how to train costomize data with lora

[baaiy0610]

在train自己的数据集的lora时，
_child_module总是为LoRACompatibleLinear导致_module._modules[name] = _tmp 没有lora_up

[Vicky0522]

Have you tried pip install peft==0.11.1?

[baaiy0610]

thanks it works, but i got the Floating point exception without any error warning. I debugthis, it seems an error comes from loss_temporal