Skip to content

[Klaud Cold] H200 Qwen3.8-27B BF16 MTP, synthetic AL 2.51 / H200 Qwen3.8-27B BF16 MTP,合成 AL 2.51 - #3306

Draft
Oseltamivir wants to merge 3 commits into
mainfrom
codex/qwen38-27b-bf16-h200-vllm-mtp
Draft

Oseltamivir wants to merge 3 commits into
mainfrom
codex/qwen38-27b-bf16-h200-vllm-mtp

Conversation

@Oseltamivir

@Oseltamivir Oseltamivir commented Sep 20, 2026

Copy link
Copy Markdown
Collaborator

Goal: Rerun H200 Qwen3.8-27B BF16 with thinking on, synthetic AL 2.51 and 3 native MTP draft tokens.
Baseline recipe: #3293 · vllm/vllm-openai:nightly-cd10ed6f9f6b37a8ace9cf380007e66fe12ec0c3
1k/1k · TP1 · concurrency 1, 2, 4, 8, 16, 32, 64, 128.

qwen3.827b-bf16-h200-vllm-mtp uses Qwen/Qwen3.8-27B, BF16 target/native MTP weights and BF16 KV cache. The image, runner, graph mode, prefix-cache setting and concurrency-sized scheduler follow the source recipe.

Throughput sets rejection_sample_method: synthetic and synthetic_acceptance_length: 2.51, using the BF16 thinking_on[3] measurement in #3304. Require THINKING_MODE=thinking_on; explicitly enable thinking in server chat-template defaults. The fixed-sequence client retains --use-chat-template, whose checkpoint default renders a thinking-on prompt. Runs requesting accuracy through EVAL_ONLY or RUN_EVAL retain real verification.

Validation: Bash syntax, whitespace, append-only changelog and full matrix checks pass (8 throughput points; the current 1k/1k policy selects no default evals). All four actual recipes reject thinking-off before GPU startup. Rendering the checkpoint template confirms its default equals explicit thinking-on. The previous real-verification sweep passed. Synthetic sweep dispatched: run 35495274838 on 725f7fef3cdc3422f8eef4a990ee17b04d51f945.

AI model disclosure

GPT-6 prepared the BF16 and synthetic-acceptance changes and validation. The exact runtime model/version identifier was not exposed and could not be verified. The source FP8 recipe was credited to Claude Code; its underlying model/version was not disclosed and could not be verified.

中文

**目标:**重跑 H200 Qwen3.8-27B BF16,使用 thinking 开启、合成 AL 2.51 和 3 个原生 MTP 草稿 token
基线配方:#3293 · vllm/vllm-openai:nightly-cd10ed6f9f6b37a8ace9cf380007e66fe12ec0c3
1k/1k · TP1 · 并发 1、2、4、8、16、32、64、128。

qwen3.827b-bf16-h200-vllm-mtp 使用 Qwen/Qwen3.8-27B,目标权重、原生 MTP 权重及 KV cache 均为 BF16。沿用原配方的镜像、runner、graph 模式、prefix-cache 设置和按并发配置的 scheduler。

吞吐测试设置 rejection_sample_method: syntheticsynthetic_acceptance_length: 2.51,采用 #3304 中 BF16 的 thinking_on[3] 测量值。要求 THINKING_MODE=thinking_on,服务端默认 chat-template 参数显式启用 thinking。固定序列长度客户端保留 --use-chat-template,checkpoint 默认模板会生成 thinking 开启的提示词。通过 EVAL_ONLYRUN_EVAL 请求准确率评测时保留真实验证。

验证:Bash 语法、空白、changelog 只追加约束及完整矩阵检查均通过(8 个吞吐测试点;当前 1k/1k 策略默认不选取 eval)。实际执行四个配方,均在 GPU 启动前拒绝 thinking 关闭。实际渲染 checkpoint 模板,确认默认行为与显式开启 thinking 相同。此前的真实验证 sweep 已通过。合成 AL sweep 已启动:run 35495274838,提交为 725f7fef3cdc3422f8eef4a990ee17b04d51f945

**AI 模型披露:**GPT-6 完成 BF16、合成接受长度修改与验证;运行时未提供精确模型/版本标识,无法核实。原 FP8 配方署名 Claude Code,未披露底层模型/版本,无法核实。

新增 H200 Qwen3.8-27B BF16 原生 MTP 配方,以相同镜像和 1k1k 并发范围重跑 #3293 的工作负载。
补充 BF16 原生 MTP 配方对应的 PR 链接。
@github-actions

Copy link
Copy Markdown
Contributor

Thanks for the contribution!

  • Review: If this PR changes files owned by someone other than a repository admin or @SemiAnalysisAI/core, ask one eligible CODEOWNER to complete the latest PR_REVIEW_CHECKLIST.md before contacting a core maintainer on Slack. Follow the template exactly, including As a PR reviewer and CODEOWNER, I have reviewed this and have, so sign-off verification triggers.
  • PR verification: Sweeps only run on labeled PRs. Add full-sweep-fail-fast (strongly recommended); use full-sweep-enabled only when matrix jobs should continue after a failure.
  • After merging: PR authors must ensure all GitHub Actions jobs pass. Transient failures often pass on rerun; see how to rerun failed jobs.
中文

感谢你的贡献!

  • **审阅:**如果 PR 修改的文件归属于仓库管理员及 @SemiAnalysisAI/core 之外的 CODEOWNER,请先联系一位有资格的 CODEOWNER 填写最新的 PR_REVIEW_CHECKLIST.md,再通过 Slack 联系核心维护者。必须严格遵循模板,并保留 As a PR reviewer and CODEOWNER, I have reviewed this and have,才能触发签核验证。
  • **PR 验证:**扫描仅在带有标签的 PR 上运行。强烈建议添加 full-sweep-fail-fast;仅当需要矩阵任务在失败后继续运行时才使用 full-sweep-enabled
  • **合并后:**PR 作者必须确保所有 GitHub Actions 任务通过。临时性失败通常可以通过重新运行恢复;参见重新运行失败任务的说明

@Oseltamivir Oseltamivir added full-sweep-fail-fast qwen3.8-27b Qwen3.8-27B (bf16) vLLM TP1 DSpark 1k1k recipes and their eager variants labels Sep 20, 2026
@github-actions

github-actions Bot commented Sep 20, 2026

Copy link
Copy Markdown
Contributor

Qwen3.8-27B BF16 吞吐测试启用 thinking,使用 3 个原生 MTP 草稿 token 和合成 AL 2.51;准确率评测保留真实验证,并同步中英文说明。
@Oseltamivir Oseltamivir changed the title [Klaud Cold] Add H200 Qwen3.8-27B BF16 vLLM native MTP / 新增 H200 Qwen3.8-27B BF16 vLLM 原生 MTP [Klaud Cold] H200 Qwen3.8-27B BF16 MTP, synthetic AL 2.51 / H200 Qwen3.8-27B BF16 MTP,合成 AL 2.51 Sep 20, 2026

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

full-sweep-fail-fast qwen3.8-27b Qwen3.8-27B (bf16) vLLM TP1 DSpark 1k1k recipes and their eager variants

Projects

Status: No status

Development

Successfully merging this pull request may close these issues.

1 participant