Publish Phoenix-MWA evaluation resources (verified batch)

Explicit evaluation-only release; no training states, raw visual data or credentials.

README.md

@@ -0,0 +1,57 @@
+---
+library_name: pytorch
+tags:
+  - robotics
+  - phoenix-mwa
+  - libero
+---
+
+# Phoenix-LIBERO
+
+Phoenix-MWA LIBERO policy initialized from pretraining step **125,000**, then fine-tuned for **25,000** steps. This is a weights-only inference release for a local **Phoenix-MWA** checkout; it is not a full training-state resume checkpoint.
+
+The policy uses `phoenix.local_joint96.weights.v1` and must be paired with the DINO110k EMA and STVAE150k weights in [OpenPhoenix/Phoenix-Encoders](https://huggingface.co/OpenPhoenix/Phoenix-Encoders). The separate RoboCasa policy is [OpenPhoenix/Phoenix-RoboCasa](https://huggingface.co/OpenPhoenix/Phoenix-RoboCasa).
+
+## Files
+
+- `policy_phoenix_v1.pt`: Phoenix-format policy weights.
+- `train_config.yaml`: saved configuration for this policy.
+- `dataset_stats.json`: matching runtime normalization statistics.
+- `normalization_source.json` and `initialization.json`: normalization and initialization provenance.
+
+Historical absolute paths in the configuration and provenance records describe the original training environment. Use the explicit resource paths below and the generated runtime configuration on your machine. Do not substitute another policy's statistics or encoder weights.
+
+## Download and smoke evaluation
+
+Run from the root of your local Phoenix-MWA checkout after following its `docs/INSTALLATION.md` and `docs/RESOURCES.md`. Install `huggingface_hub` in that environment, then download:
+
+```python
+from huggingface_hub import snapshot_download
+
+snapshot_download("OpenPhoenix/Phoenix-LIBERO", local_dir="checkpoints/libero")
+snapshot_download("OpenPhoenix/Phoenix-Encoders", local_dir="checkpoints/encoders")
+```
+
+Acquire the matching Wan video/VAE, T5 text-encoder and tokenizer components separately under `checkpoints/base/`, following the layout expected by Phoenix-MWA. They are not included in these two repositories. The saved configuration selects [Wan2.2-TI2V-5B](https://huggingface.co/Wan-AI/Wan2.2-TI2V-5B) and the [Wan2.1-T2V-1.3B](https://huggingface.co/Wan-AI/Wan2.1-T2V-1.3B) tokenizer.
+
+Prepare official [LIBERO-Plus](https://github.com/sylvestf/LIBERO-plus) resources at `resources/LIBERO-plus/libero/libero/`, containing `assets/`, `bddl_files/` and `init_files/`. Then:
+
+```bash
+python tools/prepare_eval.py --benchmark libero-plus \
+  --checkpoint checkpoints/libero/policy_phoenix_v1.pt \
+  --config checkpoints/libero/train_config.yaml \
+  --stats checkpoints/libero/dataset_stats.json \
+  --dino checkpoints/encoders/dino110k.pt \
+  --stvae checkpoints/encoders/stvae150k.pt \
+  --model-base checkpoints/base \
+  --benchmark-root resources/LIBERO-plus/libero/libero \
+  --output resources/phoenix-libero-smoke
+EVAL_GPUS=0 bash scripts/evaluate.sh libero-plus \
+  resources/phoenix-libero-smoke/eval.env.sh smoke
+```
+
+Use a new `--output` directory for each run. Preparation reads the checkpoint and writes local runtime files; it does not launch inference. Smoke evaluates one selected task per suite and checks execution, not a full benchmark success rate. For LIBERO-Pro, follow the separate benchmark setup and commands in Phoenix-MWA's `docs/EVALUATION.md`, using these same policy/configuration/statistics/encoder files. PPU execution and PPU/NVIDIA numerical parity are not certified by this model card.
+
+## Licensing
+
+See [NOTICE.md](NOTICE.md) and the accompanying component license files. Phoenix-MWA's source-code license is not a blanket license for these weights, datasets, simulators or upstream models. The shared encoder repository includes the applicable DINOv3 terms. This card does not assign an MIT license to the policy weights.

