When I evaluate GPT, It takes more than 10 hours to evaluate one time on V100, which means that it takes more than 8 days to repeat 20 times. Is this normal?
I have the same problem. Check your pytorch version, it needs to be updated. I just pip uninstall torch and pip install torch torchvision torchaudio and it works.
Thank you for your great work!
When I evaluate GPT, It takes more than 10 hours to evaluate one time on V100, which means that it takes more than 8 days to repeat 20 times. Is this normal?