pi0-libero-finetuned

Full fine-tune of Physical Intelligence's pi0 base model (gs://openpi-assets/checkpoints/pi0_base) on the LIBERO benchmark (physical-intelligence/libero dataset), using the openpi pi0_libero training config.

Base checkpoint: pi0_base
Dataset: physical-intelligence/libero
Training steps: 30,000 (checkpoint at step 29,999)
Hardware: 2x NVIDIA H200
Final train loss: ~0.0109

This repo contains only the params/ (inference weights) and assets/ (norm stats) needed to run inference — the optimizer/train_state/ used to resume training is not included.

Usage

from openpi.training import config as _config
from openpi.policies import policy_config

cfg = _config.get_config("pi0_libero")
policy = policy_config.create_trained_policy(cfg, "<path to this checkpoint>")

Downloads last month: -; Downloads are not tracked for this model. How to track

Video Preview

Robotics