Refresh DeepSeek-V4-Pro GB200 Dynamo vLLM stack / 更新 DeepSeek-V4-Pro GB200 Dynamo vLLM 推理栈 (#2139)

* perf(dsv4): refresh GB200 Dynamo vLLM stack

Update the vLLM image and Dynamo wheel, and enable multiple frontends across all five GB200 DeepSeek-V4-Pro topologies.

中文：更新 vLLM 镜像和 Dynamo wheel，并为 DeepSeek-V4-Pro 的全部五种 GB200 拓扑启用多个前端。

* chore: link performance changelog to PR #2139

Backfill the canonical pull request URL after creating the draft PR.

中文：创建草稿 PR 后，将正式 PR 链接回填至性能变更日志。

* perf(dsv4): use MessagePack for Dynamo request plane

Set DYN_REQUEST_PLANE_CODEC=msgpack on every GB200 Dynamo frontend, prefill worker, and decode worker.

中文：为所有 GB200 Dynamo 前端、预填充 worker 和解码 worker 设置 DYN_REQUEST_PLANE_CODEC=msgpack。

* fix(dsv4): reduce GB200 1P4D decode memory

Lower decode gpu-memory-utilization from 0.9 to 0.85 for the GB200 low-middle 1P4D TP8 recipe after the PR sweep OOM.

中文：针对 PR 扫描中出现的显存不足问题，将 GB200 低中段 1P4D TP8 配置的解码 gpu-memory-utilization 从 0.9 降至 0.85。

* Use MXFP4 weights for PR 2139 GB200 vLLM recipes (#2185)

Add a dedicated MXFP4 model-path alias for the five standard GB200 Dynamo-vLLM recipes while preserving the existing NVFP4 mapping for all other configurations.

* Align PR 2139 GB200 recipe identity versions (#2209)

Update the five standard GB200 Dynamo-vLLM recipe identities to the active container, Dynamo 1.3.0.dev1, and vLLM 0.1.dev18219+g4ba0a72a4.

* Use DeepGEMM-timeout vLLM setup for GB200 DSV4 (#2248)

* dsv4-fp4-gb200-dynamo-vllm: vLLM 0.25.1 container, remove setup_script (drop DeepGEMM-timeout patch)

---------

Co-authored-by: jthomson04 <jwillthomson19@gmail.com>

Author: Ankur-singh

benchmarks/multi_node/srt-slurm-recipes/vllm/deepseek-v4/8k1k/disagg-gb200-high-tpt-megamoe.yaml

@@ -8,24 +8,22 @@ name: "svf-vllm-disagg-gb200-high-tpt-megamoe"
 # 4096 with no CPU/NVMe offload.
 #
 # Local deltas vs upstream:
-#   * model.path alias renamed deepseekv4-fp4 -> deepseek-v4-pro to match
-#     SRT_SLURM_MODEL_PREFIX in runners/launch_gb200-nv.sh.
-#   * model.container set to vllm/vllm-openai:v0.20.0-ubuntu2404 to
+#   * model.path uses the deepseek-v4-pro-mxfp4 alias from
+#     runners/launch_gb200-nv.sh.
+#   * model.container set to vllm/vllm-openai:v0.25.1 to
 #     match nvidia-master.yaml image (which the launch script registers as
 #     the alias key in srtslurm.yaml). Upstream variants ship either the
 #     non-dynamo floating tag or a sha256 pin.
 #   * slurm.time_limit + health_check set to 8h / 1440 attempts to
 #     absorb cold-cache /mnt/numa1 model loads.
 model:
-  path: "deepseek-v4-pro"
-  container: "vllm/vllm-openai:v0.20.0-ubuntu2404"
+  path: "deepseek-v4-pro-mxfp4"
+  container: "vllm/vllm-openai:v0.25.1"
   precision: "fp4"
 
 dynamo:
   install: true
-  wheel: "1.2.0.dev20260426"
-
-setup_script: vllm-container-deps.sh
+  wheel: "1.3.0.dev1"
 