release_manifest.json

@@ -0,0 +1,78 @@
+{
+  "schema": "phoenix.huggingface_public_release.v1",
+  "repository": "OpenPhoenix/Phoenix-LIBERO",
+  "code_project": "Phoenix-MWA",
+  "created_at": "2026-10-15T22:51:30.700186+08:00",
+  "files": [
+    {
+      "path": "LICENSES/Apache-2.0.txt",
+      "size": 11358,
+      "sha256": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30",
+      "git_blob_sha1": "d645695673349e3947e8e5ae42332d0ac3164cd7"
+    },
+    {
+      "path": "NOTICE.md",
+      "size": 1682,
+      "sha256": "35da38790c8331ea2d262d695e46c3454b199c7b2f952943feefef95ff91dfba",
+      "git_blob_sha1": "ec761e8f309588b6dfb1626ff370735b545e2223"
+    },
+    {
+      "path": "README.md",
+      "size": 3633,
+      "sha256": "cd3f24f025c824ba43839f1e2a27829e0cfbb2c0a3160a975c9ea0793e99b5a4",
+      "git_blob_sha1": "824c88a7b78406fb03356d2734c5d2b4c2cf2eee"
+    },
+    {
+      "path": "dataset_stats.json",
+      "size": 11671,
+      "sha256": "42cd2dd6ea82642068553bdcde1b97978d5011b7c9332e07377a3a2c84e549bc",
+      "git_blob_sha1": "4182b805159f8a6e2c303b809dfc264dc66ee807"
+    },
+    {
+      "path": "initialization.json",
+      "size": 9459,
+      "sha256": "1e5f173bcc497730d62fd638da2c237e5450904eb8f404d1b3a3902789c16c9d",
+      "git_blob_sha1": "1758990ce46b7eac2bf8ed9b34d55b6b5cfeb9bf"
+    },
+    {
+      "path": "normalization_source.json",
+      "size": 11627,
+      "sha256": "5cd894665cf8bc83dd33e7c73382396c4aa67cddcadde9b7d21aad0113f06a23",
+      "git_blob_sha1": "b4e81277fced2e26a5c9494d05f77b16dd2a6bf3"
+    },
+    {
+      "path": "policy_phoenix_v1.pt",
+      "size": 12045459165,
+      "sha256": "eb5a3e13b3696b4bfaf8c70ca46d10ebce263f2be444a86f010674823afb8a07"
+    },
+    {
+      "path": "train_config.yaml",
+      "size": 18131,
+      "sha256": "75c4361dcab7a3769e50afceb156306f85a2d93d2f4e1f3ffe59cb7d5b35de49",
+      "git_blob_sha1": "2c4869426cf7de8f5369501cc8c0ff501cf16c19"
+    }
+  ],
+  "excluded": [
+    "old_policy.pt",
+    "optimizer_state",
+    "scheduler_state",
+    "rng_state",
+    "pretraining_initialization_weights",
+    "raw_training_videos",
+    "visual_feature_caches",
+    "credentials",
+    "source_acceptance.json"
+  ],
+  "runtime_hash_checks_required": false,
+  "new_gpu_evaluation_performed_during_upload": false,
+  "checkpoint": {
+    "path": "policy_phoenix_v1.pt",
+    "schema": "phoenix.local_joint96.weights.v1",
+    "pretrain_step": 125000,
+    "posttrain_step": 25000,
+    "original_training_checkpoint_sha256": "28ad7bf9164877c54d61bd49ec65543ef636599104ad2cb44acca8224a4002fe",
+    "schema_migration_parent_sha256": "7d9e2ca4017e4c3259bfcf703477f7372de21a0c0248ffb1a06f16fbea26d845",
+    "published_sha256": "eb5a3e13b3696b4bfaf8c70ca46d10ebce263f2be444a86f010674823afb8a07",
+    "tensor_equivalence_recomputed_during_upload": false
+  }
+}