From 953d8d6203b86246690c0d495987eed64e369aa4 Mon Sep 17 00:00:00 2001 From: "claude[bot]" <41898282+claude[bot]@users.noreply.github.com> Date: Sun, 11 Oct 2026 18:31:08 +0000 Subject: [PATCH 1/2] chore(dsv4): update B200 vLLM AgentX DSpark image to nightly 2eb491e MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Pin upstream vllm/vllm-openai nightly 2eb491e by digest. Upstream lacks the fork-only deep_gemm_amxf4_mega_moe backend and VLLM_DSV4_MEGA_FP8_COMBINE, so the DEP8 arm uses the upstream deep_gemm_mega_moe backend. 将 B200 vLLM AgentX DSpark 镜像更新为上游 nightly 2eb491e 并按摘要固定。上游没有 fork 专有的 deep_gemm_amxf4_mega_moe 后端和 VLLM_DSV4_MEGA_FP8_COMBINE，因此 DEP8 改用上游 deep_gemm_mega_moe 后端。 Co-Authored-By: Claude Opus 5.5 --- .../dsv4/vllm/b200-fp4-mtp/agentic.yaml | 15 +++++++-------- inferencex-e2e/configs/nvidia-master.yaml | 2 +- 2 files changed, 8 insertions(+), 9 deletions(-) diff --git a/inferencex-e2e/benchmarks/single_node/srt-slurm-recipes/dsv4/vllm/b200-fp4-mtp/agentic.yaml b/inferencex-e2e/benchmarks/single_node/srt-slurm-recipes/dsv4/vllm/b200-fp4-mtp/agentic.yaml index 44c6285845..83841e6fe2 100644 --- a/inferencex-e2e/benchmarks/single_node/srt-slurm-recipes/dsv4/vllm/b200-fp4-mtp/agentic.yaml +++ b/inferencex-e2e/benchmarks/single_node/srt-slurm-recipes/dsv4/vllm/b200-fp4-mtp/agentic.yaml @@ -6,7 +6,7 @@ base: name: dsv4-fp4-b200-vllm-agentic model: path: hf:deepseek-ai/DeepSeek-V4-Pro-0813 - container: vllm/vllm-openai:nightly-dev-x86_64-cu13.0.1-591bb95 + container: vllm/vllm-openai:nightly-2eb491e18b1ce5b8a25e4c88f32dbb5d028d525c@sha256:514cd42702d29a7b8efe20b0a916523ac473ecebdec8d53b1a0d0f5439a48634 precision: fp4 resources: gpu_type: b200 @@ -53,7 +53,6 @@ base: VLLM_PREFIX_CACHE_RETENTION_INTERVAL: '32768' VLLM_USE_V2_MODEL_RUNNER: '1' VLLM_USE_RUST_FRONTEND: '1' - VLLM_DSV4_MEGA_FP8_COMBINE: '1' VLLM_RPC_TIMEOUT: '600000' TORCH_CUDA_ARCH_LIST: '10.0' PYTHONNOUSERSITE: '1' @@ -187,7 +186,7 @@ override_dep8_c32: data-parallel-size: 8 enable-expert-parallel: true enable-ep-weight-filter: true - moe-backend: deep_gemm_amxf4_mega_moe + moe-backend: deep_gemm_mega_moe prefill-schedule-interval: 16 long-prefill-token-threshold: 512 kv-cache-memory-bytes: 37580963840 @@ -223,7 +222,7 @@ override_dep8_c64: data-parallel-size: 8 enable-expert-parallel: true enable-ep-weight-filter: true - moe-backend: deep_gemm_amxf4_mega_moe + moe-backend: deep_gemm_mega_moe prefill-schedule-interval: 16 long-prefill-token-threshold: 512 kv-cache-memory-bytes: 37580963840 @@ -259,7 +258,7 @@ override_dep8_c96: data-parallel-size: 8 enable-expert-parallel: true enable-ep-weight-filter: true - moe-backend: deep_gemm_amxf4_mega_moe + moe-backend: deep_gemm_mega_moe prefill-schedule-interval: 16 long-prefill-token-threshold: 512 kv-cache-memory-bytes: 37580963840 @@ -295,7 +294,7 @@ override_dep8_c128: data-parallel-size: 8 enable-expert-parallel: true enable-ep-weight-filter: true - moe-backend: deep_gemm_amxf4_mega_moe + moe-backend: deep_gemm_mega_moe prefill-schedule-interval: 16 long-prefill-token-threshold: 512 kv-cache-memory-bytes: 37580963840 @@ -331,7 +330,7 @@ override_dep8_c160: data-parallel-size: 8 enable-expert-parallel: true enable-ep-weight-filter: true - moe-backend: deep_gemm_amxf4_mega_moe + moe-backend: deep_gemm_mega_moe prefill-schedule-interval: 16 long-prefill-token-threshold: 512 kv-cache-memory-bytes: 36507222016 @@ -367,7 +366,7 @@ override_dep8_c192: data-parallel-size: 8 enable-expert-parallel: true enable-ep-weight-filter: true - moe-backend: deep_gemm_amxf4_mega_moe + moe-backend: deep_gemm_mega_moe prefill-schedule-interval: 16 long-prefill-token-threshold: 512 kv-cache-memory-bytes: 36507222016 diff --git a/inferencex-e2e/configs/nvidia-master.yaml b/inferencex-e2e/configs/nvidia-master.yaml index 12c4cc160e..e326c4183c 100644 --- a/inferencex-e2e/configs/nvidia-master.yaml +++ b/inferencex-e2e/configs/nvidia-master.yaml @@ -936,7 +936,7 @@ dsv4-fp4-b200-sglang-agentic-hicache-mtp: - { tp: 8, ep: 8, dp-attn: true, kv-offloading: dram, kv-offload-backend: { name: hicache }, spec-decoding: draft_model, conc-list: [64, 96, 128, 160], router: { name: sglang-router, version: "0.3.2" }, srt-recipe: benchmarks/single_node/srt-slurm-recipes/dsv4/sglang/b200-fp4-mtp/agentic.yaml } dsv4-fp4-b200-vllm-agentic-mtp: - image: vllm/vllm-openai:nightly-dev-x86_64-cu13.0.1-591bb95 + image: vllm/vllm-openai:nightly-2eb491e18b1ce5b8a25e4c88f32dbb5d028d525c@sha256:514cd42702d29a7b8efe20b0a916523ac473ecebdec8d53b1a0d0f5439a48634 model: deepseek-ai/DeepSeek-V4-Pro-0813 model-prefix: dsv4 runner: cluster:b200-nscale From 4fac5a77466034c27117c22a2c54cc1fc6b27a3a Mon Sep 17 00:00:00 2001 From: "claude[bot]" <41898282+claude[bot]@users.noreply.github.com> Date: Sun, 11 Oct 2026 20:29:06 +0000 Subject: [PATCH 2/2] chore(dsv4): add perf-changelog entry for B200 vLLM AgentX image update MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit 为 B200 vLLM AgentX 镜像更新添加 perf-changelog 条目。 Co-Authored-By: Claude Opus 5.5 --- inferencex-e2e/perf-changelog.yaml | 6 ++++++ 1 file changed, 6 insertions(+) diff --git a/inferencex-e2e/perf-changelog.yaml b/inferencex-e2e/perf-changelog.yaml index 2334965697..71dbcaf404 100644 --- a/inferencex-e2e/perf-changelog.yaml +++ b/inferencex-e2e/perf-changelog.yaml @@ -9538,3 +9538,9 @@ - "Run prefill context parallelism on the TP-only arms at conc 16, 32 and 48; conc 1 and 4 keep BCG." - "Enable SGLANG_OPT_FP8_WO_A_FUSED_INVROPE." pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/3778 + +- config-keys: + - dsv4-fp4-b200-vllm-agentic-mtp + description: + - "Update the vLLM image to upstream nightly 2eb491e and switch DEP8 to the deep_gemm_mega_moe backend." + pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/3917