Fix NaN state input on gr.Examples click; use gr.JSON output with rounded values

Pin transformers==4.57.6 (matches upstream LabVLA's exact tested env); unbounded >=4.57.0 was resolving to transformers 5.x, which changed Qwen3VLVisionModel.forward() to return a 3-field ModelOutput instead of a 2-tuple, breaking image_embs, _ = self.vlm.model.visual(...) unpacking

Fix Qwen3VL attribute path: self.vlm.visual/.language_model -> self.vlm.model.visual/.language_model (transformers 4.57.6+ nests these under .model)

Upload app.py with huggingface_hub

Upload requirements.txt with huggingface_hub

Upload app.py with huggingface_hub

Upload folder using huggingface_hub

initial commit