|
Download GRADIO_NOTES.md from DynamicIntelligence/dynamic_intelligence_sample_data: direct link, hf CLI and curl.
- Browser
- Download file 976 Bytes
-
https://huggingface.co/spaces/DynamicIntelligence/dynamic_intelligence_sample_data/resolve/main/GRADIO_NOTES.md
- Command line
-
hf download hf://spaces/DynamicIntelligence/dynamic_intelligence_sample_data/GRADIO_NOTES.md
-
curl -L -o GRADIO_NOTES.md https://huggingface.co/spaces/DynamicIntelligence/dynamic_intelligence_sample_data/resolve/main/GRADIO_NOTES.md
976 Bytes
| # Gradio Deployment Notes | |
| - `app.py` defines `build_interface()` and **only** launches the demo when run as `__main__`. Hugging Face imports the module to build the config, so expensive work (dataset download) happens lazily via `lru_cache` helpers. | |
| - Dataset paths follow the LeRobot layout (`data/chunk-000/...`, `videos/chunk-000/rgb/...`). Files are cached inside `dataset_cache/` by `snapshot_download`. | |
| - Default episode data is preloaded when the Blocks tree is constructed, so all widgets render immediately without calling the API. | |
| - User interactions run through `episode_dropdown.change` → `load_episode_payload`, which returns `[instruction, video_path, fig_x, ...]`. | |
| - `demo.queue().launch(show_api=False)` disables the “Use via API” panel (Gradio 5.7.0 had a schema bug with Plot outputs that broke API info). | |
| - If you need to expose the API again, make sure to upgrade to a Gradio release where plot schema serialization works, then set `show_api=True`. | |