指定的ModelScope gpu，但是只用到了一张2卡，怎么修改？

"torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 28.57 GiB (GPU 2; 39.59 GiB total capacity; 33.25 GiB already allocated; 5.06 GiB free; 33.60 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF我用这种方式os.environ['CUDA_VISIBLE_DEVICES']='0,1,2,3,4,5,6,7' 指定的ModelScope gpu，但是只用到了一张2卡，怎么修改？
"