CUDA out of memory

我在24G显存的3090上，进行视频推理，有如下报错
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 110.50 GiB (GPU 0; 23.69 GiB total capacity; 10.06 GiB already allocated; 12.66 GiB free; 10.40 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

但是理论上24g足够一个7b大模型跑视频推理的，为什么上面会需要110G显存

Start with the video-inference setup that produces the reported torch.cuda.OutOfMemoryError on a 24 GB RTX 3090, then compare the requested 110.50 GiB with the reported allocation and reservation figures. Confirm the cause and document a reproducible explanation or required configuration change; the issue names no files or tests.

Written by the indexing model from the issue text.