From fcb8706a6ab1ce5567d2d3f240b912b2dea3105a Mon Sep 17 00:00:00 2001 From: Wenyao Gao Date: Thu, 22 Oct 2026 18:31:03 -0700 Subject: [PATCH 1/3] chore: update Qwen3.5 FP8 B300 SGLang to v0.5.22 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit 中文：将 Qwen3.5 FP8 B300 SGLang 镜像更新至 v0.5.22。 --- inferencex-e2e/configs/nvidia-master.yaml | 2 +- inferencex-e2e/perf-changelog.yaml | 6 ++++++ 2 files changed, 7 insertions(+), 1 deletion(-) diff --git a/inferencex-e2e/configs/nvidia-master.yaml b/inferencex-e2e/configs/nvidia-master.yaml index ceca84b4d9..20b6e286c8 100644 --- a/inferencex-e2e/configs/nvidia-master.yaml +++ b/inferencex-e2e/configs/nvidia-master.yaml @@ -1281,7 +1281,7 @@ qwen3.5-fp8-b300-sglang-mtp: srt-recipe: 8k1k.yaml qwen3.5-fp8-b300-sglang: - image: lmsysorg/sglang:v0.5.21-cu130@sha256:b1259f3ea3275f66237c498ea388919729018bc9f01c3d638391e06e2cf3f469 + image: lmsysorg/sglang:v0.5.22-cu130@sha256:e3a4f06001a29542fc4a8638536d9ffc5eb9c9942b08603c09b19a56eae4a1ae model: Qwen/Qwen3.5-397B-A17B-FP8 model-prefix: qwen3.5 runner: b300 diff --git a/inferencex-e2e/perf-changelog.yaml b/inferencex-e2e/perf-changelog.yaml index a9ee808acb..fa0ca55763 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-b300-sglang + description: + - "Update SGLang image from v0.5.21-cu130 to v0.5.22-cu130." + pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/4413 From 66c3071e23cad6578d991bab317feab077f94aa6 Mon Sep 17 00:00:00 2001 From: Wenyao Gao Date: Thu, 22 Oct 2026 19:01:56 -0700 Subject: [PATCH 2/3] chore: start Qwen3.5 FP8 B300 SGLang TP4 at c1 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit 中文：让 Qwen3.5 FP8 B300 SGLang 的 TP4 从并发 1 开始。 --- .../srt-slurm-recipes/qwen3.5/sglang/b300-fp8/8k1k.yaml | 6 +++--- inferencex-e2e/configs/nvidia-master.yaml | 2 +- inferencex-e2e/perf-changelog.yaml | 1 + 3 files changed, 5 insertions(+), 4 deletions(-) diff --git a/inferencex-e2e/benchmarks/single_node/srt-slurm-recipes/qwen3.5/sglang/b300-fp8/8k1k.yaml b/inferencex-e2e/benchmarks/single_node/srt-slurm-recipes/qwen3.5/sglang/b300-fp8/8k1k.yaml index 15951f8058..196025dca8 100644 --- a/inferencex-e2e/benchmarks/single_node/srt-slurm-recipes/qwen3.5/sglang/b300-fp8/8k1k.yaml +++ b/inferencex-e2e/benchmarks/single_node/srt-slurm-recipes/qwen3.5/sglang/b300-fp8/8k1k.yaml @@ -48,8 +48,8 @@ zip_override_concurrency: roles: agg: args: - cuda-graph-max-bs-decode: [4, 8, 16, 32, 64, 128, 256] - max-running-requests: [4, 8, 16, 32, 64, 128, 256] + cuda-graph-max-bs-decode: [1, 2, 4, 8, 16, 32, 64, 128, 256] + max-running-requests: [1, 2, 4, 8, 16, 32, 64, 128, 256] benchmark: env: - CONC: ['4', '8', '16', '32', '64', '128', '256'] + CONC: ['1', '2', '4', '8', '16', '32', '64', '128', '256'] diff --git a/inferencex-e2e/configs/nvidia-master.yaml b/inferencex-e2e/configs/nvidia-master.yaml index 20b6e286c8..52d6a0f122 100644 --- a/inferencex-e2e/configs/nvidia-master.yaml +++ b/inferencex-e2e/configs/nvidia-master.yaml @@ -1296,7 +1296,7 @@ qwen3.5-fp8-b300-sglang: search-space: - tp: 4 ep: 1 - conc-start: 4 + conc-start: 1 conc-end: 256 srt-recipe: 8k1k.yaml diff --git a/inferencex-e2e/perf-changelog.yaml b/inferencex-e2e/perf-changelog.yaml index fa0ca55763..e565d51231 100644 --- a/inferencex-e2e/perf-changelog.yaml +++ b/inferencex-e2e/perf-changelog.yaml @@ -10876,4 +10876,5 @@ - qwen3.5-fp8-b300-sglang description: - "Update SGLang image from v0.5.21-cu130 to v0.5.22-cu130." + - "Start TP4/EP1 at concurrency 1 (cuda-graph-max-bs-decode and max-running-requests follow concurrency)." pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/4413 From cffb6cd12c05ae7a8e70e05c7595920c152ed719 Mon Sep 17 00:00:00 2001 From: Wenyao Gao Date: Fri, 6 Nov 2026 11:16:05 -0800 Subject: [PATCH 3/3] chore: start Qwen3.5 FP8 B300 AgentX power sweep