Local Play: CUDA out of memory.

[Koakuma-spec]

This error does not always happen, sometimes when I talk to my AI it responds just fine. Other times...

Full error summary when I tried to give a prompt to my AI:

RuntimeError: CUDA out of memory. Tried to allocate 50.00 MiB (GPU 0; 2.00 GiB total capacity; 1.23 GiB already allocated; 0 bytes free; 1.34 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

I'm new to this program so I'm not exactly sure what to do to avoid this. I can't find a documentation here that helps me with my issue and I don't know what PYTORCH_CUDA_ALLOC_CONF is. All I know is that my computer can sometimes run out of memory, so what are some settings I can manually adjust to prevent this from happening?

I am using 2.7B Nerys with 3 layers and 0 disk cache (this is the max my computer can handle, I've done a lotta trial and error to figure this out)