[nvpohanh]

[by Claude Code]

Motivation

The cookbook's Agentic Long-Context Deployment sections (#42726, #43083, #43799, #43983, #44065, #44228, #44654, #44712, #45039, #45111, #45311, #45513, #45749, #46005, #46506, #46547, #46807, #46948, #47027, #47214, #47303) mirror SemiAnalysis InferenceX AgentX submissions as raw launch commands, each linked to the run that measured it. The configurations are mirrored from SemiAnalysis submissions. This PR:

1. Syncs to InferenceX main (ea8b3439). Three PRs have merged, and their lanes now come from their dashboard runs instead of the PRs:
   • DeepSeek-V4.1-Flash B300 (InferenceX #4627, 2026-10-29): dsv41flash-fp4-b300-sglang-agentic-dspark and -dspark-lowconc, previously in progress.
   • MiniMax-M3.1-Flash B300 TP2 (InferenceX #4524), and the shortest-prefill-first TP2 points added on top of it (InferenceX #4647, run 40895671658, append-only on 2026-10-29). Each point is credited to the run that measured it. [Doc] cookbook: Add MiniMax-M3.1-Flash B300 TP2 SPF AgentX points #47303 had added the LZ Sampling Penalty #4647 points to the data by hand; they are now generated by the sync.
2. Adds Muse Spark Open updates from two in-progress InferenceX PRs:
   • GB200 SGLang, retuned per concurrency (InferenceX #4670).
   • GB300 SGLang, with two new keys, musespark-fp4-gb300-sglang-agentic-dflash-multinode and -dflash-nightly1019 (InferenceX #4454, overlaid again after [Doc] cookbook: Refresh SemiAnalysis AgentX launches to InferenceX main 70a76017 #47214).
3. Keeps the DeepSeek-V4.1-Flash GB300 overlays (InferenceX #4603, and InferenceX #4595 at its new head) and the Qwen3.5 GB300 overlay (InferenceX #4333).

Source

InferenceX main (ea8b3439), with these InferenceX PRs overlaid:

PR	Page	Lanes	Pinned to
#4670 Muse Spark Open GB200 SGLang, retuned per concurrency	Muse Spark Open	musespark-fp4-gb200-sglang-agentic-dflash	no /use yet: in progress at head 5692839d
#4454 Muse Spark Open NVFP4 GB300 SGLang	Muse Spark Open	new musespark-fp4-gb300-sglang-agentic-dflash-multinode, -dflash-nightly1019	no /use yet: in progress at head a73b274a
#4603 DeepSeek-V4.1-Flash GB300 SGLang, 20261029 nightly, TP4 c1 megakernel	DeepSeek-V4.1	dsv41flash-fp4-gb300-sglang-agentic-dspark	no /use yet: in progress at head 9bd047ac
#4595 DeepSeek-V4.1-Flash GB300 Dynamo+SGLang, bonus-fixed nightlies	DeepSeek-V4.1	dsv41flash-fp4-gb300-dynamo-sglang-agentic-agg, -dspark-disagg-hicache; retires -dspark-disagg	no /use yet: in progress at head 4145e263
#4333 Qwen3.5 NVFP4 GB300 Dynamo+SGLang	Qwen3.5	qwen3.5-fp4-gb300-dynamo-sglang-agentic-agg, -disagg; retires -agg-pareto, -pp-pareto	run 39355392627, named by the PR's /use comment (head 02b72dc0, unchanged), verified

"In progress" lanes are rendered at the PR head and labeled "In progress in SA InferenceX AgentX". A later sync pins them once the PR gets a /use run or merges.

