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

fine-tuning-openvla-oft

Fine-tune and evaluate OpenVLA-OFT policies for robot action generation.

Updated Apr 11, 2026
Search Tags:#fine-tuning#LoRA#robotics#FiLM#LIBERO#OpenVLA-OFT#ALOHA
One-click install
npx skills add https://github.com/hhhi21g/HealthCenter --skill fine-tuning-openvla-oft-hhhi21g
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: fine-tuning-openvla-oft
Source: https://github.com/hhhi21g/HealthCenter/tree/main/.codex/skills/openvla-oft
Command: npx skills add https://github.com/hhhi21g/HealthCenter --skill fine-tuning-openvla-oft-hhhi21g

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires torch, transformers, peft, draccus, accelerate, wandb, fastapi, uvicorn, tensorflow, robosuite, and includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve?

This Skill addresses the need for fine-tuning and evaluating OpenVLA-OFT and OpenVLA-OFT+ policies for robot action generation, providing a comprehensive toolset for reproducing research results and customizing action heads.

Core Features & Use Cases

  • Fine-Tuning: Customize and fine-tune OpenVLA-OFT and OpenVLA-OFT+ policies for specific tasks.
  • Evaluation: Evaluate fine-tuned models on LIBERO simulation and ALOHA real-world setups.
  • Use Case: Researchers and developers can use this Skill to reproduce the results from the OpenVLA-OFT paper, train custom VLA action heads, or deploy server-client inference for ALOHA.

Quick Start

Clone the public repo, follow the official setup, and evaluate a pretrained LIBERO checkpoint with the following command:

git clone https://github.com/moojink/openvla-oft.git
cd openvla-oft
python experiments/robot/libero/run_libero_eval.py \
  --pretrained_checkpoint moojink/openvla-7b-oft-finetuned-libero-spatial \
  --task_suite_name libero_spatial \
  --center_crop True \
  --num_trials_per_task 50 \
  --seed 7

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 for custom robot action generation tasks?▼

You can fine-tune OpenVLA-OFT policies for robot action generation by applying LoRA adaptation and FiLM conditioning to continuous action heads. The process requires PyTorch, Transformers, PEFT, and FlashAttention to train custom models on your specific robotic tasks.

What is OpenVLA-OFT and how does FiLM conditioning improve robot policy fine-tuning?▼

OpenVLA-OFT is a vision-language-action model for robot action generation that uses FiLM conditioning and continuous action heads. FiLM conditioning improves fine-tuning by allowing the model to adapt action outputs continuously based on visual and language inputs, enabling more precise robotic control.

Can I evaluate fine-tuned OpenVLA-OFT models on both LIBERO and ALOHA simulation environments?▼

Yes, you can evaluate fine-tuned OpenVLA-OFT models on both LIBERO simulation and ALOHA environments. The Skill supports evaluating pretrained LIBERO checkpoints and deploying server-client inference specifically designed for ALOHA real-world robotic setups.

Do I need TensorFlow and PyTorch to run OpenVLA-OFT fine-tuning and evaluation?▼

Yes, running OpenVLA-OFT fine-tuning and evaluation requires both PyTorch and TensorFlow. PyTorch, Transformers, and PEFT handle model training and LoRA adaptation, while TensorFlow is specifically required for running evaluation processes across LIBERO and ALOHA environments.

What's the best way to reproduce OpenVLA-OFT research results on the LIBERO benchmark?▼

The best way to reproduce OpenVLA-OFT research results on the LIBERO benchmark is to clone the official repository, complete the environment setup, and run the provided evaluation script with a pretrained checkpoint like openvla-7b-oft-finetuned-libero-spatial using specified task suite and trial parameters.

Does OpenVLA-OFT support LoRA adaptation for customizing continuous action heads?▼

Yes, OpenVLA-OFT supports LoRA adaptation for customizing continuous action heads. This allows researchers and developers to efficiently fine-tune the vision-language-action model for specific robotic tasks while updating only a small subset of model parameters.

Related Skills

Explore More Software Engineering Skills→
jonnabiojonnabio
6

fine-tuning-openvla-oft

Fine-tunes and evaluates OpenVLA-OFT robot action generation policies on LIBERO and ALOHA environments.

Community
Advanced
qcmuuqcmuu

fine-tuning-openvla-oft

Fine-tune and evaluate OpenVLA-OFT policies for robot action generation in LIBERO and ALOHA.

Community
Advanced
KappTech88KappTech88

fine-tuning-openvla-oft

Fine-tune and evaluate OpenVLA-OFT models with LoRA and FiLM conditioning.

Community
Advanced
tadod12tadod12

fine-tuning-openvla-oft

Fine-tune and evaluate OpenVLA-OFT policies with LoRA adaptation.

Community
Advanced