GitLab

[feat]: refactor to adapt new config system and align precision for qwen-image & qwen-image-edit (#383)

[feat]: refactor to adapt new config system and align precision for
qwen-image & qwen-image-edit

主要用于步数蒸馏实验
- 支持在 highnoise / lownoise 文件夹下指定特定的checkpoint用于加载
- 删除冗余的 Wan22MoeDistill 类
- 修改同时发起多个请求时，请求中没有 seed 参数导致的 bug

Co-authored-by: root <root@pt-731f15f5c62740f495519352dc90d9c4-worker-0.pt-731f15f5c62740f495519352dc90d9c4.ns-devsft-3460edd0.svc.cluster.local>

support self-forcing wan2.1 dmd

Co-authored-by: PengGao <peng.gaoc@gmail.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Yang Yong (雍洋) <yongyang1030@163.com>

Co-authored-by: Yang Yong(雍洋) <yongyang1030@163.com>

* update seko talk

* update

* Update configs

* update

---------
Co-authored-by: gushiqiao <975033167@qq.com>

* feature: qwen-image support cpu offload (block) and refactor transfomrer model

* bugfix

* update seko file names

* update

* [Update] update seko-models configs

* Add seko scripts

---------
Co-authored-by: gushiqiao <975033167@qq.com>

* [Feat] audio driven for wan2.2-r2v-5b

* refactor: reduce vae args dim &&  timestep init

* update audio pre_infer

* update runner

* update scripts

* [Fix] move ref timestep padding to Scheduler

* fix wan i2v

---------
Co-authored-by: helloyongyang <yongyang1030@163.com>

* [Fix] Fix audio model offload bug.

* Fix

* Fix

* Fix

* Fix

* Fix

* Fix

* feature: support qwen-image-edit(i2i)

* bugfix

* Add magcache

* fix

* lint

* fix

* update script

* Support cfg infer

* support cfg parallel

* update cfgsp/t2v config

* support magcache calibration

* calibration

* feature: support qwen-image t2i

* pass ci

---------
Co-authored-by: wushuo1 <wushuo1@sensetime.com>