From e9bee3539eb3a4f1e7bcbb62394a234bbc770a5a Mon Sep 17 00:00:00 2001 From: Yifan Qiao Date: Wed, 19 Aug 2026 10:23:09 +0000 Subject: [PATCH 1/3] perf(gb300): add AgentX aggregate TP4 points MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit 为 GB300 AgentX 聚合式 TP4 扫描新增并发 1、2、4 和 6，同时保留现有并发 8 数据点。 --- configs/nvidia-master.yaml | 2 +- perf-changelog.yaml | 9 +++++++++ 2 files changed, 10 insertions(+), 1 deletion(-) diff --git a/configs/nvidia-master.yaml b/configs/nvidia-master.yaml index 238aaf320f..731bb63b79 100644 --- a/configs/nvidia-master.yaml +++ b/configs/nvidia-master.yaml @@ -7787,7 +7787,7 @@ dsv4-fp4-gb300-dynamo-vllm-agentic-mtp-agg: ep: 1 dp-attn: false - spec-decoding: mtp - conc-list: [8] + conc-list: [1, 2, 4, 6, 8] prefill: num-worker: 1 tp: 4 diff --git a/perf-changelog.yaml b/perf-changelog.yaml index 52c74bf168..b0a67a950f 100644 --- a/perf-changelog.yaml +++ b/perf-changelog.yaml @@ -6195,3 +6195,12 @@ - "Use AITER INT4 quick-reduce, ptpc_fp8 online quantization excluding embeddings, lm_head, gates, and experts, an FP8 KV cache, and three-token MTP with golden synthetic acceptance length 2.99 for benchmark runs." - "Set max-num-seqs to twice the concurrency, select CUDA graph capture sizes by concurrency, and cap max-num-batched-tokens at 16384." pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2576 + +- config-keys: + - dsv4-fp4-gb300-dynamo-vllm-agentic-mtp-agg + scenario-type: + - agentic-coding + description: + - "Expand the GB300 AgentX aggregate TP4 sweep to concurrency 1, 2, 4, 6, and 8." + pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2676 + append-only: true From 3b29f667ceb2d3987bfb9b07952e795d9232dd81 Mon Sep 17 00:00:00 2001 From: Cameron Quilici Date: Fri, 21 Aug 2026 09:53:36 -0500 Subject: [PATCH 2/3] Update perf-changelog.yaml --- perf-changelog.yaml | 1 + 1 file changed, 1 insertion(+) diff --git a/perf-changelog.yaml b/perf-changelog.yaml index b0a67a950f..7d1f4b75fd 100644 --- a/perf-changelog.yaml +++ b/perf-changelog.yaml @@ -6204,3 +6204,4 @@ - "Expand the GB300 AgentX aggregate TP4 sweep to concurrency 1, 2, 4, 6, and 8." pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2676 append-only: true + From 1cad75f9c6626d4abe3024a49f51487466ec7b3a Mon Sep 17 00:00:00 2001 From: Cameron Quilici Date: Fri, 21 Aug 2026 09:54:58 -0500 Subject: [PATCH 3/3] Update perf-changelog.yaml --- perf-changelog.yaml | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/perf-changelog.yaml b/perf-changelog.yaml index c2483663e5..cfb0b4daa0 100644 --- a/perf-changelog.yaml +++ b/perf-changelog.yaml @@ -6324,4 +6324,4 @@ description: - "Expand the GB300 AgentX aggregate TP4 sweep to concurrency 1, 2, 4, 6, and 8." pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2676 - \ No newline at end of file +