How to use RTX 4090 *2 running train

[Fucapisun]

When I train with two GPUs using CUDA, I get a memory error message, torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 416.00 MiB (GPU 0; 23.64 GiB total capacity; 21.07 GiB already allocated; 405.69 MiB free; 22.77 GiB reserved in total by PyTorch)indicating that only one GPU is being called. Is there any way to train with two GPUs simultaneously?

[wozuicai]

May I ask which size of model you are using: base, medium, or large? And what is the input size? I hope to use this model on other datasets, but I am currently concerned about the VRAM usage

[Fucapisun]

I used base model to train.And I follow nnUNet to preprocess my dataset.