ContextVLA

KO | EN

gitlite — search

#javascript #hacktoberfest #react #typescript #machine-learning #rust #deep-learning #linux #macos #java #awesome #cli #ios #open-source

(★ 20)

ContextVLA: Vision-Language-Action Model with Amortized Multi-Frame Context

ContextVLA 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

<h1 align="center"> ContextVLA: Vision-Language-Action Model with Amortized Multi-Frame Context </h1>
<div align="center">
  <a href="https://huiwon-jang.github.io/" target="_blank">Huiwon&nbsp;Jang</a><sup>1,2</sup>
  &ensp; <b>&middot;</b> &ensp;
  <a href="https://sihyun.me/" target="_blank">Sihyun&nbsp;Yu</a><sup>1</sup>
  &ensp; <b>&middot;</b> &ensp;
  <a href="https://arunos728.github.io/" target="_blank">Heeseung&nbsp;Kwon</a><sup>1,2</sup>
  <br>
  Hojin Jeon<sup>1</sup>
  &ensp; <b>&middot;</b> &ensp;
  <a href="https://younggyo.me/" target="_blank">Younggyo&nbsp;Seo</a><sup>3*</sup>
  &ensp; <b>&middot;</b> &ensp;
  <a href="https://alinlab.kaist.ac.kr/shin.html" target="_blank">Jinwoo&nbsp;Shin</a><sup>1,2*</sup>
  <br>
  <sup>1</sup> KAIST &emsp; <sup>2</sup>RLWRLD &emsp; <sup>3</sup>UC Berkeley &emsp; *Equal Advising<br>
</div>
<h3 align="center">[<a href="https://huiwon-jang.github.io/contextvla/">project page</a>] [<a href="https://arxiv.org/abs/2510.04246">arxiv</a>]</h3>

### 1. Environment Setup
```bash
conda create -n contextvla python=3.11 -y
pip install torch==2.7.1 torchvision==0.22.1 torchaudio==2.7.1 --index-url https://download.pytorch.org/whl/cu126
pip install -r requirements.txt
cd qwen_fast/packages/openpi-client && pip install -e . && cd ../../..
```

### 2. OXE Pre-trained Checkpoints
We provide checkpoints of ContextVLA pretrained on <a href="https://github.com/dusty-nv/openvla/blob/main/prismatic/vla/datasets/rlds/oxe/mixtures.py" target="_blank">OXE Magic Soup</a> in the below. If the link is blocked, please contact me.
- <a href="https://huggingface.co/huiwon/ContextVLA-3B-Qwen2.5VL-FAST" target="_blank">ContextVLA-3B-Qwen2.5VL-FAST</a> (180K iter pretrained with batch size of 128)

### 3. Fine-tuning
E.g., Fine-tuning ContextVLA on Libero dataset

(1) `qwen_fast/src/training/config.py`
 - Set `_CONFIGS`: TrainConfig.data.repo_id (e.g., "physical-intelligence/libero")
 - See <a href="https://github.com/Physical-Intelligence/openpi/tree/main?tab=readme-ov-file#fine-tuning-base-models-on-your-own-data" target="_blank">Physical-intelligence/openpi</a> for more details.

(2) run training script

```bash
cd qwen_fast
bash train_scripts/train_contextvla_libero.sh
```

### 4. Evaluation
TODO

### 5. Fine-tuning Pi0 / Pi0-FAST / GR00T N1.5 using ContextVLA
TODO

### Acknowledgement
This code is mainly built upon <a href="https://github.com/Physical-Intelligence/openpi/tree/main" target="_blank">Physical-intelligence/openpi</a>. We also appreciate <a href="https://github.com/declare-lab/nora" target="_blank">NORA</a> for providing pre-training code implementation using OXE Magic Soup dataset.

ContextVLA

ContextVLA 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

ContextVLA

ContextVLA 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README