Repository navigation
[Doc] cookbook: Add Qwen3.8-Flash-Next B300 NVFP4 AgentX recipes - #47560
Merged
nvpohanh merged 2 commits intoOct 31, 2026
Merged
Conversation
hshrivastava-droid
requested review from
wisclmy0611 and
zijiexia
as code owners
October 31, 2026 00:43
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
Contributor
Preview approval requiredThis pull request comes from a fork, so Mintlify won't build a preview until someone with write access approves commit |
hshrivastava-droid
requested review from
JustinTong0323 and
sogalin
as code owners
October 31, 2026 00:43
github-actions
Bot
added
the
documentation
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
Collaborator
|
Merging for now. Will clean up next Monday |
nvpohanh
changed the title
nvpohanh
added a commit
to nvpohanh/sglang
that referenced
this pull request
Nov 2, 2026
…X module (sgl-project#47560) This reverts commit 26452a6. sgl-project#47560 added a hand-written data module (qwen3.8-flash-next-b300.jsx) next to the generated Qwen3.8-Flash-Next AgentX data and merged it into the page, for configurations from the still-open InferenceX sgl-project#4683. The Agentic Long-Context sections are generated from SemiAnalysis submissions by the cookbook sync; the next commit overlays InferenceX sgl-project#4683 through it, at the run sgl-project#47560 cited. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Yuxingwang-intel
pushed a commit
to Yuxingwang-intel/sglang
that referenced
this pull request
Nov 6, 2026
Yuxingwang-intel
pushed a commit
to Yuxingwang-intel/sglang
that referenced
this pull request
Nov 6, 2026
…in c5e8b9b3 and revert sgl-project#47560 (sgl-project#47929) Co-authored-by: Claude Opus 5.5 <noreply@anthropic.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
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.
The Qwen3.8-Flash-Next cookbook's B300 AgentX entry uses the RadixArk checkpoint. Add the NVIDIA NVFP4 configurations from InferenceX #4683 to its existing interactive builder, with the digest-pinned SGLang image and checkpoint revisions.
Cover TP4/EP4 at concurrency 1 and 2, TP1 GPU KV at 1, 2, 12 and 24, and TP1 HiCache at 48 and 80. Preserve per-configuration kernel, graph, memory, admission, speculative-decoding and cache settings. Explain the 298 GB HiCache host-memory allocation and the cache behavior assumptions for replay traffic. Retain the generated historical dataset unchanged; a separate source module adds the new entries without changing indexed references in existing entries. Expose only the submitted SGLang topology and cache tier for these configurations.
Validation: run 41117066222, attempt 1 passed all eight benchmark configurations, both eval jobs and result collection on source head
d06564bc533cc54584ef6a35837336a2b4b84421. Entries are marked verified against the source PR, pending its merge. These production commands omit synthetic acceptance controls; no throughput or accuracy claim is made. The pinned nightly's flags must not be assumed compatible with other releases.AI model disclosure
OpenAI GPT-6-based Codex prepared the cookbook changes and local validation. The exact runtime model identifier could not be verified.
中文
为现有交互式部署工具添加 InferenceX #4683 中的 NVIDIA Qwen3.8-Flash-Next NVFP4 B300 配置，覆盖 TP4/EP4、TP1 GPU KV 和 TP1 HiCache，共八个并发配置。保留固定 digest 的镜像、模型 revision 以及各配置的优化参数，并说明主机内存需求和回放流量的缓存假设。
完整 sweep 的八个 benchmark 配置、两个 eval job 和结果收集均已通过；新增配置标记为源 PR 已验证、待合并。生产命令不包含合成接受长度设置，不宣称性能或准确率已经验证。
CI States
Latest PR Test (Base): ✅ Run #41154512417
Latest PR Test (Extra): ❌ Run #41154512289
Latest PR Test (AMD ROCm 10): ➖ No AMD PR run found for this commit.