Regarding multi-machine multi-GPU training for FLUX kontext

[wangjiyuan9]

Hello, I have carefully reviewed the previous issues and found that you don't seem to plan to support single-card multi-batch training for various models. But, may I ask, can I simply replace:

accelerate launch \ examples/flux/model_training/train.py \

with

accelerate launch \ --multi_gpu \ --num_processes 3 \ --gpu_ids 5,6,7 \ --mixed_precision bf16 \ examples/flux/model_training/train.py \

to achieve multi-GPU training, and will subsequent multi-machine training also be done by directly modifying this, while ensuring the training logic is completely correct?

[Artiprocher]

@wangjiyuan9 Yes. It is correct. If you need more fine-grained multi-GPU training features, you can use accelerate config in your terminal. If you need a larger batch size, please use --gradient_accumulation_steps xxx.