lerobot_train_model_SmolVLAmulti_gpu_wandb.sh
Humanoid/fmc3-robotics-main/projects/lerobot/demo_scripts/lerobot_train_model_SmolVLAmulti_gpu_wandb.sh
accelerate launch \
--multi_gpu \
--num_processes=1 \
$(which lerobot-train) \
--dataset.repo_id=${HF_USER}/my_eval_smolvla_single_cam \
--policy.path=lerobot/smolvla_base \
--policy.repo_id=${HF_USER}/lerobot_pick_and_place_smolvla \
--output_dir=outputs/train/act_multi_gpu/lerobot_model \
--job_name=pick_and_place_bottle_from_A_to_B \
--policy.device=cuda \
--batch_size=8 \
--steps=50000 \
--optimizer.lr=2e-4 \
--wandb.enable=true
相关文章
run.py
run.py — python source code from the Humanoid learning materials (Humanoid/May_12/RoboOS/deploy/run.py).
阅读文章 →utils.py
utils.py — python source code from the Humanoid learning materials (Humanoid/May_12/RoboOS/deploy/utils.py).
阅读文章 →agent.py
agent.py — python source code from the Humanoid learning materials (Humanoid/May_12/RoboOS/master/agents/agent.py).
阅读文章 →planner.py
planner.py — python source code from the Humanoid learning materials (Humanoid/May_12/RoboOS/master/agents/planner.py).
阅读文章 →prompts.py
prompts.py — python source code from the Humanoid learning materials (Humanoid/May_12/RoboOS/master/agents/prompts.py).
阅读文章 →run.py
run.py — python source code from the Humanoid learning materials (Humanoid/May_12/RoboOS/master/run.py).
阅读文章 →