Add final full-suite aligned v0 1/4 checkpoint (training only)

README.md

@@ -9,7 +9,7 @@ tags:
 
 # LoopWAM final checkpoints
 
-Final trained policy weights from the LoopWAM experiments. This repository contains **13 checkpoints**, separated by training data scope. Both original and repeated 4/4 runs are retained. All production checkpoints are from completed ten-epoch runs with global batch 128 and training seed 42.
+Final trained policy weights from the LoopWAM experiments. This repository contains **14 checkpoints**, separated by training data scope. Both original and repeated 4/4 runs are retained. All production checkpoints are from completed ten-epoch runs with global batch 128 and training seed 42.
 
 ## October 9 additions: all-loop KV
 
@@ -32,10 +32,19 @@ original three-seed decision remains inconclusive. This grid overlaps the
 original states and must not be pooled as independent evidence.
 See the [updated results report](https://github.com/anhdao69/LoopWAM_NT/blob/KV_concat/plans/performance/LoopWAM_v0_V4A1_KVconcat_20261008.md).
 
+## October 10 addition: full-suite aligned 1/4
+
+[libero-all-suites/v0-video1-action4](libero-all-suites/v0-video1-action4)
+contains the final ten-epoch full-suite v0 aligned 1/4 policy (21,700 updates,
+global batch 128, training seed 42). Training took 10h 58m 46s on two H100s.
+**Not evaluated:** this run finished under the requested training-only workflow.
+Normalization, training metadata, fairness checks and export hashes are included.
+
 ## Checkpoint index
 
 | Directory | Model | Video/action loops | Parameters | Final evaluation |
 |---|---|---|---:|---|
+| [libero-all-suites/v0-video1-action4](libero-all-suites/v0-video1-action4) | v0 aligned | 1/4 | 584536135 | Not evaluated (training only) |
 | [libero-long/v0-video4-action4-original](libero-long/v0-video4-action4-original) | v0 | 4/4 | 584,536,135 | 96/100 (96%) |
 | [libero-long/v1-video4-action4](libero-long/v1-video4-action4) | v1 | 4/4 | 584,536,135 | 82/100 (82%) |
 | [libero-long/dense-s12](libero-long/dense-s12) | dense_s12 | 1/1 | 584,536,135 | 81/100 (81%) |

checkpoint_index.json

@@ -335,7 +335,36 @@
       "source_revision": "3a89bfac5cfccad66bfebbe00014f7f739a78adc",
       "compatible_code_revision": "3a89bfac5cfccad66bfebbe00014f7f739a78adc",
       "normalization_sha256": "78058a689c41382a2759541d5590cb3ada8c576e736285aaeebc344874366f1b"
+    },
+    {
+      "id": "v0_full_1_4",
+      "repo_path": "libero-all-suites/v0-video1-action4",
+      "training_path": "kv_campaign_job4771/campaign_3a89bfa/full_14/train",
+      "source_sha256": "f7f3f20c721c5072ab8b476e45819d400bdb4e0635402c09681a24d45deebd71",
+      "source_bytes": 7015569623,
+      "export_sha256": "801922e30df320cd7d9f3c642d90315865d49807ec20b6e27ea93c219d2f9a19",
+      "export_bytes": 2338635000,
+      "format": "loopwam-s-v1",
+      "optimizer_included": false,
+      "step": 21700,
+      "version": "v0",
+      "video_loops": 1,
+      "action_loops": 4,
+      "action_kv_mode": "aligned",
+      "parameters": 584536135,
+      "dataset_scope": "full_libero",
+      "epochs": 10,
+      "global_batch": 128,
+      "training_seed": 42,
+      "training_seconds": 39525.92356655188,
+      "evaluation_status": "not_run_training_only",
+      "success_rate": null,
+      "episodes": 0,
+      "tensor_equality_verified": true,
+      "source_revision": "3a89bfac5cfccad66bfebbe00014f7f739a78adc",
+      "compatible_code_revision": "3a89bfac5cfccad66bfebbe00014f7f739a78adc",
+      "normalization_sha256": "d5ad351eb3272f416d34082e8af5483ba1118ceca355a37d9f268fdb269fb3e4"
     }
   ],
-  "updated_utc": "2026-10-09T16:24:40.158405+00:00"
+  "updated_utc": "2026-10-10T02:41:05.842410+00:00"
 }

