When reasoning, how to multi-card on a single machine?

[tiga-dudu]

torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 1.55 GiB (GPU 1; 23.64 GiB total capacity; 21.21 GiB already allocated; 120.25 MiB free; 21.39 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

Even if I set model = model.to(gpu) -> model = model.half().to(gpu), it still doesn't work.

export CUDA_VISIBLE_DEVICES=0,1,2,3 Equally doesn't work， Can someone help me?