search

huggingface / diffusers

🤗 Diffusers: State-of-the-art diffusion models for image and audio generation in PyTorch and FLAX.
https://huggingface.co/docs/diffusers
Apache License 2.0
26.4k stars 5.43k forks source link

Parallel Inference support ffor Dit model #10023

Open enozhu opened 2 days ago

enozhu commented 2 days ago

does diffusers have any plan to support Parallel Inference like TP/PP/SP？

a-r-r-o-w commented 2 days ago

Nope, Diffusers does not support TP/SP at the moment. You can get PP to work more easily though and there are examples available for that.

We do plan to eventually support TP/SP (cc @yiyixuxu). For now, good reference would be ParaAttention by chengzeyi

enozhu commented 2 days ago

thanks for replying, any plan to support cpp(chunked pipeline parallel)? like pipefusion in xdit, It seems faster than common pipeline parrallel