[Docs] Cookbook: refresh SemiAnalysis AgentX launches (DeepSeek-V4.1 GB200/GB300, GLM-5.2 FP8 B300)

[nvpohanh]

[by Claude Code]

Motivation

Refresh the cookbook's interactive Agentic Long-Context Deployment sections (#42726, #43083, #43799) on GLM-5.2, DeepSeek-V4, DeepSeek-V4.1, Qwen3.5 and Qwen3.8-Flash-Next against the latest SemiAnalysis InferenceX AgentX submissions. The configurations are mirrored from SemiAnalysis submissions as raw launch commands, each linked to the run that measured it.

This PR supersedes #43934. It brings in the GLM-5.2 FP8 B300 update from InferenceX #3836 (c20/c24/c28 added to the existing c1–c16) through the AgentX explorer and leaves the Deployment panel unchanged.

Source

InferenceX main (029d019d), with these open InferenceX PRs overlaid:

PR	Page	Lane(s)	Pinned to
#3927 DSV4.1 GB200 SGLang	DeepSeek-V4.1	dsv41flash-fp4-gb200-sglang-agentic-dspark	run 38266927261 (/use), verified
#3859 DSV4.1 GB300 SGLang	DeepSeek-V4.1	dsv41flash-fp4-gb300-sglang-agentic-dspark	run 38221097558 (/use), verified
#3836 GLM-5.2 FP8 B300 SGLang	GLM-5.2	glm5.2-fp8-b300-sglang-agentic-mtp	no /use yet: in progress, PR head 3120bc1b
#3817 DSV4 B300 Dynamo+SGLang	DeepSeek-V4	dsv4-fp4-b300-dynamo-sglang-agentic-agg, -disagg	run 37715955859 (/use), verified
#2978 Qwen3.5 B300 Dynamo+SGLang	Qwen3.5	qwen3.5-fp4-b300-dynamo-sglang-agentic-pareto	run 34548026742 (/reuse-sweep-run), verified
#3606 DSV4 B300 SGLang	DeepSeek-V4	dsv4-fp4-b300-sglang-agentic-hicache-mtp	no /use yet: in progress, PR head 5b394d39
#3860 DSV4.1 GB300 Dynamo+SGLang	DeepSeek-V4.1	dsv41flash-fp4-gb300-dynamo-sglang-agentic-disagg	no /use yet: in progress, PR head b2061c77

A lane from a PR without an authorized /use run is rendered at the PR head and labeled "In progress in SA InferenceX AgentX". It switches to the measured run on the next sync after a /use comment or a merge. Two overlays from #43799, InferenceX #3890 and InferenceX #3701, have since merged into InferenceX main and are now mirrored from there, at their dashboard runs.

What this PR does

Page	Lanes	Deployments	Points	Hardware
GLM-5.2	15	25	60	GB300, GB200, B300, B200, H200
DeepSeek-V4	12	26	75	GB300, GB200, B300, B200, H200
DeepSeek-V4.1	9	24	71	GB300, GB200, B300, B200, H200, H100
Qwen3.5	19	50	221	GB300, GB200, B300, B200, H200, H100
Qwen3.8-Flash-Next	4	4	22	GB300, B300, B200, H200

Only the generated data files change. The component (_agentx.jsx), page text, Deployment panels and Playgrounds are untouched. Rendered commands change only for these three lanes:

• DeepSeek-V4.1 GB200 SGLang (InferenceX #3927). The lane is now launched from an srt recipe instead of the bash launcher used by InferenceX #3347. Its two curves, "Agg TP2" and "Agg TP4" (c1–c128, on the pruned nightly-dev-cu13-20260923-06008c17 tag), are replaced by "Agg TP4·EP4" (c1/c4/c8/c16) and "Agg DEP4" (c32/c64/c128/c192). Both new deployments run on lmsysorg/sglang:nightly-dev-cu13-20261011-d32d8fa0@sha256:8346c975…, which is available on Docker Hub.
• DeepSeek-V4.1 GB300 SGLang (InferenceX #3859). Now verified at its /use run; [Docs] Cookbook: refresh SemiAnalysis AgentX launches (GLM-5.2 B300, DeepSeek-V4.1 GB300) #43799 had it in progress at an earlier head. The curve has 7 points: "Agg TP4" c1/c4/c16, "Agg TP2·EP2" c16/c32 and "Agg DEP2" c64/c128. TP4 c1, TP2·EP2 c16/c32 and DEP2 c64/c128 are byte-identical to main, and TP4 c4/c16 are new. TP2·EP2 c1–c8, TP4 c2 and the "Agg TP4·EP4" deployment (c4–c128) are no longer in the submission and are removed.
• GLM-5.2 FP8 B300 SGLang (InferenceX #3836). Adds c20, c24 and c28 to the "Agg TP8 · MTP 3-1-4" deployment, with --max-running-requests and --cuda-graph-max-bs set to 40, 48 and 56. The c1–c16 commands are byte-identical. The PR was force-pushed to a single commit and its earlier /use comment was removed, so the whole lane (c1–c28) now shows as in progress at the PR head (3120bc1b) until it gets a new /use run.

