yyshi0619/pi05-base-pytorch

π0.5 base — PyTorch weights (format conversion of OpenPI pi05_base)

This repository contains the π0.5 base vision-language-action model by Physical Intelligence, converted from the original JAX checkpoint to a PyTorch safetensors file for OpenPI's PI0Pytorch model.

The weights are not retrained, fine-tuned or otherwise modified — this is a format conversion only.

Provenance

Original model	π0.5 base, Physical Intelligence — openpi
Original checkpoint	gs://openpi-assets/checkpoints/pi05_base/params
Converter	OpenPI examples/convert_jax_model_to_pytorch.py, --precision float32
Architecture	PaliGemma (Gemma 2B + SigLIP) with a Gemma-300M action expert, pi05=True
Action horizon / dim (config)	50 / 32
model.safetensors SHA256	f722b8d13a9c70ee2fe7451568fce1ded46d28d30f0fb7b89b801237514d85b1

Files

• model.safetensors — all model weights (float32, about 14.5 GB)
• config.json — the converter's model settings

Usage

Requires OpenPI's PyTorch model code, including its patched transformers files (see the OpenPI README, "PyTorch support").

import safetensors.torch
from openpi.models.pi0_config import Pi0Config
from openpi.models_pytorch.pi0_pytorch import PI0Pytorch

model = PI0Pytorch(Pi0Config(pi05=True, action_horizon=50, action_dim=32))
safetensors.torch.load_model(model, "model.safetensors", strict=True)

Verify the download with sha256sum model.safetensors against the hash above.

License and terms

Gemma is provided under and subject to the Gemma Terms of Use found at ai.google.dev/gemma/terms. This model is derived from PaliGemma/Gemma; any use must comply with the Gemma Terms of Use and the Gemma Prohibited Use Policy. The OpenPI code is released under the Apache License 2.0.

All credit for the model belongs to Physical Intelligence and the OpenPI authors; please cite their work if you use these weights.