Enable model download statistics

Add a checkpoint catalog with published weight paths, sizes and SHA256 digests, and include the standard config.yaml entry point in the documented downloads.

Ego4WAM_VideoPretrained/README.md

@@ -42,7 +42,7 @@ from starVLA.model.framework.base_framework import baseframework
 variant = "Ego4WAM_VideoPretrained"
 root = snapshot_download(
     "HorizonRobotics/Ego4WAM",
-    allow_patterns=[f"{variant}/**"],
+    allow_patterns=["config.yaml", f"{variant}/**"],
 )
 model_dir = f"{root}/{variant}"
 model = baseframework.from_pretrained(f"{model_dir}/model.safetensors")

Ego4WAM_VideoPretrained_RoboDojoJoint/README.md

@@ -41,7 +41,7 @@ from starVLA.model.framework.base_framework import baseframework
 variant = "Ego4WAM_VideoPretrained_RoboDojoJoint"
 root = snapshot_download(
     "HorizonRobotics/Ego4WAM",
-    allow_patterns=[f"{variant}/**"],
+    allow_patterns=["config.yaml", f"{variant}/**"],
 )
 model = baseframework.from_pretrained(f"{root}/{variant}/model.safetensors")
 ```

README.md

@@ -43,6 +43,7 @@ RoboDojo Base weights are not included in this release.
 ```text
 Ego4WAM/
 ├── README.md
+├── config.yaml
 ├── LICENSE
 ├── .gitattributes
 ├── Ego4WAM_VideoPretrained/
@@ -63,7 +64,9 @@ by the code release, not by an internal experiment manifest.
 
 ## Download and load a single model
 
-Use `allow_patterns` to avoid downloading both large checkpoints:
+The root `config.yaml` indexes the released variants, their runtime configuration
+paths, weight sizes and SHA256 digests. Include it when downloading one variant
+with `allow_patterns` to keep the release index alongside the selected bundle:
 
 ```python
 from huggingface_hub import snapshot_download
@@ -72,7 +75,7 @@ from starVLA.model.framework.base_framework import baseframework
 variant = "Ego4WAM_VideoPretrained_RoboDojoJoint"
 root = snapshot_download(
     "HorizonRobotics/Ego4WAM",
-    allow_patterns=[f"{variant}/**"],
+    allow_patterns=["config.yaml", f"{variant}/**"],
 )
 model_dir = f"{root}/{variant}"
 model = baseframework.from_pretrained(f"{model_dir}/model.safetensors")

config.yaml

@@ -0,0 +1,21 @@
+# Released checkpoints, with repository-relative paths and SHA256 digests.
+# Use the model-specific project loader and runtime configuration.
+catalog_version: 1
+repo_id: HorizonRobotics/Ego4WAM
+checkpoints:
+  Ego4WAM_VideoPretrained:
+    directory: Ego4WAM_VideoPretrained
+    runtime_config: Ego4WAM_VideoPretrained/inference_config.yaml
+    model_assets: Ego4WAM_VideoPretrained/model_assets/
+    weights:
+    - path: Ego4WAM_VideoPretrained/model.safetensors
+      size_bytes: 26913478024
+      sha256: c5ca7480e9fbf504704e2e5bc3d9aba0e415838ef50bc993213e810a18b99f05
+  Ego4WAM_VideoPretrained_RoboDojoJoint:
+    directory: Ego4WAM_VideoPretrained_RoboDojoJoint
+    runtime_config: Ego4WAM_VideoPretrained_RoboDojoJoint/inference_config.yaml
+    model_assets: Ego4WAM_VideoPretrained_RoboDojoJoint/model_assets/
+    weights:
+    - path: Ego4WAM_VideoPretrained_RoboDojoJoint/model.safetensors
+      size_bytes: 26913478024
+      sha256: ae1b4595e1909e4573ad3214e1c5112f1920a34675340f6adfd4f2e9ec6c12c4