fine-tuning-openvla-oft

Automate fine-tuning and evaluation of OpenVLA-OFT policies with LoRA adaptation.

Updated Apr 3, 2026
One-click install
npx skills add https://github.com/handsomelong922/my-codex-skills --skill fine-tuning-openvla-oft-handsomelong922
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: fine-tuning-openvla-oft
Source: https://github.com/handsomelong922/my-codex-skills/tree/main/skills/openvla-oft
Command: npx skills add https://github.com/handsomelong922/my-codex-skills --skill fine-tuning-openvla-oft-handsomelong922

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

OpenVLA-OFT fine-tunes and evaluates robot action generation policies with LoRA adaptation and FiLM conditioning, enabling reproducible workflows across LIBERO and ALOHA setups.

Core Features & Use Cases

  • Fine-tuning OpenVLA-OFT and OFT+ policies with LoRA adaptation on VLA backbones.
  • Supporting LIBERO simulation and ALOHA real-world pipelines for reproducible results.
  • Debugging and deployment workflows including server-client inference and cross-GPU consistency.

Quick Start

Run end-to-end fine-tuning and evaluation of OpenVLA-OFT on LIBERO/ALOHA to reproduce paper results.

Frequently Asked Questions about fine-tuning-openvla-oft

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

FAQPage Schema
How do I fine-tune OpenVLA-OFT policies with LoRA adaptation for robotics action generation?

This Skill supports ALOHA real-world pipelines alongside LIBERO simulation workflows for OpenVLA-OFT fine-tuning. You can reproduce paper results and deploy trained policies using server-client inference for robotic action generation tasks.

What PyTorch and Transformers versions do I need to run OpenVLA-OFT fine-tuning?

During OpenVLA-OFT fine-tuning, you must ensure training and inference parity for parameters like use_l1_regression, use_diffusion, use_film, num_images_in_input, use_proprio, and lora_rank. This Skill automates parity checks to prevent mismatches and supports cross-GPU debugging.

Can I train custom action heads with OpenVLA-OFT using multi-GPU setups?

Yes, OpenVLA-OFT fine-tuning supports multi-GPU training for custom action heads. The Skill applies LoRA adaptation and FiLM conditioning to the VLA backbone, enabling scalable training across GPUs for both LIBERO and ALOHA workflows.

What is the difference between OpenVLA-OFT and OFT+ policies when fine-tuning robot action generation?

OpenVLA-OFT and OFT+ are both robot action generation policies supported by this Skill for fine-tuning with LoRA adaptation. They enable reproducible results across LIBERO simulation and ALOHA real-world pipelines, with OFT+ extending the base OFT approach for enhanced action generation.

Why are my OpenVLA-OFT training and inference results inconsistent across different GPUs?

Inconsistencies often stem from training and inference parity mismatches in parameters like use_l1_regression, use_diffusion, use_film, num_images_in_input, use_proprio, and lora_rank. This Skill automates cross-GPU debugging and parity checks to resolve these alignment issues.