"collapse/effective_rank": 4.03470884645853,
|
| 84 |
+
"collapse/effective_rank_ratio": 0.015760581431478633,
|
| 85 |
+
"collapse/embedding_count": 192.0,
|
| 86 |
+
"collapse/embedding_norm_mean": 14.392418570878986,
|
| 87 |
+
"collapse/latent_dim": 256.0,
|
| 88 |
+
"collapse/nearest_neighbor_entropy": 4.822367445946839,
|
| 89 |
+
"collapse/pairwise_cosine_mean": 0.006447293492730005,
|
| 90 |
+
"collapse/per_dim_variance_max": 3.456151289617992,
|
| 91 |
+
"collapse/per_dim_variance_median": 0.7004040151172777,
|
| 92 |
+
"collapse/per_dim_variance_min": 0.04232150622637062,
|
| 93 |
+
"loss/action_swap_contrastive": 0.0048749265260994434,
|
| 94 |
+
"loss/action_swap_contrastive_weighted": 0.0009749853052198887,
|
| 95 |
+
"loss/action_use_margin": 0.007081065326929092,
|
| 96 |
+
"loss/action_use_margin_weighted": 0.001770266331732273,
|
| 97 |
+
"loss/inverse_action_reconstruction": 0.21137171983718872,
|
| 98 |
+
"loss/inverse_action_reconstruction_weighted": 0.021137172356247902,
|
| 99 |
+
"loss/prediction_mse": 0.004846842493861914,
|
| 100 |
+
"loss/sigreg": 1.0546875,
|
| 101 |
+
"loss/sigreg_weighted": 0.0947265625,
|
| 102 |
+
"loss/total": 0.12345582246780396,
|
| 103 |
+
"train/examples": 3833852.0,
|
| 104 |
+
"train/examples_per_second": 1323.9171764663713,
|
| 105 |
+
"train/gradient_norm": 2.023609161376953,
|
| 106 |
+
"val/action_diagnostics/positive_distance": 0.1494137846997806,
|
| 107 |
+
"val/action_diagnostics/swap_distance_gap": 0.03513682243369874,
|
| 108 |
+
"val/action_diagnostics/swapped_distance": 0.18455060713347934,
|
| 109 |
+
"val/loss/action_swap_contrastive": 0.059425451925822666,
|
| 110 |
+
"val/loss/action_swap_contrastive_weighted": 0.011885090686735652,
|
| 111 |
+
"val/loss/action_use_margin": 0.05836626293048972,
|
| 112 |
+
"val/loss/action_use_margin_weighted": 0.01459156573262243,
|
| 113 |
+
"val/loss/inverse_action_reconstruction": 0.552373686007091,
|
| 114 |
+
"val/loss/inverse_action_reconstruction_weighted": 0.055237369877951484,
|
| 115 |
+
"val/loss/prediction_mse": 0.14941378363541194,
|
| 116 |
+
"val/loss/sigreg": 2.168621886344183,
|
| 117 |
+
"val/loss/sigreg_weighted": 0.19517597804466882,
|
| 118 |
+
"val/loss/total": 0.4263037897291638
|
| 119 |
+
},
|
| 120 |
+
"run_id": "codelewm_scaled_v0_2_action_swap_inverse_gpu_a10g",
|
| 121 |
+
"schema_version": "codelewm.training_run.v1",
|
| 122 |
+
"seed": 240119,
|
| 123 |
+
"step_count": 60000
|
| 124 |
+
},
|
| 125 |
+
"parent_artifacts": [
|
| 126 |
+
"dataset-daecac9f9965c563"
|
| 127 |
+
],
|
| 128 |
+
"schema_version": "codelewm.artifact_manifest.v1",
|
| 129 |
+
"source_git_sha": "7c7cb0b8fe132e4819f05a77585c254267e77574"
|
| 130 |
+
}
|
checkpoints/codelewm-v0-2-action-swap-rerun-20260520-7c7cb0b/metrics.jsonl
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
{"metrics": {"action_diagnostics/positive_distance": 0.004852294921875, "action_diagnostics/swap_distance_gap": 0.26171875, "action_diagnostics/swapped_distance": 0.265625, "collapse/effective_rank": 4.03470884645853, "collapse/effective_rank_ratio": 0.015760581431478633, "collapse/embedding_count": 192.0, "collapse/embedding_norm_mean": 14.392418570878986, "collapse/latent_dim": 256.0, "collapse/nearest_neighbor_entropy": 4.822367445946839, "collapse/pairwise_cosine_mean": 0.006447293492730005, "collapse/per_dim_variance_max": 3.456151289617992, "collapse/per_dim_variance_median": 0.7004040151172777, "collapse/per_dim_variance_min": 0.04232150622637062, "loss/action_swap_contrastive": 0.0048749265260994434, "loss/action_swap_contrastive_weighted": 0.0009749853052198887, "loss/action_use_margin": 0.007081065326929092, "loss/action_use_margin_weighted": 0.001770266331732273, "loss/inverse_action_reconstruction": 0.21137171983718872, "loss/inverse_action_reconstruction_weighted": 0.021137172356247902, "loss/prediction_mse": 0.004846842493861914, "loss/sigreg": 1.0546875, "loss/sigreg_weighted": 0.0947265625, "loss/total": 0.12345582246780396, "train/examples": 3833852.0, "train/examples_per_second": 1323.9171764663713, "train/gradient_norm": 2.023609161376953, "val/action_diagnostics/positive_distance": 0.1494137846997806, "val/action_diagnostics/swap_distance_gap": 0.03513682243369874, "val/action_diagnostics/swapped_distance": 0.18455060713347934, "val/loss/action_swap_contrastive": 0.059425451925822666, "val/loss/action_swap_contrastive_weighted": 0.011885090686735652, "val/loss/action_use_margin": 0.05836626293048972, "val/loss/action_use_margin_weighted": 0.01459156573262243, "val/loss/inverse_action_reconstruction": 0.552373686007091, "val/loss/inverse_action_reconstruction_weighted": 0.055237369877951484, "val/loss/prediction_mse": 0.14941378363541194, "val/loss/sigreg": 2.168621886344183, "val/loss/sigreg_weighted": 0.19517597804466882, "val/loss/total": 0.4263037897291638}, "run_id": "codelewm_scaled_v0_2_action_swap_inverse_gpu_a10g", "schema_version": "codelewm.training_metrics.v1", "step": 60000}
|
checkpoints/codelewm-v0-2-action-swap-rerun-20260520-7c7cb0b/reports/metrics_report.json
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"final_metrics": {
|
| 3 |
+
"action_diagnostics/positive_distance": 0.004852294921875,
|
| 4 |
+
"action_diagnostics/swap_distance_gap": 0.26171875,
|
| 5 |
+
"action_diagnostics/swapped_distance": 0.265625,
|
| 6 |
+
"collapse/effective_rank": 4.03470884645853,
|
| 7 |
+
"collapse/effective_rank_ratio": 0.015760581431478633,
|
| 8 |
+
"collapse/embedding_count": 192.0,
|
| 9 |
+
"collapse/embedding_norm_mean": 14.392418570878986,
|
| 10 |
+
"collapse/latent_dim": 256.0,
|
| 11 |
+
"collapse/nearest_neighbor_entropy": 4.822367445946839,
|
| 12 |
+
"collapse/pairwise_cosine_mean": 0.006447293492730005,
|
| 13 |
+
"collapse/per_dim_variance_max": 3.456151289617992,
|
| 14 |
+
"collapse/per_dim_variance_median": 0.7004040151172777,
|
| 15 |
+
"collapse/per_dim_variance_min": 0.04232150622637062,
|
| 16 |
+
"loss/action_swap_contrastive": 0.0048749265260994434,
|
| 17 |
+
"loss/action_swap_contrastive_weighted": 0.0009749853052198887,
|
| 18 |
+
"loss/action_use_margin": 0.007081065326929092,
|
| 19 |
+
"loss/action_use_margin_weighted": 0.001770266331732273,
|
| 20 |
+
"loss/inverse_action_reconstruction": 0.21137171983718872,
|
| 21 |
+
"loss/inverse_action_reconstruction_weighted": 0.021137172356247902,
|
| 22 |
+
"loss/prediction_mse": 0.004846842493861914,
|
| 23 |
+
"loss/sigreg": 1.0546875,
|
| 24 |
+
"loss/sigreg_weighted": 0.0947265625,
|
| 25 |
+
"loss/total": 0.12345582246780396,
|
| 26 |
+
"train/examples": 3833852.0,
|
| 27 |
+
"train/examples_per_second": 1323.9171764663713,
|
| 28 |
+
"train/gradient_norm": 2.023609161376953,
|
| 29 |
+
"val/action_diagnostics/positive_distance": 0.1494137846997806,
|
| 30 |
+
"val/action_diagnostics/swap_distance_gap": 0.03513682243369874,
|
| 31 |
+
"val/action_diagnostics/swapped_distance": 0.18455060713347934,
|
| 32 |
+
"val/loss/action_swap_contrastive": 0.059425451925822666,
|
| 33 |
+
"val/loss/action_swap_contrastive_weighted": 0.011885090686735652,
|
| 34 |
+
"val/loss/action_use_margin": 0.05836626293048972,
|
| 35 |
+
"val/loss/action_use_margin_weighted": 0.01459156573262243,
|
| 36 |
+
"val/loss/inverse_action_reconstruction": 0.552373686007091,
|
| 37 |
+
"val/loss/inverse_action_reconstruction_weighted": 0.055237369877951484,
|
| 38 |
+
"val/loss/prediction_mse": 0.14941378363541194,
|
| 39 |
+
"val/loss/sigreg": 2.168621886344183,
|
| 40 |
+
"val/loss/sigreg_weighted": 0.19517597804466882,
|
| 41 |
+
"val/loss/total": 0.4263037897291638
|
| 42 |
+
},
|
| 43 |
+
"run_id": "codelewm_scaled_v0_2_action_swap_inverse_gpu_a10g",
|
| 44 |
+
"schema_version": "codelewm.training_metrics.v1",
|
| 45 |
+
"step_count": 60000
|
| 46 |
+
}
|
checkpoints/codelewm-v0-2-action-swap-rerun-20260520-7c7cb0b/reports/torch_training_report.json
ADDED
|
@@ -0,0 +1,86 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"collapse_report": {
|
| 3 |
+
"effective_rank": 4.03470884645853,
|
| 4 |
+
"effective_rank_ratio": 0.015760581431478633,
|
| 5 |
+
"embedding_count": 192,
|
| 6 |
+
"embedding_norm_mean": 14.392418570878986,
|
| 7 |
+
"latent_dim": 256,
|
| 8 |
+
"nearest_neighbor_entropy": 4.822367445946839,
|
| 9 |
+
"pairwise_cosine_mean": 0.006447293492730005,
|
| 10 |
+
"per_dim_variance_max": 3.456151289617992,
|
| 11 |
+
"per_dim_variance_median": 0.7004040151172777,
|
| 12 |
+
"per_dim_variance_min": 0.04232150622637062,
|
| 13 |
+
"schema_version": "codelewm.eval.collapse_report.v1"
|
| 14 |
+
},
|
| 15 |
+
"dataset": {
|
| 16 |
+
"action_view": "text",
|
| 17 |
+
"train_rows": 18019,
|
| 18 |
+
"val_rows": 1291
|
| 19 |
+
},
|
| 20 |
+
"metrics": {
|
| 21 |
+
"action_diagnostics/positive_distance": 0.004852294921875,
|
| 22 |
+
"action_diagnostics/swap_distance_gap": 0.26171875,
|
| 23 |
+
"action_diagnostics/swapped_distance": 0.265625,
|
| 24 |
+
"collapse/effective_rank": 4.03470884645853,
|
| 25 |
+
"collapse/effective_rank_ratio": 0.015760581431478633,
|
| 26 |
+
"collapse/embedding_count": 192.0,
|
| 27 |
+
"collapse/embedding_norm_mean": 14.392418570878986,
|
| 28 |
+
"collapse/latent_dim": 256.0,
|
| 29 |
+
"collapse/nearest_neighbor_entropy": 4.822367445946839,
|
| 30 |
+
"collapse/pairwise_cosine_mean": 0.006447293492730005,
|
| 31 |
+
"collapse/per_dim_variance_max": 3.456151289617992,
|
| 32 |
+
"collapse/per_dim_variance_median": 0.7004040151172777,
|
| 33 |
+
"collapse/per_dim_variance_min": 0.04232150622637062,
|
| 34 |
+
"loss/action_swap_contrastive": 0.0048749265260994434,
|
| 35 |
+
"loss/action_swap_contrastive_weighted": 0.0009749853052198887,
|
| 36 |
+
"loss/action_use_margin": 0.007081065326929092,
|
| 37 |
+
"loss/action_use_margin_weighted": 0.001770266331732273,
|
| 38 |
+
"loss/inverse_action_reconstruction": 0.21137171983718872,
|
| 39 |
+
"loss/inverse_action_reconstruction_weighted": 0.021137172356247902,
|
| 40 |
+
"loss/prediction_mse": 0.004846842493861914,
|
| 41 |
+
"loss/sigreg": 1.0546875,
|
| 42 |
+
"loss/sigreg_weighted": 0.0947265625,
|
| 43 |
+
"loss/total": 0.12345582246780396,
|
| 44 |
+
"train/examples": 3833852.0,
|
| 45 |
+
"train/examples_per_second": 1323.9171764663713,
|
| 46 |
+
"train/gradient_norm": 2.023609161376953,
|
| 47 |
+
"val/action_diagnostics/positive_distance": 0.1494137846997806,
|
| 48 |
+
"val/action_diagnostics/swap_distance_gap": 0.03513682243369874,
|
| 49 |
+
"val/action_diagnostics/swapped_distance": 0.18455060713347934,
|
| 50 |
+
"val/loss/action_swap_contrastive": 0.059425451925822666,
|
| 51 |
+
"val/loss/action_swap_contrastive_weighted": 0.011885090686735652,
|
| 52 |
+
"val/loss/action_use_margin": 0.05836626293048972,
|
| 53 |
+
"val/loss/action_use_margin_weighted": 0.01459156573262243,
|
| 54 |
+
"val/loss/inverse_action_reconstruction": 0.552373686007091,
|
| 55 |
+
"val/loss/inverse_action_reconstruction_weighted": 0.055237369877951484,
|
| 56 |
+
"val/loss/prediction_mse": 0.14941378363541194,
|
| 57 |
+
"val/loss/sigreg": 2.168621886344183,
|
| 58 |
+
"val/loss/sigreg_weighted": 0.19517597804466882,
|
| 59 |
+
"val/loss/total": 0.4263037897291638
|
| 60 |
+
},
|
| 61 |
+
"objective": {
|
| 62 |
+
"action_swap_contrastive_margin": 0.05,
|
| 63 |
+
"action_swap_contrastive_weight": 0.2,
|
| 64 |
+
"action_use_margin": 0.02,
|
| 65 |
+
"action_use_margin_weight": 0.25,
|
| 66 |
+
"enable_action_swap_contrastive": true,
|
| 67 |
+
"enable_action_use_margin": true,
|
| 68 |
+
"enable_inverse_action_reconstruction": true,
