From fc1716efa4d79436b6e07eddd5cf6f72c4ba3150 Mon Sep 17 00:00:00 2001 From: yhyang201 Date: Sun, 26 Apr 2026 23:50:24 +0800 Subject: [PATCH] dsv4-b300-sglang: better performance for dp-attention Co-Authored-By: Claude Opus 4.6 --- perf-changelog.yaml | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/perf-changelog.yaml b/perf-changelog.yaml index 4c85924b4..e4c46268e 100644 --- a/perf-changelog.yaml +++ b/perf-changelog.yaml @@ -1887,3 +1887,10 @@ - "Three CONC bands: A=TP8 (1-8), B=TP4 (16-128), C=DP4 dp-attn (64-512); B/C overlap at conc 64,128" - "Configs: 1k1k and 8k1k, no validation.py / launcher / yaml-field changes (knob-free)" pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1166 + +- config-keys: + - dsv4-fp4-b300-sglang + description: + - "better performance for dp-attention" + - "Recipes from https://docs.sglang.io/cookbook/autoregressive/DeepSeek/DeepSeek-V4" + pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1178