Skip to content

README.md

Latest commit

 

History

History
63 lines (46 loc) · 2.15 KB

README.md

File metadata and controls

63 lines (46 loc) · 2.15 KB
Raw

LeRobot GPU

This page is part of the sample applications guide. Follow README first and stop interactive workloads when finished.

(Optional) Create a docker image for LeRobot GPU by following the upstream LeRobot docker setup and using the pinned v0.4.4 source in this repository's Dockerfile.

docker build -f docker/lerobot-gpu/Dockerfile . -t j3soon/runai-lerobot-gpu:0.4.4
docker push j3soon/runai-lerobot-gpu:0.4.4

docker/lerobot-gpu/Dockerfile is a self-contained local variant adapted from the upstream docker/Dockerfile.internal at LeRobot v0.4.4.

Environment command:

/run.sh "uv pip install jupyterlab" "uv run jupyter lab --ip=0.0.0.0 --no-browser --allow-root --NotebookApp.base_url=/${RUNAI_PROJECT}/${RUNAI_JOB_NAME} --NotebookApp.token='' --notebook-dir=/"

Run In Jupyter Lab

Below is a simple training flow based on the LeRobot training guide.

The Dockerfile clones the LeRobot source tree into /lerobot during the image build.

cd /lerobot
git config --global --add safe.directory /lerobot
hf auth login
wandb login

If you already have a dataset on the Hugging Face Hub, set your username and start training:

HF_USER=$(NO_COLOR=1 hf auth whoami | awk -F': *' 'NR==1 {print $2}')

lerobot-train \
  --dataset.repo_id=${HF_USER}/so101_test \
  --policy.type=act \
  --output_dir=outputs/train/act_so101_test \
  --job_name=act_so101_test \
  --policy.device=cuda \
  --wandb.enable=true \
  --policy.repo_id=${HF_USER}/my_policy \
  --policy.private=true

Checkpoints are written to:

ls outputs/train/act_so101_test/checkpoints

To resume from the latest checkpoint:

lerobot-train \
  --config_path=outputs/train/act_so101_test/checkpoints/last/pretrained_model/train_config.json \
  --resume=true

If you need to create a dataset first, follow the upstream Imitation Learning on Real-World Robots guide for teleoperation, recording, replay, and evaluation steps.