You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hello, you guys implemented Sequence parallel approach for Flux. But, the inference speed for 3 and 4gpus is slower as compared to 1 GPU. Do you have ideas to speed it up using distrifuser or pipefuser? Why is it not possible in pipefusion?
The text was updated successfully, but these errors were encountered:
Hello, you guys implemented Sequence parallel approach for Flux. But, the inference speed for 3 and 4gpus is slower as compared to 1 GPU. Do you have ideas to speed it up using distrifuser or pipefuser? Why is it not possible in pipefusion?
The text was updated successfully, but these errors were encountered: