skills.rest
hhhi21g
hhhi21gCOMMUNITY
Explore all skills by this author →
GitHub

fine-tuning-serving-openpi

Fine-tune and serve OpenPI models for robot policy inference.

Updated Apr 11, 2026
Search Tags:#fine-tuning#PyTorch#robotics#Physical Intelligence#policy serving#JAX#OpenPI
One-click install
npx skills add https://github.com/hhhi21g/HealthCenter --skill fine-tuning-serving-openpi-hhhi21g
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: fine-tuning-serving-openpi
Source: https://github.com/hhhi21g/HealthCenter/tree/main/.codex/skills/openpi
Command: npx skills add https://github.com/hhhi21g/HealthCenter --skill fine-tuning-serving-openpi-hhhi21g

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires uv>=0.4.0, jax>=0.4.30, torch>=2.1.0, transformers>=4.53.2, and includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve?

This Skill solves the problem of fine-tuning and serving Physical Intelligence's OpenPI models, enabling robot policy inference across various environments.

Core Features & Use Cases

  • Fine-Tuning: Adapt pi0, pi0-fast, and pi0.5 models to custom datasets using JAX or PyTorch.
  • Checkpoint Conversion: Convert JAX checkpoints to PyTorch for deployment.
  • Policy Serving: Run policy inference servers for real-time control.
  • Use Case: Use this Skill to fine-tune an OpenPI model on a custom dataset and serve it for a robot's policy inference in an ALOHA, DROID, or LIBERO environment.

Quick Start

Use the fine-tuning-serving-openpi skill to fine-tune a pi0.5 model on your dataset and serve it using:

uv run scripts/train.py <config_name> --exp-name=<run_name>
uv run scripts/serve_policy.py --env ALOHA

Frequently Asked Questions about fine-tuning-serving-openpi

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I fine-tune Physical Intelligence OpenPI models for robot policy inference?▼

You can fine-tune OpenPI models like pi0, pi0-fast, and pi0.5 for robot policy inference by adapting them to custom datasets using JAX or PyTorch, followed by serving the policy for real-time control.

What's the best way to convert JAX checkpoints to PyTorch for OpenPI deployment?▼

Converting JAX checkpoints to PyTorch for OpenPI deployment is supported directly within the workflow, allowing you to transition from fine-tuning in JAX to running a PyTorch-based policy inference server.

Can I serve a fine-tuned OpenPI policy in an ALOHA or DROID environment?▼

Yes, you can serve a fine-tuned OpenPI policy in ALOHA, DROID, or LIBERO environments by running the policy inference server for real-time robot control.

Do I need JAX and PyTorch to run OpenPI policy serving?▼

Yes, running OpenPI policy serving and fine-tuning requires both JAX and PyTorch, along with the transformers library and the uv package manager to execute the training and serving scripts.

How does fine-tuning OpenPI models on custom datasets work?▼

Fine-tuning OpenPI models on custom datasets works by executing a training script with a specified configuration and experiment name, adapting the base pi0 or pi0.5 model to your specific robotic data.

Are there limitations when using OpenPI models for real-time robot control?▼

While OpenPI supports real-time policy inference across environments like ALOHA and LIBERO, limitations depend on your custom dataset quality and the successful conversion of JAX checkpoints to PyTorch for deployment.

Related Skills

Explore More Software Engineering Skills→
jonnabiojonnabio
6

fine-tuning-serving-openpi

Fine-tune and serve OpenPI robot policy models across ALOHA, DROID, and LIBERO.

Community
Advanced
KappTech88KappTech88

fine-tuning-serving-openpi

Fine-tune and serve OpenPI robot policy models for real-time control.

Community
Advanced
qcmuuqcmuu

fine-tuning-serving-openpi

Fine-tune and serve OpenPI robot policies for manipulation datasets.

Community
Advanced
tadod12tadod12

fine-tuning-serving-openpi

Fine-tune and serve OpenPI robot policies across ALOHA, DROID, and LIBERO environments.

Community
Advanced