Upload README.md with huggingface_hub

README.md

@@ -7,6 +7,8 @@ tags:
   - lerobot
   - libero
   - manipulation
+datasets:
+  - HuggingFaceVLA/libero
 ---
 
 # SmolVLA Fine-tuned on LIBERO-Spatial
@@ -14,6 +16,14 @@ tags:
 This is a fine-tuned version of [lerobot/smolvla_base](https://huggingface.co/lerobot/smolvla_base) 
 trained on the LIBERO-Spatial benchmark using the LeRobot framework.
 
+## Demo Video
+
+Task 8 success episode (70% success rate on this task):
+
+<video controls>
+  <source src="https://huggingface.co/Beeface/smolvla-libero-spatial/resolve/main/videos/task_8_success.mp4" type="video/mp4">
+</video>
+
 ## Model Details
 
 - **Base model:** lerobot/smolvla_base
@@ -39,9 +49,7 @@ trained on the LIBERO-Spatial benchmark using the LeRobot framework.
 | task_9 | 30% |
 | **Overall** | **36%** |
 
-## Training
-
-Fine-tuned using the LeRobot training pipeline:
+## Training Command
 ```bash
 lerobot-train \
   --policy.type=smolvla \
@@ -52,6 +60,20 @@ lerobot-train \
   --seed=42
 ```
 
+## Ablation Study — Training Duration
+
+We evaluated checkpoints at multiple steps to understand convergence:
+
+| Training Steps | Success Rate |
+|---------------|-------------|
+| 2,000 | 2% |
+| 6,000 | 17% |
+| 10,000 | 31% |
+| 20,000 | 36% |
+
+Performance improves consistently but with diminishing returns, suggesting 
+convergence begins around 10K steps on LIBERO-Spatial.
+
 ## Framework
 
 - [LeRobot](https://github.com/huggingface/lerobot)