Upload tinyvla2/scripts/train.py with huggingface_hub

train.py: tokenize robot text for vlm_native too (enables all-channels v4: demos-in-LM + numeric morph + text)

Upload folder using huggingface_hub