Make HunyuanDiT2DModel and HunyuanDiT2DControlNetModel hidden states contiguous#14203
Open
menglcai wants to merge 1 commit into
Open
Make HunyuanDiT2DModel and HunyuanDiT2DControlNetModel hidden states contiguous#14203menglcai wants to merge 1 commit into
HunyuanDiT2DModel and HunyuanDiT2DControlNetModel hidden states contiguous#14203menglcai wants to merge 1 commit into
Conversation
Member
|
Thanks for your PR. Instead of Hunyuan, do we want to treat another model that has popularity and more usage? |
Contributor
|
Hi @menglcai, thanks for the PR! It does not appear to link an issue it fixes. If this PR addresses an existing issue, please add a closing keyword (e.g. |
Contributor
Author
Understood, I'll look into more popular or widely used models and open a follow-up PR if a similar fix applies. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What does this PR do?
Similar to #14186
The non-contiguous tensor originates from
PatchEmbed.forward—flatten(2).transpose(1, 2) (BCHW → BNC)produces a non-contiguous layout whose overhead accumulates across transformer blocks inHunyuanDiT2DModelandHunyuanDiT2DControlNetModel.Unlike SD3 where the benefit is limited to ROCm, this patch improves performance on both ROCm and NVIDIA for HunyuanDiT, though the gain on NVIDIA is smaller.
Benchmarks (HunyuanDiT-v1.1-Diffusers, 512×512, 50 steps, fp16):
Benchmarks (HunyuanDiT-v1.2-Diffusers, 512×512, 50 steps, fp16):
Benchmarks (HunyuanDiT-v1.2-ControlNet-Diffusers-Canny + HunyuanDiT-v1.2-Diffusers-Distilled, 512×512, 50 steps, fp16):
Before submitting
self-reviewskill on the diff?documentation guidelines, and
here are tips on formatting docstrings.
Who can review?
@sayakpaul @yiyixuxu
Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.