libero-all-suites/v0-video1-action4/README.md

@@ -0,0 +1,20 @@
+# LoopWAM v0 aligned 1/4 — all four LIBERO suites
+
+Final policy from fresh canonical initialization, training seed 42, global batch
+128, ten epochs (21,700 updates; 2,777,130 real windows). Video/action loops 1/4;
+effective depths 12/30. Training completed in 10h 58m 46s on two H100s.
+
+No simulator evaluation was run for this checkpoint: the campaign was changed
+to training-only at the owner's request. No success rate is claimed.
+
+Use the KV_concat branch of https://github.com/anhdao69/LoopWAM_NT,
+revision 3a89bfac5cfccad66bfebbe00014f7f739a78adc or compatible later code.
+Use the supplied dataset_stats.json for normalization. Wan VAE, text embeddings,
+and observation preprocessing are external dependencies. Prefer eager BF16
+inference with FP32 policy weights as in the existing evaluation setup.
+
+All model tensors and retained metadata were checked exactly against the final
+native checkpoint; only the Adam optimizer state was removed. This export is
+for inference or fresh-optimizer fine-tuning, not exact training resume. The
+full resumable checkpoint remains on the training server. See export.json for
+source/export SHA-256 hashes and training provenance.

libero-all-suites/v0-video1-action4/dataset_stats.json

@@ -0,0 +1,50 @@
+{
+  "action": {
+    "default": {
+      "global_max": [
+        0.9375,
+        0.9375,
+        0.9375,
+        0.3557142913341522,
+        0.375,
+        0.375,
+        1.0
+      ],
+      "global_min": [
+        -0.9375,
+        -0.9375,
+        -0.9375,
+        -0.24214285612106323,
+        -0.375,
+        -0.3642857074737549,
+        0.0
+      ]
+    }
+  },
+  "num_episodes": 1712,
+  "num_transition": 277713,
+  "state": {
+    "default": {
+      "global_max": [
+        0.2103137969970703,
+        0.3904264271259308,
+        1.472778081893921,
+        3.7248642444610596,
+        3.5618896484375,
+        1.3863215446472168,
+        0.04232141748070717,
+        0.0013126095291227102
+      ],
+      "global_min": [
+        -0.48278069496154785,
+        -0.3309336006641388,
+        0.008128181099891663,
+        1.002794623374939,
+        -3.6312508583068848,
+        -1.842738389968872,
+        -0.005453015677630901,
+        -0.042015016078948975
+      ]
+    }
+  }
+}

libero-all-suites/v0-video1-action4/export.json

@@ -0,0 +1,29 @@
+{
+  "id": "v0_full_1_4",
+  "repo_path": "libero-all-suites/v0-video1-action4",
+  "training_path": "kv_campaign_job4771/campaign_3a89bfa/full_14/train",
+  "source_sha256": "f7f3f20c721c5072ab8b476e45819d400bdb4e0635402c09681a24d45deebd71",
+  "source_bytes": 7015569623,
+  "export_sha256": "801922e30df320cd7d9f3c642d90315865d49807ec20b6e27ea93c219d2f9a19",
+  "export_bytes": 2338635000,
+  "format": "loopwam-s-v1",
+  "optimizer_included": false,
+  "step": 21700,
+  "version": "v0",
+  "video_loops": 1,
+  "action_loops": 4,
+  "action_kv_mode": "aligned",
+  "parameters": 584536135,
+  "dataset_scope": "full_libero",
+  "epochs": 10,
+  "global_batch": 128,
+  "training_seed": 42,
+  "training_seconds": 39525.92356655188,
+  "evaluation_status": "not_run_training_only",
+  "success_rate": null,
+  "episodes": 0,
+  "tensor_equality_verified": true,
+  "source_revision": "3a89bfac5cfccad66bfebbe00014f7f739a78adc",
+  "compatible_code_revision": "3a89bfac5cfccad66bfebbe00014f7f739a78adc",
+  "normalization_sha256": "d5ad351eb3272f416d34082e8af5483ba1118ceca355a37d9f268fdb269fb3e4"
+}

