From 75c59c6d03cb9db51736601db53d7df65c9f933d Mon Sep 17 00:00:00 2001
From: Oseltamivir <58582368+Oseltamivir@users.noreply.github.com>
Date: Fri, 31 Jul 2026 12:24:01 +0800
Subject: [PATCH] chore(amd): update DSR1 MI355X SGLang image
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Update the 8k1k MTP configuration to the July 29 SGLang ROCm image and append the corresponding performance changelog entry.

中文：将 DSR1 MI355X 8k1k MTP 配置更新至 7 月 29 日构建的 SGLang ROCm 镜像，并追加对应的性能变更记录。
---
 configs/amd-master.yaml | 2 +-
 perf-changelog.yaml     | 6 ++++++
 2 files changed, 7 insertions(+), 1 deletion(-)

diff --git a/configs/amd-master.yaml b/configs/amd-master.yaml
index b8f6f4fc31..41288c4849 100644
--- a/configs/amd-master.yaml
+++ b/configs/amd-master.yaml
@@ -983,7 +983,7 @@ dsr1-fp4-mi355x-sglang-disagg:
           - "DECODE_MTP_SIZE=0"
 
 dsr1-fp4-mi355x-sglang-disagg-8k1k-mtp:
-  image: lmsysorg/sglang-rocm:v0.5.12.post1-rocm720-mi35x-20260529
+  image: lmsysorg/sglang-rocm:v0.5.16-rocm720-mi35x-20260729
   model: amd/DeepSeek-R1-0528-MXFP4-v2
   model-prefix: dsr1
   runner: mi355x-disagg
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 6eebd97070..d9bc37e6f0 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -5355,3 +5355,9 @@
     - "Apply the accuracy-gated Kimi-K2.5 MXFP4 settings: tuned AITER MXFP4 MoE, fused shared experts, FP8 KV cache, block size 16, 16384 batched tokens, 512 sequences, async scheduling, gpu-memory-utilization 0.85 (headroom for CUDA-graph capture on MI355X), and the AITER BF16 GEMM path"
     - "Extend the TP4 and TP8 8k1k concurrency sweep from 64 to 128 (1k1k deprecated per #2263)"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2213
+
+- config-keys:
+    - dsr1-fp4-mi355x-sglang-disagg-8k1k-mtp
+  description:
+    - "Update the SGLang ROCm image to v0.5.16-rocm720-mi35x-20260729, which includes the AITER MLA LDS synchronization fix."
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2435