From 48c21404b92b17e80066da747ea92273dee402ce Mon Sep 17 00:00:00 2001 From: Wenyao Gao Date: Thu, 22 Oct 2026 18:43:35 -0700 Subject: [PATCH 1/2] chore: extend Qwen3.5 FP8 H200 SGLang 8k1k to c1-c128 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit 中文：将 Qwen3.5 FP8 H200 SGLang 8k1k 并发范围扩展至 c1-c128。 --- .../srt-slurm-recipes/qwen3.5/sglang/h200-fp8/8k1k.yaml | 4 ++-- inferencex-e2e/configs/nvidia-master.yaml | 4 ++-- inferencex-e2e/perf-changelog.yaml | 6 ++++++ 3 files changed, 10 insertions(+), 4 deletions(-) diff --git a/inferencex-e2e/benchmarks/single_node/srt-slurm-recipes/qwen3.5/sglang/h200-fp8/8k1k.yaml b/inferencex-e2e/benchmarks/single_node/srt-slurm-recipes/qwen3.5/sglang/h200-fp8/8k1k.yaml index 5c8b24c9d0..0b7c3761b9 100644 --- a/inferencex-e2e/benchmarks/single_node/srt-slurm-recipes/qwen3.5/sglang/h200-fp8/8k1k.yaml +++ b/inferencex-e2e/benchmarks/single_node/srt-slurm-recipes/qwen3.5/sglang/h200-fp8/8k1k.yaml @@ -44,7 +44,7 @@ zip_override_concurrency: roles: agg: args: - cuda-graph-max-bs: [4, 8, 16, 32, 64] + cuda-graph-max-bs: [1, 2, 4, 8, 16, 32, 64, 128] benchmark: env: - CONC: ['4', '8', '16', '32', '64'] + CONC: ['1', '2', '4', '8', '16', '32', '64', '128'] diff --git a/inferencex-e2e/configs/nvidia-master.yaml b/inferencex-e2e/configs/nvidia-master.yaml index ceca84b4d9..5a4733aad7 100644 --- a/inferencex-e2e/configs/nvidia-master.yaml +++ b/inferencex-e2e/configs/nvidia-master.yaml @@ -1677,8 +1677,8 @@ qwen3.5-fp8-h200-sglang: search-space: - tp: 8 ep: 8 - conc-start: 4 - conc-end: 64 + conc-start: 1 + conc-end: 128 srt-recipe: 8k1k.yaml qwen3.5-fp8-h200-sglang-mtp: diff --git a/inferencex-e2e/perf-changelog.yaml b/inferencex-e2e/perf-changelog.yaml index a9ee808acb..8e53d7edaa 100644 --- a/inferencex-e2e/perf-changelog.yaml +++ b/inferencex-e2e/perf-changelog.yaml @@ -10871,3 +10871,9 @@ - "Update the vLLM ROCm image from v0.31.0 to v0.32.0." pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/4395 scenario-type: [agentic-coding] + +- config-keys: + - qwen3.5-fp8-h200-sglang + description: + - "Extend TP8/EP8 8k1k concurrency from 4-64 to 1-128 (cuda-graph-max-bs follows concurrency)." + pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/XXX From b7ad5ed31f850f59abdc8538bc030804ff8cf436 Mon Sep 17 00:00:00 2001 From: Wenyao Gao Date: Thu, 22 Oct 2026 18:44:38 -0700 Subject: [PATCH 2/2] chore: set H200 changelog PR link MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit 中文：填写 H200 changelog 的 PR 链接。 --- inferencex-e2e/perf-changelog.yaml | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/inferencex-e2e/perf-changelog.yaml b/inferencex-e2e/perf-changelog.yaml index 8e53d7edaa..dbb276b4bb 100644 --- a/inferencex-e2e/perf-changelog.yaml +++ b/inferencex-e2e/perf-changelog.yaml @@ -10876,4 +10876,4 @@ - qwen3.5-fp8-h200-sglang description: - "Extend TP8/EP8 8k1k concurrency from 4-64 to 1-128 (cuda-graph-max-bs follows concurrency)." - pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/XXX + pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/4415