RuntimeError: CUDA out of memory. about casanovo HOT 3 CLOSED

You can try decreasing the following parameters:

val_batch_size: 1024
test_batch_size: 1024

However, the error message is a little weird here:

 7.79 GiB total capacity; 2.46 GiB already allocated; 632.94 MiB free; 3.65 GiB reserved in total by PyTorch

This implies to me that there may be another process running on your GPU.

from casanovo.

Thanks. Would it be possible to catch this error and issue a more informative message such as this?

RuntimeError: CUDA out of memory. Tried to allocate 714.00 MiB (GPU 0; 7.79 GiB total capacity; 2.46 GiB already allocated; 632.94 MiB free; 3.65 GiB reserved in total by PyTorch) 
Consider reducing the value of val_batch_size and test_batch_size.

from casanovo.

It's not super nice design because the error would need to be caught at a very high level, basically in the outermost main function. Additionally, we'd need to string match the error message to figure out when catching a RuntimeError to check whether it matches a GPU OOM, which is a bit iffy.

I think the message is reasonably informative, and I'd just add a note about it in the FAQ section.

from casanovo.