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.

Sign up for GitHub

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

Jump to bottom

use pytorch built-in SiLU function to save GPU memory usage #331

Open

pppoe wants to merge 1 commit into Stability-AI:main from pppoe:main

pppoe commented Oct 5, 2023

Replaced the custom implemented nonlinearity function with the PyTorch built-in function SiLU.
The built-in function saves significant GPU memory when specifying a large output size.
According to PyTorch documentation, the formulation/result is identical.

On my 12GB GPU card, the following command gives a GPU OOM error without this PR while runs through with it.

python3 scripts/txt2img.py --H 1024 --W 1024 --plms --ckpt /data/models/SD/v2-1_512-ema-pruned.ckpt --config configs/stable-diffusion/v2-inference.yaml  --device cuda --prompt "pytorch logo" --n_sample 1  --n_iter 1

Tested under the default environment requirements. SiLU function has been available in PyTorch after 1.7.0.


          use pytorch built-in SiLU function to save GPU memory usage

51c8130

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet