Fine Tuning Serving Openpi

If you're working with Physical Intelligence's OpenPI models for robot manipulation, this handles the full pipeline from blank machine setup through fine-tuning to serving. It covers both JAX and PyTorch training paths, checkpoint conversion between frameworks, and spins up policy inference servers for pi0, pi0-fast, and pi0.5. The quick start gets you serving a pretrained policy in four commands, which is about as friction-free as robotics tooling gets. Useful if you're prototyping manipulation tasks and don't want to spend days wiring up the training infrastructure yourself. The repository integration is straightforward, though you'll need comfort with both JAX and PyTorch ecosystems depending on your path.

skills.sh installs: 925
GitHub stars: 14.4K
last updated: Aug 21, 2026

View on GitHub

Install

$ npx -y skills add orchestra-research/ai-research-skills --skill fine-tuning-serving-openpi --agent claude-code

Installs into .claude/skills of the current project.

Files

OpenPI Fine-Tuning and Serving

End-to-end workflows for fine-tuning and serving Physical Intelligence's OpenPI models (pi0, pi0-fast, pi0.5) on robot manipulation tasks from the public openpi repository. Covers blank-machine setup, JAX training, PyTorch training, checkpoint conversion, and policy inference serving.

Quick start

Clone the public repo, install the workspace, then serve a pretrained policy:

git clone --recurse-submodules https://github.com/Physical-Intelligence/openpi.git
cd openpi
GIT_LFS_SKIP_SMUDGE=1 uv sync
GIT_LFS_SKIP_SMUDGE=1 uv pip install -e .
uv run scripts/serve_policy.py --env DROID

from openpi_client import websocket_client_policy

client = websocket_client_policy.WebsocketClientPolicy(host="localhost", port=8000)
result = client.infer(observation)
actions = result["actions"]  # numpy array of shape (chunk_size, action_dim)