H20 96G显存 qwen-image-edit lora微调为什么还是报错

torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 18.00 MiB. GPU  has a total capacity of 95.00 GiB of which 6.50 MiB is free. Process 3927658 has 73.02 GiB memory in use. Process 4105349 has 21.96 GiB memory in use. Of the allocated memory 21.65 GiB is allocated by PyTorch, and 1.51 MiB is reserved by PyTorch but unallocated. 