libero-all-suites/v0-video1-action4/fairness.json

@@ -0,0 +1,52 @@
+{
+  "passed": true,
+  "compared_fields": [
+    "asset_sha256",
+    "seed",
+    "epochs",
+    "global_batch",
+    "train_windows",
+    "planned_updates",
+    "planned_windows",
+    "policy_dtype",
+    "optimizer_state_dtype",
+    "compute_dtype",
+    "initialization_mode"
+  ],
+  "data_fields": [
+    "action_offsets",
+    "available_episodes",
+    "available_windows",
+    "camera_order",
+    "codebase_version",
+    "complete_500_demo_set",
+    "content_files",
+    "coverage",
+    "dataset_dir",
+    "dataset_scope",
+    "episode_metadata_sha256",
+    "fps",
+    "info_sha256",
+    "metadata_files",
+    "normalization_scope",
+    "normalization_sha256",
+    "normalization_source",
+    "padding_policy",
+    "seed",
+    "split",
+    "split_algorithm",
+    "suites",
+    "task_counts",
+    "text_cache_dir",
+    "text_cache_files",
+    "text_encoder_provenance",
+    "train_episodes",
+    "train_windows",
+    "validation_episodes",
+    "validation_windows",
+    "video_offsets"
+  ],
+  "ignored_data_fields": [
+    "normalization_path"
+  ]
+}

libero-all-suites/v0-video1-action4/policy.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:801922e30df320cd7d9f3c642d90315865d49807ec20b6e27ea93c219d2f9a19
+size 2338635000

libero-all-suites/v0-video1-action4/training_timing.json

@@ -0,0 +1,45 @@
+{
+  "status": "complete",
+  "completed_updates": 21700,
+  "windows_seen": 2777130,
+  "planned_updates": 21700,
+  "elapsed_training_seconds": 39525.92356655188,
+  "checkpoint_seconds_total": 77.87684620916843,
+  "checkpoint_count": 10,
+  "measured_mean_update_seconds": 1.8147799186571616,
+  "estimated_total_training_hours": 10.939090065239002,
+  "estimate_note": "Projection from measured updates; excludes setup/checkpoint overhead",
+  "latest": {
+    "update": 21700,
+    "epoch": 10,
+    "windows": 81,
+    "windows_seen": 2777130,
+    "effective_passes": 10.0,
+    "seconds": 1.7924169301986694,
+    "lr": 1.0000000000000002e-06,
+    "grad_norm": 0.2220783531665802,
+    "branch_grad_after_clip": {
+      "video": 0.190041646361351,
+      "action": 0.11335917562246323,
+      "proprio": 0.018779480829834938
+    },
+    "latent_cache_rank_local": {
+      "hits": 1388640,
+      "misses": 0,
+      "dummy_encodes": 0,
+      "encoder_samples": 0,
+      "batches": 86790,
+      "read_seconds": 4171.644230587408,
+      "encode_seconds": 0.0,
+      "write_seconds": 0.0
+    },
+    "peak_allocated_gb": 53.634735107421875,
+    "peak_reserved_gb": 60.72303771972656,
+    "exit1/video_raw": 0.05690508712956935,
+    "exit1/action_raw": 0.04461105075883277,
+    "exit1/video": 0.06293870784618237,
+    "exit1/action": 0.021066600893750603,
+    "loss_video": 0.06293870784618237,
+    "loss_action": 0.021066600893750603
+  }
+}