the adapter trained on the v2 collection

Upload openpi/model-embed_tokens.safetensors with huggingface_hub

the same weights in openpi layout, for RL

pi0.5 LoRA adapter at 20k steps

initial commit