fix: improve Qwen3.5 MTP accept rate - #1402
Merged
Merged
Conversation
Contributor
|
Caution The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased. |
sufubao
force-pushed
the
fix-qwen35-mtp-accept-rate
branch
from
July 24, 2026 08:08
d315b20 to
af19707
Compare
sufubao
force-pushed
the
fix-qwen35-mtp-accept-rate
branch
from
July 24, 2026 10:13
af19707 to
e27ef4e
Compare
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Qwen3.5 MTP draft 需要使用 final norm 后的 hidden，当前实现直接传入 raw hidden，导致接受长度偏低。
本 PR 像共享 embedding/lm_head 一样把 main final norm 共享给 draft，并在 draft 输入融合前执行 norm；改动仅限
qwen3_5_mtp。验证：H200、Qwen3.5-0.8B、TP=1、
eagle_with_att、mtp_step=3，8 个固定请求各生成 256 tokens；按相同口径汇总的 accept length 从1.804提升到2.738（+51.7%），输出与 producer-side post-norm 版本逐项一致。相关 MTP CUDA 单测 2/2 通过，Black、Flake8 和py_compile通过。