参考文献如下
[1] 通过设置PYTORCH_CUDA_ALLOC_CONF中的max_split_size_mb解决Pytorch的显存碎片化导致的CUDA:Out Of Memory问题
https://blog.csdn.net/MirageTanker/article/details/127998036
[2] shell环境变量说明
https://blog.csdn.net/JOJOY_tester/article/details/90738717
具体解决步骤
报错信息如下:
RuntimeError: CUDA out of memory. Tried to allocate 6.18 GiB (GPU 0; 24.00 GiB total capacity; 11.39 GiB already allocated; 3.43 GiB free; 17.62 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计算 reserved – allocated = 17.62 – 11.39 = 6.23 > 6.18 (暂且不用管如何来的，更多说明参考文献[1])
查看CUDA中管理缓存的环境变量
echo $PYTORCH_CUDA_ALLOC_CONF设置环境变量的值（这里用到6.18这个数了，简单理解6.18表示缓存空间6.18GB）
export PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:6110（6110的由来简单理解为6110MB，我们要选择比6.18GB小的最大空间，推荐直接设置为6.1*1000MB）
问题圆满解决，可喜可贺 可喜可贺