New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
OOM error #8
Comments
Hi, I used a 32GB GPU for the XSUM experiments. You could either switch to a GPU with larger memory, or you could reduce the bsz and increase the gradient_accumulation_steps. |
Hi, I used one GPU (Tesla V100 SXM2 32GB) and used the command in the homepage. |
Hi, it's the command to reproduce. Could you check if you have --fp16 yes and whether this turn on half-precision? This should turn on half precision, so that bsz=16 could fit. Side Note: I used AWS single GPU (I think it's A100) to run all XSUM experiments. |
I have Side Note: the A100 in AWS has 40GB GPU memory rather than 32GB. |
Maybe check if your stdout contains this: |
Thanks! By the way, what does the |
It means the dim of the MLP's middle layer! (we use an MLP for re-parametrization.) |
Got it! Thanks for your answer! |
Thanks! My mistake. |
Hi, I tried the seq2seq prefixtuning and found:
RuntimeError: CUDA out of memory. Tried to allocate 1.20 GiB (GPU 0; 15.90 GiB total capacity; 4.63 GiB already allocated; 797.50 MiB free; 5.81 GiB reserved in total by PyTorch)
I run the expr on a 16GB GPU. Am I supposed to use a 32GB GPU instead? Thanks!
The text was updated successfully, but these errors were encountered: