[Klaud Cold] Update minimaxm3-fp8-mi300x-vllm-agentic-mtp vLLM ROCm image to v0.29.0 / 将 minimaxm3-fp8-mi300x-vllm-agentic-mtp 的 vLLM ROCm 镜像更新至 v0.29.0

configs/amd-master.yaml

@@ -1529,7 +1529,7 @@ dsv4-fp4-mi355x-atom-disagg:
           - "DECODE_NODES=1"
       # 1P1D TP8
 minimaxm3-fp8-mi300x-vllm-agentic-mtp:
-  image: vllm/vllm-openai-rocm:nightly-d9105ea8001e0a6d77a96327d17515bb5791fb36
+  image: vllm/vllm-openai-rocm:v0.29.0
   model: MiniMaxAI/MiniMax-M3-MXFP8
   model-prefix: minimaxm3
   runner: cluster:mi300x-amd

perf-changelog.yaml

@@ -7441,3 +7441,10 @@
     - "Replace the removed decode-only alias --cuda-graph-max-bs with its canonical name --cuda-graph-max-bs-decode, retaining the same min(2*CONC, 64) cap. The 20260910 image includes sgl-project/sglang#38375, so argparse interprets the removed spelling as an ambiguous prefix of the decode and prefill options and exits before serving. Leave --cuda-graph-max-bs-prefill unset: it controls a separate prefill token-capture ceiling, not request concurrency."
     - "Switch DSA prefill and decode from tilelang to triton (sgl-project/sglang#30575 Fast Triton Sparse MLA). The 20260910 image already contains that backend; kv-cache-dtype stays fp8_e4m3, which the Triton DSA path requires."
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2900
+
+- config-keys:
+    - minimaxm3-fp8-mi300x-vllm-agentic-mtp
+  description:
+    - "Update vLLM ROCm image from vllm/vllm-openai-rocm:nightly-d9105ea8001e0a6d77a96327d17515bb5791fb36 (2026-09-07 upstream ROCm nightly, tag commit vllm-project/vllm@d9105ea8) to the v0.29.0 release image vllm/vllm-openai-rocm:v0.29.0 (digest sha256:e5e47f6aaab675c252c381f0dac237b31b10d87bb74d092b07fb4065efd7f5a1, Docker Hub last pushed 2026-09-09T05:41:55Z, release tag commit vllm-project/vllm@98dff2a8; the release branch diverged from main on 2026-08-31 and does not include the ROCm MiniMax-M3 decode indexer/top-k optimization vllm-project/vllm#54682 or the AITER v0.1.21.post1 base bump vllm-project/vllm#52826). benchmarks/single_node/agentic/minimaxm3_fp8_mi300x_mtp.sh, the search space and the default evals are unchanged."
+    - "将 vLLM ROCm 镜像从 vllm/vllm-openai-rocm:nightly-d9105ea8001e0a6d77a96327d17515bb5791fb36（2026-09-07 上游 ROCm nightly，标签提交 vllm-project/vllm@d9105ea8）更新为 v0.29.0 发布镜像 vllm/vllm-openai-rocm:v0.29.0（digest sha256:e5e47f6aaab675c252c381f0dac237b31b10d87bb74d092b07fb4065efd7f5a1，Docker Hub 最后推送时间 2026-09-09T05:41:55Z，发布标签提交 vllm-project/vllm@98dff2a8；该发布分支于 2026-08-31 从 main 分出，不含 ROCm MiniMax-M3 解码索引/top-k 优化 vllm-project/vllm#54682 及 AITER v0.1.21.post1 基础镜像升级 vllm-project/vllm#52826）。benchmarks/single_node/agentic/minimaxm3_fp8_mi300x_mtp.sh、搜索空间与默认评测保持不变。"
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/3063