|
| 69 |
+
"enable_retrieval_loss": false,
|
| 70 |
+
"inverse_action_reconstruction_weight": 0.1,
|
| 71 |
+
"retrieval_temperature": 0.1,
|
| 72 |
+
"retrieval_weight": 0.0,
|
| 73 |
+
"sigreg_knots": 17,
|
| 74 |
+
"sigreg_num_proj": 1024,
|
| 75 |
+
"sigreg_weight": 0.09
|
| 76 |
+
},
|
| 77 |
+
"run_id": "codelewm_scaled_v0_2_action_swap_inverse_gpu_a10g",
|
| 78 |
+
"runtime": {
|
| 79 |
+
"device": "cuda",
|
| 80 |
+
"dtype": "torch.bfloat16",
|
| 81 |
+
"precision": "bf16-mixed",
|
| 82 |
+
"torch": "2.12.0+cu130"
|
| 83 |
+
},
|
| 84 |
+
"schema_version": "codelewm.torch_training_report.v1",
|
| 85 |
+
"step_count": 60000
|
| 86 |
+
}
|
checkpoints/codelewm-v0-2-action-swap-rerun-20260520-7c7cb0b/training_manifest.json
ADDED
|
@@ -0,0 +1,103 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"artifact_manifest_id": "training_run-0a41863d1da33737",
|
| 3 |
+
"artifact_manifest_path": "manifest.json",
|
| 4 |
+
"checkpoint_files": [
|
| 5 |
+
{
|
| 6 |
+
"bytes": 254388041,
|
| 7 |
+
"path": "checkpoints/checkpoint.pt",
|
| 8 |
+
"sha256": "f2c5ba50ee0ec5e32ff5c3ceed848020e989ebdb1c98a917f17589ee523c6d7e"
|
| 9 |
+
},
|
| 10 |
+
{
|
| 11 |
+
"bytes": 511,
|
| 12 |
+
"path": "checkpoints/checkpoint.pt.manifest.json",
|
| 13 |
+
"sha256": "98b3be6594bcb3d6e59999b47017ea146083d0d7ce54d675bfdc707d09c0b5bf"
|
| 14 |
+
}
|
| 15 |
+
],
|
| 16 |
+
"config_path": "config.json",
|
| 17 |
+
"config_sha256": "6bf543fe67dc43831a0c450b746986641f31cdf33179d8cb669fc05968ff967b",
|
| 18 |
+
"dataset_manifest_path": ".artifacts/hf/codelewm-v0-2-action-swap-rerun-20260520-7c7cb0b/pack/manifest.json",
|
| 19 |
+
"final_metrics": {
|
| 20 |
+
"action_diagnostics/positive_distance": 0.004852294921875,
|
| 21 |
+
"action_diagnostics/swap_distance_gap": 0.26171875,
|
| 22 |
+
"action_diagnostics/swapped_distance": 0.265625,
|
| 23 |
+
"collapse/effective_rank": 4.03470884645853,
|
| 24 |
+
"collapse/effective_rank_ratio": 0.015760581431478633,
|
| 25 |
+
"collapse/embedding_count": 192.0,
|
| 26 |
+
"collapse/embedding_norm_mean": 14.392418570878986,
|
| 27 |
+
"collapse/latent_dim": 256.0,
|
| 28 |
+
"collapse/nearest_neighbor_entropy": 4.822367445946839,
|
| 29 |
+
"collapse/pairwise_cosine_mean": 0.006447293492730005,
|
| 30 |
+
"collapse/per_dim_variance_max": 3.456151289617992,
|
| 31 |
+
"collapse/per_dim_variance_median": 0.7004040151172777,
|
| 32 |
+
"collapse/per_dim_variance_min": 0.04232150622637062,
|
| 33 |
+
"loss/action_swap_contrastive": 0.0048749265260994434,
|
| 34 |
+
"loss/action_swap_contrastive_weighted": 0.0009749853052198887,
|
| 35 |
+
"loss/action_use_margin": 0.007081065326929092,
|
| 36 |
+
"loss/action_use_margin_weighted": 0.001770266331732273,
|
| 37 |
+
"loss/inverse_action_reconstruction": 0.21137171983718872,
|
| 38 |
+
"loss/inverse_action_reconstruction_weighted": 0.021137172356247902,
|
| 39 |
+
"loss/prediction_mse": 0.004846842493861914,
|
| 40 |
+
"loss/sigreg": 1.0546875,
|
| 41 |
+
"loss/sigreg_weighted": 0.0947265625,
|
| 42 |
+
"loss/total": 0.12345582246780396,
|
| 43 |
+
"train/examples": 3833852.0,
|
| 44 |
+
"train/examples_per_second": 1323.9171764663713,
|
| 45 |
+
"train/gradient_norm": 2.023609161376953,
|
| 46 |
+
"val/action_diagnostics/positive_distance": 0.1494137846997806,
|
| 47 |
+
"val/action_diagnostics/swap_distance_gap": 0.03513682243369874,
|
| 48 |
+
"val/action_diagnostics/swapped_distance": 0.18455060713347934,
|
| 49 |
+
"val/loss/action_swap_contrastive": 0.059425451925822666,
|
| 50 |
+
"val/loss/action_swap_contrastive_weighted": 0.011885090686735652,
|
| 51 |
+
"val/loss/action_use_margin": 0.05836626293048972,
|
| 52 |
+
"val/loss/action_use_margin_weighted": 0.01459156573262243,
|
| 53 |
+
"val/loss/inverse_action_reconstruction": 0.552373686007091,
|
| 54 |
+
"val/loss/inverse_action_reconstruction_weighted": 0.055237369877951484,
|
| 55 |
+
"val/loss/prediction_mse": 0.14941378363541194,
|
| 56 |
+
"val/loss/sigreg": 2.168621886344183,
|
| 57 |
+
"val/loss/sigreg_weighted": 0.19517597804466882,
|
| 58 |
+
"val/loss/total": 0.4263037897291638
|
| 59 |
+
},
|
| 60 |
+
"metadata": {
|
| 61 |
+
"executor": {
|
| 62 |
+
"checkpoint_schema_version": "codelewm.checkpoint.v1",
|
| 63 |
+
"device": "cuda",
|
| 64 |
+
"executor": "torch",
|
| 65 |
+
"objective": {
|
| 66 |
+
"action_swap_contrastive_margin": 0.05,
|
| 67 |
+
"action_swap_contrastive_weight": 0.2,
|
| 68 |
+
"action_use_margin": 0.02,
|
| 69 |
+
"action_use_margin_weight": 0.25,
|
| 70 |
+
"enable_action_swap_contrastive": true,
|
| 71 |
+
"enable_action_use_margin": true,
|
| 72 |
+
"enable_inverse_action_reconstruction": true,
|
| 73 |
+
"enable_retrieval_loss": false,
|
| 74 |
+
"inverse_action_reconstruction_weight": 0.1,
|
| 75 |
+
"retrieval_temperature": 0.1,
|
| 76 |
+
"retrieval_weight": 0.0,
|
| 77 |
+
"sigreg_knots": 17,
|
| 78 |
+
"sigreg_num_proj": 1024,
|
| 79 |
+
"sigreg_weight": 0.09
|
| 80 |
+
},
|
| 81 |
+
"precision": "bf16-mixed",
|
| 82 |
+
"torch": "2.12.0+cu130",
|
| 83 |
+
"train_rows": 18019,
|
| 84 |
+
"val_rows": 1291
|
| 85 |
+
}
|
| 86 |
+
},
|
| 87 |
+
"metrics_path": "metrics.jsonl",
|
| 88 |
+
"metrics_report_path": "reports/metrics_report.json",
|
| 89 |
+
"parent_artifacts": [
|
| 90 |
+
"dataset-daecac9f9965c563"
|
| 91 |
+
],
|
| 92 |
+
"report_files": [
|
| 93 |
+
{
|
| 94 |
+
"bytes": 3608,
|
| 95 |
+
"path": "reports/torch_training_report.json",
|
| 96 |
+
"sha256": "7a669bfeef22ff0e3420fc4b7453fdb436b76472bfa633b0ebef28cc9d616165"
|
| 97 |
+
}
|
| 98 |
+
],
|
| 99 |
+
"run_id": "codelewm_scaled_v0_2_action_swap_inverse_gpu_a10g",
|
| 100 |
+
"schema_version": "codelewm.training_run.v1",
|
| 101 |
+
"seed": 240119,
|
| 102 |
+
"step_count": 60000
|
| 103 |
+
}
|