Training IDM model #9
Description
🐛 Describe the bug
Hello,
Thank you for your amazing jobs.
when I Train Dreams for my own embodiment, I got the belowing error:
File ".../GR00T-Dreams/gr00t/model/action_head/flow_matching_action_head_idm.py", line 122, in forward
return torch.bmm(x, selected_W) + selected_b.unsqueeze(1)
RuntimeError: expected scalar type BFloat16 but found Double
However, when I Train Gr1 datasets, I also got the same error,
Belowing is my running code:
CUDA_VISIBLE_DEVICES=0
WANDB_DISABLED=true PYTHONPATH=.
python scripts/idm_training.py
--dataset-path demo_data/robot_sim.PickNPlace
--data-config gr1_arms_only
--embodiment-tag gr1
--output-dir runs/idm_gr1_smoke
--batch-size 4
--max-steps 5
--num-gpus 1
--video-backend torchvision_av
--dataloader-num-workers 1
--report-to none