[Question] Random seed in mult gpu training

[liyang-7]

Question

Training loss is fixed in single GPU training, but it is different in multi GPU training even with fixed seed in TrainingArguments How to get same loss and weights in multi GPU training based on current scripts？