joon-stack/univla-lerobot-banpot-hfrad-step5000
Full merged UniVLA checkpoint for LeRobot cache fine-tuning.
- dataset:
put_banana_in_pot_merge
- model:
hyperbolic_factorized_radprog
- base VLA:
hf_univla_bridge_factorized_lam30k_rad16dir16_50k_step050000
- step:
5000
- latent action token length:
5
- codebook size:
32
- predicted action horizon:
10
This repository contains full merged VLA weights and the matching action_decoder-5000.pt.
