torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 92.99 GiB

#2
by chenrq2005 - opened

when using GPU (A100-40GB) encountering the torch.cuda.OutOfMemoryError error, any suggestion?

SpeechBrain org

Hi,
Sorry for the late response, in this case you may need to reduce the inference batch size and check your audio length. The input should be a speech turn instead of an entire conversation. If you have a long audio, you can first segment with a vad and then pass the segments to the model.

Thanks Yingzhi!

Sign up or log in to comment