PI0.5 no-red-left LoRA checkpoints
OpenPI PI0.5 VLM-LoRA/full-action checkpoints trained on
ZhihanKang/panda-sort-three-no-red-left-v2.1.
The repository contains inference parameters and normalization assets for steps 20,000 and 30,000. Training optimizer state is intentionally excluded.
Training config: sort_three_no_red_left_vlm_lora_full_action_bs16.