Status- and provenance-only changes (the commands are byte-identical to main): GLM-5.2 NVFP4 B300 (InferenceX #3890) and DeepSeek-V4 GB200 Dynamo agg/disagg (InferenceX #3701) are now pinned to their dashboard runs on InferenceX main.

How the commands are produced

The pipeline and rules are the same as in #42726, #43083 and #43799, and no keep/remove/rewrite rule was added. Each lane is rendered at the exact commit of the run that produced its dashboard data. For an open PR, that is the run named by its authorized /use comment, or the PR head when there is none. SA-only settings are removed: the benchmark harness, cluster paths and caches, telemetry, and InferenceX's synthetic speculative acceptance (disclosed per deployment). Placeholders replace IPs, ports and NICs. Router and KV-offload variants that were not run are derived and labeled "Derived — not run".

• Not mirrored. qwen3.5-fp8-h100-sglang-agentic (H100 without MTP) has never produced dashboard data, so there is no run to pin it to.
• DeepSeek-V4.1 HiCache donor. Unchanged from [Docs] Cookbook: refresh SemiAnalysis AgentX launches (GLM-5.2 B300, DeepSeek-V4.1 GB300) #43799: the measured main version of dsv41flash-fp4-gb300-dynamo-sglang-agentic-disagg keeps donating the derived HiCache flags while InferenceX #3860's revision is unmeasured.

The data files are generated, so there is no hand-written code to review here.

Validation

• mint validate (strict build) and mint broken-links --check-anchors --check-redirects pass. node docs/scripts/check_cookbook_configs.mjs passes, and so does pre-commit run --all-files.
• A headless browser clicked through every combination of hardware, checkpoint, deployment, concurrency, KV tier and router on all five pages under mint dev, opening every tab. That is 2,274 command sets: GLM-5.2 300, DeepSeek-V4 424, DeepSeek-V4.1 426, Qwen3.5 1,048 and Qwen3.8-Flash-Next 76. The rendered commands match the expected text exactly. For every deployment, the "Copy link" URL of a non-default selection was reopened in a fresh tab and restored the same selection and tab (129 links).
• The Deployment-panel footer link still appears on GLM-5.2, DeepSeek-V4, DeepSeek-V4.1 and Qwen3.8-Flash-Next and scrolls to the section. Qwen3.5's anchor sits directly above the explorer, and pages that use the Deployment panel without the prop still show no link.
• The component's decoder and KV-tier transform, run through Node, reproduce the Python-generated commands for all 2,274 combinations, including every point's measured tier. Every KV-cache flag in a shown command exists in the SGLang commit of its deployment's image (188 image/flag pairs).
• Compared with main, every deployment outside the three lanes above renders byte-identical commands on all five pages, including the derived router and KV-offload variants.
• Single-node commands from bash launchers were accepted only where a dry run of the pinned launcher reproduced the server command in the run's job log (221 points).
• Every submitted and derived sglang.launch_server, dynamo.sglang, sglang_router.launch_router and dynamo.frontend command (1,934 distinct, across 28 images) parses with the argument parsers of its own submission image (CPU only). The 80 commands that are new in this PR were parsed now, each in its own image: 50 in nightly-dev-cu13-20261011-d32d8fa0, 12 in nightly-dev-cu13-20261012-e6072935 and 18 in the GLM-5.2 FP8 B300 image lmsysorg/sglang@sha256:2d53500c…. The rest are identical to commands parsed for [Docs] Cookbook: interactive SemiAnalysis InferenceX AgentX launches for GLM-5.2, DeepSeek-V4 and DeepSeek-V4.1 #42726, [Docs] Cookbook: AgentX launches for Qwen3.5 / Qwen3.8-Flash-Next; refresh GLM-5.2, DeepSeek-V4, DeepSeek-V4.1 #43083 and [Docs] Cookbook: refresh SemiAnalysis AgentX launches (GLM-5.2 B300, DeepSeek-V4.1 GB300) #43799.

No serving or benchmark runs were performed for this PR; the numbers live in the linked SemiAnalysis runs. SGLang CI was not triggered.

🤖 Generated with Claude Code

---

CI States

Latest PR Test (Base): ✅ Run #38348415549
Latest PR Test (Extra): ❌ Run #38348415375
Latest PR Test (AMD ROCm 10): ➖ No AMD PR run found for this commit.

[nvpohanh]

Verified locally. UI looks good. Force merge to unblock SA AgentX submission.