Normalize Motius Pipeline.from_pretrained metadata

README.md

@@ -1,5 +1,5 @@
 ---
-library_name: hftrainer
+library_name: motius
 pipeline_tag: other
 tags:
 - motion-generation
@@ -13,10 +13,10 @@ license: other
 
 # MDM — Human Motion Diffusion Model
 
-Text-to-motion baseline integrated into the hftrainer Model Zoo. Our
+Text-to-motion baseline integrated into the motius Model Zoo. Our
 reproduction is **fully self-contained and independent of `ref_repo`**: the
 network, the Gaussian-diffusion schedule, the classifier-free-guidance sampler
-and the collate are all vendored into `hftrainer.models.motion.mdm._mdm`, and verified
+and the collate are all vendored into `motius.models.motion.mdm._mdm`, and verified
 to be **bit-identical** to the released checkpoint (`max-abs-diff = 0.0` for the
 same seed/input).
 
@@ -24,7 +24,7 @@ same seed/input).
 |---|---|
 | **Task** | Text-to-Motion (T2M) |
 | **Bundle / Pipeline** | `MDMBundle` / `MDMPipeline` |
-| **Processed HF artifact** | [`ZeyuLing/hftrainer-mdm-humanml3d`](https://huggingface.co/ZeyuLing/hftrainer-mdm-humanml3d) |
+| **Processed HF artifact** | [`ZeyuLing/Motius-MDM-HumanML3D`](https://huggingface.co/ZeyuLing/Motius-MDM-HumanML3D) |
 | **Motion representation** | **HumanML3D-263** (263-dim, 20 fps, 22 joints) |
 | **Text encoder** | CLIP ViT-B/32 (frozen) |
 | **Paper** | *Human Motion Diffusion Model*, Tevet et al., ICLR 2023 — [arXiv:2209.14916](https://arxiv.org/abs/2209.14916) |
@@ -34,26 +34,26 @@ same seed/input).
 
 ## Weights
 
-Current hftrainer artifact (diffusers-style `from_pretrained`):
+Current motius artifact (diffusers-style `from_pretrained`):
 
 | Artifact | Location | Contents | Status |
 |---|---|---|---|
-| MDM HumanML3D | [`ZeyuLing/hftrainer-mdm-humanml3d`](https://huggingface.co/ZeyuLing/hftrainer-mdm-humanml3d) | `model.safetensors` + `mdm_config.json` + `Mean.npy` / `Std.npy` | public Hub artifact; complete CLIP packaging pending |
+| MDM HumanML3D | [`ZeyuLing/Motius-MDM-HumanML3D`](https://huggingface.co/ZeyuLing/Motius-MDM-HumanML3D) | `model.safetensors` + `mdm_config.json` + `Mean.npy` / `Std.npy` | public Hub artifact; complete CLIP packaging pending |
 | local mirror | `checkpoints/mdm/humanml_trans_enc_512` | same layout | optional local cache |
 
 **Use directly from the Hub:**
 
 ```python
-from hftrainer.pipelines.mdm import MDMPipeline
+from motius.pipelines.mdm import MDMPipeline
 
-pipe = MDMPipeline.from_pretrained("ZeyuLing/hftrainer-mdm-humanml3d", device="cuda")
+pipe = MDMPipeline.from_pretrained("ZeyuLing/Motius-MDM-HumanML3D", device="cuda")
 motions = pipe.infer_t2m(["a person walks forward then sits down"], [120])  # list of (T, 263)
 ```
 
 **Or download to disk first:**
 
 ```bash
-huggingface-cli download ZeyuLing/hftrainer-mdm-humanml3d \
+huggingface-cli download ZeyuLing/Motius-MDM-HumanML3D \
     --local-dir checkpoints/mdm/humanml_trans_enc_512
 ```
 
@@ -82,7 +82,7 @@ CLIP ViT-B/32 is currently resolved by name rather than stored inside the repo.
 | `local_vel` | 66 | local joint velocities (22×3) |
 | `foot_contact` | 4 | binary foot-contact labels |
 
-Convert to/from other spaces with `hftrainer.motion.representation.convert`
+Convert to/from other spaces with `motius.motion.representation.convert`
 (e.g. `hml263_to_joints`, `hml263_to_motion135`, `hml263_to_motion272`).
 
 ---
@@ -90,7 +90,7 @@ Convert to/from other spaces with `hftrainer.motion.representation.convert`
 ## Evaluation
 
 Generation under the **official HumanML3D protocol** (standard test split, native
-263-dim @ 20 fps, first caption) and scoring with the two persisted hftrainer
+263-dim @ 20 fps, first caption) and scoring with the two persisted motius
 evaluators. Reproduce with:
 
 ```bash
@@ -103,7 +103,7 @@ python3 scripts/eval/verify_evaluators.py --which hml263 \
 
 ### HumanML3D-263 evaluator (native space, n=3970)
 
-| Metric | hftrainer | MDM paper | Note |
+| Metric | motius | MDM paper | Note |
 |---|---|---|---|
 | **FID** ↓ | **0.509** | 0.544 | ✅ reproduced (within noise) |
 | **Diversity** → | **9.563** | 9.559 | ✅ matches |
@@ -136,10 +136,18 @@ is the `263→272` representation mismatch.
 
 ## Implementation notes
 
-- **Vendored, ref_repo-independent**: `hftrainer/models/mdm/_mdm/` holds the
+- **Vendored, ref_repo-independent**: `motius/models/mdm/_mdm/` holds the
   network (`network.py`), diffusion (`diffusion/`), CFG sampler and collate.
   Training-only deps are stubbed (inference-only).
 - **Normalization travels with the checkpoint**: `Mean.npy` / `Std.npy` are the
   HumanML3D *training* stats (not the evaluator stats) and are embedded in the
   artifact, eliminating the recurring "wrong Mean/Std → forward drift" bug.
 - **Guidance**: classifier-free, default scale `2.5`.
+
+## Direct Loading
+
+```python
+from motius import Pipeline
+
+pipeline = Pipeline.from_pretrained("ZeyuLing/Motius-MDM-HumanML3D")
+```

model_index.json

@@ -0,0 +1,23 @@
+{
+  "_class_name": "MDMPipeline",
+  "_library_name": "motius",
+  "format_version": 1,
+  "pipeline_class": "motius.pipelines.mdm.MDMPipeline",
+  "bundle_class": "motius.models.mdm.MDMBundle",
+  "tasks": [
+    "text_to_motion"
+  ],
+  "required_files": [
+    "Mean.npy",
+    "Std.npy",
+    "mdm_config.json",
+    "model.safetensors"
+  ],
+  "api": {
+    "loader": "motius.Pipeline.from_pretrained",
+    "task_methods": [
+      "infer_text_to_motion"
+    ]
+  },
+  "artifact_format": "motius-mdm-v1"
+}