Closed HXZhong1997 closed 1 year ago
https://github.com/HarderThenHarder/transformers_tasks/blob/c56bcc4a19d960cb9481ff13d796fad3c303d749/LLM/finetune/train_multi_gpu.py#L190 之前没有接触过llm微调。请问大佬,这里多卡训练的num_update_steps_per_epoch数没有除以(num_gpus*batch_size),是有什么原因吗?
应该只会影响显示,实际训练流程不会受到影响。close了
https://github.com/HarderThenHarder/transformers_tasks/blob/c56bcc4a19d960cb9481ff13d796fad3c303d749/LLM/finetune/train_multi_gpu.py#L190 之前没有接触过llm微调。请问大佬,这里多卡训练的num_update_steps_per_epoch数没有除以(num_gpus*batch_size),是有什么原因吗?