pi05_simvla_post_train_task1_real_200

SimVLA pi05 post-training on the 200-episode real-robot task1 dataset, initialised from the SimVLA pretrain checkpoint at step 60000.

All training checkpoints are included as step subfolders: 10000, 20000, 30000, 40000, 50000. Each holds params/ and assets/ only (optimizer train_state/ excluded), so they are for inference and evaluation, not for resuming training.

from openpi.shared import download
path = download.maybe_download("hf://exaFLOPs09/pi05_simvla_post_train_task1_real_200/50000")