pi05-base-pytorch

pi05-base-pytorch

PyTorch conversion of Physical Intelligence's official pi05_base checkpoint (gs://openpi-assets/checkpoints/pi05_base), converted via openpi examples/convert_jax_model_to_pytorch.py with --config_name pi05_libero (openpi has no standalone pi05_base config; the π₀.₅ architecture is identical), precision float32. Uploaded for RobotTrain subnet validator pipeline testing — serves as a not-fine-tuned negative control (zero-score anchor) for benchmark discrimination tests.

Note: the base checkpoint ships no norm_stats of its own; assets/physical-intelligence/libero/norm_stats.json from the official pi05_libero checkpoint was added so the model can be served with the pi05_libero config (the validator's submission convention).

Important: the LIBERO score of this repo does NOT measure the original checkpoint's capability. The base model was never fine-tuned on LIBERO and is not bound to any normalization statistics; the LIBERO norm_stats were added purely to satisfy the serving convention. Treat it strictly as a format-valid negative control for validator pipeline testing — never as an evaluation of the underlying model.

Layout: model.safetensors (fp32) + assets/physical-intelligence/libero/norm_stats.json. Load with openpi: create_trained_policy(get_config("pi05_libero"), <this dir>).