Repository navigation
Conversation
|
Important Draft PR not reviewedDraft PRs are not automatically reviewed by default.
To automatically review draft PRs, update your CodeRabbit configuration: reviews:
auto_review:
drafts: true
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. CodeRabbit Verify
|
sufubao
force-pushed
the
support-qwen38-flash-next
branch
from
November 3, 2026 07:19
7c20a86 to
1c67342
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
支持 Qwen3.8-Flash-Next(checkpoint 类型
qwen4_exp),包括 gated residual、n-gram embedding、QSA/GDN、视觉及 MTP 集成。数值组件独立放在 common;CPU embedding offload 默认关闭,TopK 优化仅在匹配的 H100/Torch/Triton 配置启用。已 rebase 到
aab72d919。QSA 索引键、位置数据接入主线 GQA State,词法历史使用独立 bounded State;CPU/PD/DP 传输和 MTP 接受状态提交均沿用 AttentionStateManager,不恢复已删除的请求/缓存管理器。模型与 draft 使用懒加载注册,HTTP 保留主线原子批量槽分配及 PD 超时语义。本次检查:1022 项通过,930 项 GPU 用例跳过;170 个 Python 文件通过语法、Black、Flake8。DeepSeek V4 集成模块因缺少 FlashMLA 未收集;两项 State identity 测试因无 CUDA 无法分配 pinned memory,在未修改的 upstream/main 上复现相同失败。
历史 8×H100、BF16 TP8、MTP2 的 4× Mooncake 回放为 LightLLM 886–887 tok/s、SGLang 835、vLLM 524;1319 个数学样例与原实现 token 一致、正确率 96.44%。这些结果属于迁移前的历史快照,不能证明本次 rebase 后的精度和速度。低负载部分延迟仍落后,历史四页块压力恢复有文本差异。
保留草稿状态:本次未进行 GPU 服务精度、完整 PD 或性能验证。测试口径和迁移范围见
docs/qwen38_adaptation.rst、docs/qwen38_h100_performance.rst。已发布的旧镜像仍对应7c20a86e1,未自动重新构建或部署。