Skip to content

Omni如何进行分布式训练? #210

@zlf0307

Description

@zlf0307

我使用如下命令进行多卡分布式训练:
CUDA_VISIBLE_DEVICES=5,6 python -m torch.distributed.run
main.py
--data_root ./text_spotting_datasets/
--output_folder ./output/pretrain/stage1/
--train_dataset totaltext_train mlt_train ic13_train ic15_train syntext1_train syntext2_train
--lr 0.0005
--max_steps 400000
--warmup_steps 5000
--checkpoint_freq 10000
--batch_size 6
--tfm_pre_norm
--train_max_size 768
--rec_loss_weight 2
--use_fpn
--use_char_window_prompt
但是实际上只有5号卡在训练,6号卡没有显存占用

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions