Views
No views yet
1python scripts/gr00t_finetune.py \
2 --dataset-path ./demo_data/so101-Pickup \
3 --num-gpus 1 \
4 --output-dir ./so101-Pickup-checkpoints00 \
5 --max-steps 10000 \
6 --data-config so100_dualcam \
7 --video-backend decord \
8 --batch-size 4 \
9 --dataloader-num-workers 1 \
10 --dataloader-prefetch-factor 1./demo_data/so101-Pickup1./so101-Pickup-checkpoints0010000so100_dualcamdecord411如调整超参，请在此文件追加新的命令块与参数记录，确保可追溯。
conda activate gr00t1python scripts/eval_policy.py --plot \
2 --embodiment_tag new_embodiment \
3 --model_path ./dataset/pickup-boxs16-model/checkpoint-10000 \
4 --data_config so100_dualcam \
5 --dataset_path ./dataset/pickup-boxs16 \
6 --video_backend torchvision_av \
7 --modality_keys single_arm gripper./dataset/pickup-boxs16-model/eval_traj0.pngstate_joints vs timegt_action_joints vs time (150, 6)pred_action_joints vs time (150, 6)备注：日志包含后端加载与权重分片装载信息（略）。如需完整原始日志，请将终端输出另存为文件并在此处追加链接或引用。