Does Wan2.2 Inference support parrallelim like TP/CP or some cache feature ?

[pengchengneo]

As above, or it has other accellerate method for Wan2.2

[entrpn]

@SiqiLi-Fighting it supports context parallelism using the default configuration and command described here: https://github.com/AI-Hypercomputer/maxdiffusion?tab=readme-ov-file#wan22

The vae uses a caching mechanism like in the original pytorch model, is that what you're referring to?

[syhuang22]

CFG Cache is now supported in Wan 2.2, to enable it, use the configurable paramater use_cfg_cache=True.