calculate-flops.pytorch
calculate-flops.pytorch copied to clipboard
How to calculate the FLOPs when the model has some freezed variables, or using LoRA?
I suppose the corresponding backward calculations should be discarded.