[Performance] Improve MiMo-Audio tokenizer decoding performance#2183
Merged
Conversation
Signed-off-by: 齐保元 <qibaoyuan@xiaomi.com>
Signed-off-by: 齐保元 <qibaoyuan@xiaomi.com>
Signed-off-by: 齐保元 <qibaoyuan@xiaomi.com>
# Conflicts: # vllm_omni/model_executor/models/mimo_audio/mimo_audio_code2wav.py
Signed-off-by: 齐保元 <qibaoyuan@xiaomi.com>
Signed-off-by: 齐保元 <qibaoyuan@xiaomi.com>
Signed-off-by: 齐保元 <qibaoyuan@xiaomi.com>
Collaborator
|
will this pr solve the acc issue? |
Collaborator
|
can you provide the e2e improvement from this PR? |
Contributor
Author
CUDA Graph vs Eager Execution Performance ComparisonCompared with Eager execution, CUDA Graph delivers significant latency reduction and inference acceleration in both non-streaming and streaming modes. Performance Table
Key Observations
|
…r.synchronize Signed-off-by: 齐保元 <qibaoyuan@xiaomi.com>
qibaoyuan
requested review from
ZeldaHuang,
linyueqian,
princepride,
yenuo26 and
yuanheng-zhao
as code owners
clodaghwalsh17
pushed a commit
to clodaghwalsh17/nm-vllm-omni-ent
that referenced
this pull request
May 12, 2026
…-project#2183) Signed-off-by: 齐保元 <qibaoyuan@xiaomi.com> Co-authored-by: Jialong Liu <88185941+Galleons2029@users.noreply.github.com> Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com>
Galleons2029
added a commit
to Galleons2029/vllm-omni-ljl
that referenced
this pull request
May 18, 2026
…-project#2183) Signed-off-by: 齐保元 <qibaoyuan@xiaomi.com> Co-authored-by: Jialong Liu <88185941+Galleons2029@users.noreply.github.com> Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com> Signed-off-by: Jialong Liu <88185941+Galleons2029@users.noreply.github.com>
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
5 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.
Purpose
To improve the decoding capability of the audio tokenizer in the MiMo-Audio model, we focus on optimizing its efficiency, as it is frequently invoked in asynchronous scenarios. Improving its performance is therefore critical. Our approach leverages CUDA Graphs to accelerate execution.
Key changes include:
flash_attn_varlen_funcwithF.scaled_dot_product_attention, operating on 3D tensors [B, L, D], thereby avoiding variable-length packing.self_attn.forward_fixedwith thefeed-forward network (FFN).masked_select.dconv1 → transformer layers → dconv2 → vocoder.decode_vq, padding, anddecoder.forward_fixed.Test Plan
Test Result
0_3581f0d8-1ec1-4063-a223-72fa6a95b4a1.wav
Essential Elements of an Effective PR Description Checklist
supported_models.mdandexamplesfor a new model. Please runmkdocs serveto sync the documentation editions to./docs.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)