• DeepSeek-V4 page unchanged. InferenceX #4664 retired DeepSeek-V4-Pro from active benchmarks: InferenceX main no longer has dsv4-* keys, though the published results remain. This PR does not touch deepseek-v4.jsx. Its section keeps the launches [Doc] cookbook: Refresh SemiAnalysis AgentX launches to InferenceX main 70a76017 #47214 put live (InferenceX main 70a76017 and InferenceX #4556's /use run), each still linked to the run that measured it. Later refreshes will not update it.
• No longer overlaid:
  • DeepSeek-V4-Pro B200 disagg (InferenceX #4556): its model was retired, see above.
• Retired keys (as in [Doc] cookbook: Refresh SemiAnalysis AgentX launches to InferenceX main 70a76017 #47214):
  • InferenceX #4595 deletes dsv41flash-fp4-gb300-dynamo-sglang-agentic-dspark-disagg and replaces it with -dspark-disagg-hicache.
  • InferenceX #4333 deletes qwen3.5-fp4-gb300-dynamo-sglang-agentic-agg-pareto and -pp-pareto.
• Not mirrored (unmeasured, unchanged): musespark-fp4-h100-sglang-agentic-dflash and qwen3.5-fp8-h100-sglang-agentic. The dashboard has no rows for them.
• InferenceX main was frozen for this PR at ea8b3439. Later merges are left for the next refresh.

What this PR does

Page	Lanes	Deployments	Points	Hardware
MiniMax-M3.1	4	19 (17 in #47214)	65 (59 in #47214)	GB300, GB200, B300, B200
GLM-5.2	14	34	75	GB300, GB200, B300, B200, H200
GLM-5.3	1	1	10	B300
DeepSeek-V4	unchanged
DeepSeek-V4.1	10	21	65	GB300, GB200, B300, B200, H200, H100
Qwen3.5	16	51	220	GB300, GB200, B300, B200, H200, H100
Qwen3.8-Flash-Next	4	5	24	GB300, B300, B200, H200
Muse Spark Open	8 (was 6)	11 (was 9)	70 (was 66)	GB300, GB200, B300, B200, H200

• No new keep / remove rules were needed. The GLM-5.2, GLM-5.3, Qwen3.5 and Qwen3.8-Flash-Next data change only in their shared source block (the InferenceX main SHA and the overlay list).

Validation

• Browser checks: every page this PR changes was rendered in a headless browser (mint dev, Mintlify as in CI).

  • Every hardware / checkpoint / deployment / concurrency / KV tier / router combination renders exactly the decoded commands: 2,868 command sets, 0 mismatches, 0 unseen, no browser errors.
  • Every tab was clicked, and every "Copy link" URL was reopened.
  • The section links from the Deployment panels (DeepSeek-V4 included) and the copy buttons were also checked.

• Decode and KV-flag checks:

  • The component's own decoder reproduces every generated plan (2,868 router × KV-tier variants).
  • Every KV flag exists in its image's SGLang commit (218 image × flag pairs).
  • The round-trip / leak check reports 0 problems.

• Arg-parse in the submission images: all 96 commands not validated in an earlier round parse with their image's own parsers:

  Image	Commands
  dev-minimax-m31-flash	36
  nightly-dev-20261026-8252e3af	48
  nightly-dev-cu13-20261020-ab580b2e	12

  The amd64 and arm64 builds of each nightly come from the same SGLang commit.

• node scripts/check_cookbook_configs.mjs, mint broken-links --check-anchors --check-redirects, mint validate and pre-commit run --all-files (in lmsysorg/sglang:dev-cu13) pass.

🤖 Generated with Claude Code

---

CI States

Latest PR Test (Base): ✅ Run #41046383160
Latest PR Test (Extra): ❌ Run #41046383120
Latest PR Test (AMD ROCm 10): ➖ No AMD PR run found for this commit.

[claude]

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.

[mintlify]

✅ Preview approved

@nvpohanh approved a Mintlify preview for commit 6a9e8b1.

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated
sglang-doc	🟢 Ready	View Preview	Oct 30, 2026, 11:19 AM

💡 Tip: Enable Automations to automatically generate PRs for you.

[nvpohanh]

Preview looks good. Merge to unblock SA AgentX submission