Skip to content

[feature] add TeaCache support for OmniGen2#2257

Draft
zzhuoxin1508 wants to merge 6 commits into
vllm-project:mainfrom
zzhuoxin1508:omnigen2-teacache
Draft

[feature] add TeaCache support for OmniGen2#2257
zzhuoxin1508 wants to merge 6 commits into
vllm-project:mainfrom
zzhuoxin1508:omnigen2-teacache

Conversation

@zzhuoxin1508
Copy link
Copy Markdown
Contributor

PLEASE FILL IN THE PR DESCRIPTION HERE ENSURING ALL CHECKLIST ITEMS (AT THE BOTTOM) HAVE BEEN CONSIDERED.

Purpose

This PR implements TeaCache support for OmniGen2

Test Plan

I will provide the detailed benchmark and test results as soon as possible.

Test Result


Essential Elements of an Effective PR Description Checklist
  • The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
  • The test plan. Please provide the test scripts & test commands. Please state the reasons if your codes don't require additional test scripts. For test file guidelines, please check the test style doc
  • The test results. Please paste the results comparison before and after, or the e2e results.
  • (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model. Please run mkdocs serve to sync the documentation editions to ./docs.
  • (Optional) Release notes update. If your change is user-facing, please update the release notes draft.

BEFORE SUBMITTING, PLEASE READ https://github.com/vllm-project/vllm-omni/blob/main/CONTRIBUTING.md (anything written below this line will be removed by GitHub Actions)

@wtomin
Copy link
Copy Markdown
Collaborator

wtomin commented Mar 30, 2026

A recent PR changed the diffusion features docs strucure. Pls PTAL #1928.

@zzhuoxin1508
Copy link
Copy Markdown
Contributor Author

A recent PR changed the diffusion features docs strucure. Pls PTAL #1928.

Noted, thanks!

Signed-off-by: zhou zhuoxin <zhouzhuoxin1508@outlook.com>
Copy link
Copy Markdown
Collaborator

@lishunyang12 lishunyang12 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

left a couple comments

Comment thread vllm_omni/diffusion/cache/teacache/extractors.py
Comment thread vllm_omni/diffusion/cache/teacache/extractors.py
@@ -37,6 +37,15 @@
# Bagel transformer coefficients
# Using Qwen's coefficients as reasonable default given shared architecture
"Bagel": [1.33313129e06, -1.68644226e05, 7.95050740e03, -1.63747873e02, 1.26352397e00],
Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The test plan says results will be provided "as soon as possible" — are these coefficients validated at all yet? Shipping untuned polynomial coefficients could silently degrade output quality. Would be good to at least confirm basic generation quality before merging.

@zzhuoxin1508
Copy link
Copy Markdown
Contributor Author

Sorry for the delay.Right now TeaCache only supports 2-branch CFG ,but omnigen2 is 3-branch . I've filed this as issue#2371. So these coefficients only cover the 2-branch path; fuller validation depends on #2371. @lishunyang12

Signed-off-by: zhou zhuoxin <zhouzhuoxin1508@outlook.com>
Signed-off-by: zhou zhuoxin <zhouzhuoxin1508@outlook.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants