Skip to content

Kennyp-Chen/VLA_WAM_Practice

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

VLA & WAM 学习实践

从理论到实践,掌握 VLA (Vision-Language-Action) 和 WAM (World Action Model) 的训练全流程。

项目结构

VLA_WAM_Practice/
├── starVLA/                      # StarVLA 上游代码(需自行 clone,不在本仓库中)
├── VLA_WAM_Docs/                 # 学习文档
│   ├── Learning_Plan.md          # 主学习计划 & 进度追踪(← 从这里开始)
│   ├── Repo_Overview.md          # StarVLA 仓库速览 + VLA 算法简介
│   ├── OFT_Study.md              # OFT 深入学习文档
│   └── Papers_Summary/           # 论文研读
│       └── OpenVLA_OFT_Papers.md # OpenVLA & OpenVLA-OFT 论文研读
├── README.md                     # 本文
└── .gitignore

当前进度

Phase 1(VLA 入门)进行中 — 已完成环境搭建、模型和数据下载、Smoke Test。

详见 VLA_WAM_Docs/Learning_Plan.mdPhase 1 章节。

硬件说明

本项目的初始开发环境是 NVIDIA RTX 5090 (32GB, Blackwell),操作系统 Linux。
其他设备克隆后请根据自己硬件调整:

  • GPU 非 5090 → 可以尝试安装 flash-attn(见 StarVLA 官方文档)
  • GPU 显存 < 24GB → 调小 per_device_batch_size
  • CUDA 版本不同 → 安装对应 PyTorch 版本

数据 & 模型

以下文件在开发机上已下载完成,因体积较大不纳入版本控制:

内容 路径(相对 starVLA/ 大小
Qwen3-VL-2B-Instruct playground/Pretrained_models/Qwen3-VL-2B-Instruct/ ~4GB
LIBERO 数据集 playground/Datasets/LIBERO/libero/ ~3.6GB
数据集软链 playground/Datasets/LEROBOT_LIBERO_DATA → LIBERO/libero -

其他机器需按 StarVLA 官方指南自行下载。

快速启动(仅做参考,按原项目 README 为准)

# 1. Clone
git clone https://github.com/starVLA/starVLA.git
cd starVLA
git checkout starVLA_dev  # 对齐此项目的开发分支

# 2. Conda 环境
conda create -n starvla python=3.10 -y
conda activate starvla
pip install -r requirements.txt
pip install -e .

# 3. 下载模型 & 数据(huggingface-cli / git LFS)
# 4. 运行 smoke test
python starVLA/model/framework/VLM4A/QwenOFT.py

学习文档

About

VLA & WAM 学习实践 - StarVLA-OFT入门 → WM4A → LaWAM → RL

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors