[NV] Add Qwen3.5 FP4 GB300 TRT-LLM Disaggregated AgentX / 新增 Qwen3.5 FP4 GB300 TRT-LLM 分离式 AgentX

[richardhuo-nv]

Add Qwen3.5 FP4 GB300 TRT-LLM Disaggregated AgentX / 新增 Qwen3.5 FP4 GB300 TRT-LLM 分离式 AgentX

Summary / 摘要

Registers nine disaggregated Dynamo-TRT agentic benchmark recipes for Qwen3.5-397B-A17B-NVFP4-V2 on GB300, covering a range of topologies from single-node small-scale to large multi-node high-throughput.

新增九个 Qwen3.5-397B-A17B-NVFP4-V2 GB300 分离式 Dynamo-TRT AgentX Benchmark 配置，覆盖从单节点小规模到大型多节点高吞吐量的一系列拓扑结构。

Changes / 变更内容

configs/nvidia-master.yaml

• New config key qwen3.5-fp4-gb300-dynamo-trt-agentic-disagg with scenario-type: agentic-coding, image nvcr.io/nvidia/tensorrt-llm/release:1.3.0rc24, model nvidia/Qwen3.5-397B-A17B-NVFP4-V2, Dynamo router 1.4.0.dev20260807, and NIXL KV transfer.
• Nine search-space entries with MTP speculative decoding, covering tep (attention-DP disabled) and dep (attention-DP enabled) topology families:

Recipe	Prefill	Decode	conc
1p1d-tep2-tep4	1×TP2/EP2	1×TP4/EP4	8
2p3d-tep2-tep4	2×TP2/EP2	3×TP4/EP4	6, 24
2p3d-tep2-tep8	2×TP2/EP2	3×TP8/EP8	6
4p1d-dep4-dep8	4×TP4/EP4	1×TP8/EP8	896
15p1d-dep1-dep16	15×TP1/EP1	1×TP16/EP16	606
17p3d-dep2-dep8	17×TP2/EP2	3×TP8/EP8	1452
19p1d-dep1-dep16	19×TP1/EP1	1×TP16/EP16	225, 449

benchmarks/multi_node/srt-slurm-recipes/trtllm/qwen3.5/gb300-fp4/disagg/agentx/

• Nine recipe YAMLs added, sourced from srt-slurm/recipes/trtllm/qwen3.5-fp4/agentx/submission/.

runners/launch_gb300-nv.sh

• New branch for IS_AGENTIC=1 && dynamo-trt && qwen3.5: clones NVIDIA/srt-slurm at v1.0.48 and overlays the agentx recipe directory.

perf-changelog.yaml

• New entry for qwen3.5-fp4-gb300-dynamo-trt-agentic-disagg.

Validation / 验证

Matrix generator produces all 9 expected agentic-coding rows:

qwen3.5_p1x2ep2_d1x4ep4_conc8
qwen3.5_p2x2ep2_d3x4ep4_conc6
qwen3.5_p2x2ep2_d3x8ep8_conc6
qwen3.5_p2x2ep2_d3x4ep4_conc24
qwen3.5_p4x4ep4dpa_d1x8ep8dpa_conc896
qwen3.5_p15x1dpa_d1x16ep16dpa_conc606
qwen3.5_p17x2ep2dpa_d3x8ep8dpa_conc1452
qwen3.5_p19x1dpa_d1x16ep16dpa_conc225
qwen3.5_p19x1dpa_d1x16ep16dpa_conc449

[github-actions]

Thanks for the contribution! Please reach out to respective companies' CODEOWNER to fill in the latest PR_REVIEW_CHECKLIST.md before pinging core maintainer on Slack for review. In order for the signoff PR check bot to trigger, you must follow the PR_REVIEW_CHECKLIST.md template correctly, including the phrase As a PR reviewer and CODEOWNER, I have reviewed this and have.

For PR verification, add the full-sweep-fail-fast label (strongly recommended) to this PR — the benchmark sweep only runs on labeled PRs. Use full-sweep-enabled only if you need matrix jobs to keep running past a failure.

PR authors are responsible for ensuring that after merging, all GitHub Action jobs fully pass. A lot of the time, failures are just flakes and simply re-running the failed jobs will fix it. See GitHub's docs on re-running failed jobs

---

感谢你的贡献！请联系相应公司的 CODEOWNER 填写最新的 PR_REVIEW_CHECKLIST.md，然后再在 Slack 上联系核心维护者进行审阅。为了触发 signoff PR 检查机器人，你必须正确遵循 PR_REVIEW_CHECKLIST.md 模板，包括保留英文语句 As a PR reviewer and CODEOWNER, I have reviewed this and have。

如需进行 PR 验证，请为此 PR 添加 full-sweep-fail-fast 标签（强烈推荐）— 基准测试 sweep 仅在带有标签的 PR 上运行。仅当需要矩阵任务在失败后继续运行时才使用 full-sweep-enabled。

PR 作者有责任确保合并后所有 GitHub Action 任务完全通过。 很多时候失败只是偶发抖动（flake），重新运行失败的任务即可解决。参见 GitHub 关于重新运行失败任务的文档

[github-actions]

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=31648775819
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=31648775819

[github-actions]

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=31648851366
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=31648851366