RoboBrain-X0-FlagOS

Quick Start

Pull Docker Image

docker pull harbor.baai.ac.cn/flagrelease-public/robobrain_x0_flagscale

You can use cuda12.4.1-cudnn9.5.0-python3.12-torch2.6.0-time250928-ssh tag for now.

Download Model Ckpts

You should download RoboBrain-X0's weights to /share/

Run Container

docker run -itd --name robotics_pretrain --privileged --gpus all --net=host --ipc=host --device=/dev/infiniband --shm-size 512g --ulimit memlock=-1 -v /share/:/models harbor.baai.ac.cn/flagrelease-public/robobrain_x0_flagscale

Train

cd /root/robotics_pretrain/flag-scale
conda activate flagscale-train
python run.py --config-path ./examples/qwen2_5_vl/conf --config-name train_3b_action_S6_subtask_agilex_eval5_demo action=run

Serve & Inference

cd /root/robotics_pretrain/flag-scale
conda activate flagscale-inference
python run.py --config-path ./examples/qwen2_5_vl/conf/ --config-name serve_3b action=run

For more information

You can refer to the origin repo of RoboBrain-X0 https://github.com/FlagOpen/RoboBrain-X0.

The weight files and evaluation results will be released at a later date after the native model is officially launched.