 slurm:
   time_limit: "8:00:00"
@@ -48,11 +46,14 @@ infra:
 
 frontend:
   type: dynamo
-  enable_multiple_frontends: false
+  enable_multiple_frontends: true
+  env:
+    DYN_REQUEST_PLANE_CODEC: "msgpack"
 backend:
   type: vllm
   connector: null
   prefill_environment:
+    DYN_REQUEST_PLANE_CODEC: "msgpack"
     TILELANG_CLEANUP_TEMP_FILES: "1"
     VLLM_USE_NCCL_SYMM_MEM: "1"
     TORCH_SYMMMEM: "NVSHMEM"
@@ -70,6 +71,7 @@ backend:
     UCX_CUDA_IPC_ENABLE_MNNVL: "y"
     NCCL_P2P_LEVEL: NVL
   decode_environment:
+    DYN_REQUEST_PLANE_CODEC: "msgpack"
     TILELANG_CLEANUP_TEMP_FILES: "1"
     VLLM_USE_NCCL_SYMM_MEM: "1"
     TORCH_SYMMMEM: "NVSHMEM"
@@ -147,9 +149,9 @@ benchmark:
 identity:
   model:
     repo: "deepseek-ai/DeepSeek-V4-Pro"
-    revision: "0366e4e064385807ea86b088a5c6c878ff23343b"
+    revision: "b5968e9190ef611bbf34a7229255be88a0e937c1"
   container:
-    image: "vllm/vllm-openai:v0.20.0-ubuntu2404"
+    image: "vllm/vllm-openai:v0.25.1"
   frameworks:
-    dynamo: "1.2.0.dev20260426"
-    vllm: "0.20.0"
+    dynamo: "1.3.0.dev1"
+    vllm: "0.25.1"

benchmarks/multi_node/srt-slurm-recipes/vllm/deepseek-v4/8k1k/disagg-gb200-low-latency.yaml

@@ -7,24 +7,22 @@ name: "svf-vllm-disagg-gb200-low-latency"
 # dedicated NATS/etcd infra node. Single-concurrency point for low latency.
 #
 # Local deltas vs upstream:
-#   * model.path alias renamed deepseekv4-fp4 -> deepseek-v4-pro to match
-#     SRT_SLURM_MODEL_PREFIX in runners/launch_gb200-nv.sh.
-#   * model.container set to vllm/vllm-openai:v0.20.0-ubuntu2404 to
+#   * model.path uses the deepseek-v4-pro-mxfp4 alias from
+#     runners/launch_gb200-nv.sh.
+#   * model.container set to vllm/vllm-openai:v0.25.1 to
 #     match nvidia-master.yaml image (which the launch script registers as
 #     the alias key in srtslurm.yaml). Upstream variants ship either the
 #     non-dynamo floating tag or a sha256 pin.
 #   * slurm.time_limit + health_check set to 8h / 1440 attempts to
 #     absorb cold-cache /mnt/numa1 model loads.
 model:
-  path: "deepseek-v4-pro"
-  container: "vllm/vllm-openai:v0.20.0-ubuntu2404"
+  path: "deepseek-v4-pro-mxfp4"
+  container: "vllm/vllm-openai:v0.25.1"
   precision: "fp4"
 
 dynamo:
   install: true
-  wheel: "1.2.0.dev20260426"
-
-setup_script: vllm-container-deps.sh
+  wheel: "1.3.0.dev1"
 
 slurm:
   time_limit: "8:00:00"
@@ -47,11 +45,14 @@ infra:
 
 frontend:
   type: dynamo
-  enable_multiple_frontends: false
+  enable_multiple_frontends: true
+  env:
+    DYN_REQUEST_PLANE_CODEC: "msgpack"
 backend:
   type: vllm
   connector: null
   prefill_environment:
+    DYN_REQUEST_PLANE_CODEC: "msgpack"
     TILELANG_CLEANUP_TEMP_FILES: "1"
     VLLM_USE_NCCL_SYMM_MEM: "1"
     NCCL_CUMEM_ENABLE: "1"
@@ -68,6 +69,7 @@ backend:
     UCX_CUDA_IPC_ENABLE_MNNVL: "y"
     NCCL_P2P_LEVEL: NVL
   decode_environment:
+    DYN_REQUEST_PLANE_CODEC: "msgpack"
     TILELANG_CLEANUP_TEMP_FILES: "1"
     VLLM_USE_NCCL_SYMM_MEM: "1"
     NCCL_CUMEM_ENABLE: "1"
@@ -142,8 +144,11 @@ benchmark:
   custom_tokenizer: "sa_bench_tokenizers.vllm_deepseek_v4.VLLMDeepseekV4Tokenizer"
 
 identity:
+  model:
+    repo: "deepseek-ai/DeepSeek-V4-Pro"
+    revision: "b5968e9190ef611bbf34a7229255be88a0e937c1"
   container:
-    image: "vllm/vllm-openai:v0.20.0-ubuntu2404"
+    image: "vllm/vllm-openai:v0.25.1"
   frameworks:
-    dynamo: "1.2.0.dev20260426"
-    vllm: "0.20.0"
+    dynamo: "1.3.0.dev1"
+    vllm: "0.25.1"

benchmarks/multi_node/srt-slurm-recipes/vllm/deepseek-v4/8k1k/disagg-gb200-low-middle-curve.yaml

@@ -8,24 +8,22 @@ name: "svf-vllm-disagg-gb200-low-middle-curve"
 # 256 and 512.
 #
 # Local deltas vs upstream:
-#   * model.path alias renamed deepseekv4-fp4 -> deepseek-v4-pro to match
-#     SRT_SLURM_MODEL_PREFIX in runners/launch_gb200-nv.sh.
-#   * model.container set to vllm/vllm-openai:v0.20.0-ubuntu2404 to
+#   * model.path uses the deepseek-v4-pro-mxfp4 alias from
+#     runners/launch_gb200-nv.sh.
+#   * model.container set to vllm/vllm-openai:v0.25.1 to
 #     match nvidia-master.yaml image (which the launch script registers as
 #     the alias key in srtslurm.yaml). Upstream variants ship either the
 #     non-dynamo floating tag or a sha256 pin.
 #   * slurm.time_limit + health_check set to 8h / 1440 attempts to
 #     absorb cold-cache /mnt/numa1 model loads.
 model:
-  path: "deepseek-v4-pro"
-  container: "vllm/vllm-openai:v0.20.0-ubuntu2404"
+  path: "deepseek-v4-pro-mxfp4"
+  container: "vllm/vllm-openai:v0.25.1"
   precision: "fp4"
 
 dynamo:
   install: true
-  wheel: "1.2.0.dev20260426"
-
-setup_script: vllm-container-deps.sh
+  wheel: "1.3.0.dev1"
 
 slurm:
   time_limit: "8:00:00"
@@ -48,11 +46,14 @@ infra:
 
 frontend:
   type: dynamo
-  enable_multiple_frontends: false
+  enable_multiple_frontends: true
+  env:
+    DYN_REQUEST_PLANE_CODEC: "msgpack"
 backend:
   type: vllm
   connector: null
   prefill_environment:
+    DYN_REQUEST_PLANE_CODEC: "msgpack"
     TILELANG_CLEANUP_TEMP_FILES: "1"
     VLLM_USE_NCCL_SYMM_MEM: "1"
     NCCL_CUMEM_ENABLE: "1"
@@ -69,6 +70,7 @@ backend:
     UCX_CUDA_IPC_ENABLE_MNNVL: "y"
     NCCL_P2P_LEVEL: NVL
   decode_environment:
+    DYN_REQUEST_PLANE_CODEC: "msgpack"
     TILELANG_CLEANUP_TEMP_FILES: "1"
     VLLM_USE_NCCL_SYMM_MEM: "1"
     NCCL_CUMEM_ENABLE: "1"
@@ -129,7 +131,7 @@ backend:
       no-enable-flashinfer-autotune: true
       block-size: 256
       compilation-config: '{"cudagraph_mode":"FULL_DECODE_ONLY","mode":0}'
-      gpu-memory-utilization: 0.9
+      gpu-memory-utilization: 0.85
       stream-interval: 50
       no-disable-hybrid-kv-cache-manager: true
       enable-sleep-mode: true
@@ -144,8 +146,11 @@ benchmark:
   custom_tokenizer: "sa_bench_tokenizers.vllm_deepseek_v4.VLLMDeepseekV4Tokenizer"
 
 identity:
+  model:
+    repo: "deepseek-ai/DeepSeek-V4-Pro"
+    revision: "b5968e9190ef611bbf34a7229255be88a0e937c1"
   container:
-    image: "vllm/vllm-openai:v0.20.0-ubuntu2404"
+    image: "vllm/vllm-openai:v0.25.1"
   frameworks:
-    dynamo: "1.2.0.dev20260426"
-    vllm: "0.20.0"
+    dynamo: "1.3.0.dev1"
+    vllm: "0.25.1"

benchmarks/multi_node/srt-slurm-recipes/vllm/deepseek-v4/8k1k/disagg-gb200-max-tpt-megamoe.yaml

@@ -8,24 +8,22 @@ name: "svf-vllm-disagg-gb200-max-tpt-megamoe"
 # 4096 with no CPU/NVMe offload.
 #
 # Local deltas vs upstream:
-#   * model.path alias renamed deepseekv4-fp4 -> deepseek-v4-pro to match
-#     SRT_SLURM_MODEL_PREFIX in runners/launch_gb200-nv.sh.
-#   * model.container set to vllm/vllm-openai:v0.20.0-ubuntu2404 to
+#   * model.path uses the deepseek-v4-pro-mxfp4 alias from
+#     runners/launch_gb200-nv.sh.
+#   * model.container set to vllm/vllm-openai:v0.25.1 to
 #     match nvidia-master.yaml image (which the launch script registers as
 #     the alias key in srtslurm.yaml). Upstream variants ship either the
 #     non-dynamo floating tag or a sha256 pin.
 #   * slurm.time_limit + health_check set to 8h / 1440 attempts to
 #     absorb cold-cache /mnt/numa1 model loads.
 model:
-  path: "deepseek-v4-pro"
-  container: "vllm/vllm-openai:v0.20.0-ubuntu2404"
+  path: "deepseek-v4-pro-mxfp4"
+  container: "vllm/vllm-openai:v0.25.1"
   precision: "fp4"
 
 dynamo:
   install: true
-  wheel: "1.2.0.dev20260426"
-
-setup_script: vllm-container-deps.sh
+  wheel: "1.3.0.dev1"
 
 slurm:
   time_limit: "8:00:00"
@@ -48,11 +46,14 @@ infra:
 
 frontend:
   type: dynamo
-  enable_multiple_frontends: false
+  enable_multiple_frontends: true
+  env:
+    DYN_REQUEST_PLANE_CODEC: "msgpack"
 backend:
   type: vllm
   connector: null
   prefill_environment:
+    DYN_REQUEST_PLANE_CODEC: "msgpack"
     TILELANG_CLEANUP_TEMP_FILES: "1"
     VLLM_USE_NCCL_SYMM_MEM: "1"
     TORCH_SYMMMEM: "NVSHMEM"
@@ -70,6 +71,7 @@ backend:
     UCX_CUDA_IPC_ENABLE_MNNVL: "y"
     NCCL_P2P_LEVEL: NVL
   decode_environment:
+    DYN_REQUEST_PLANE_CODEC: "msgpack"
     TILELANG_CLEANUP_TEMP_FILES: "1"
     VLLM_USE_NCCL_SYMM_MEM: "1"
     TORCH_SYMMMEM: "NVSHMEM"
@@ -147,9 +149,9 @@ benchmark:
 identity:
   model:
     repo: "deepseek-ai/DeepSeek-V4-Pro"
-    revision: "0366e4e064385807ea86b088a5c6c878ff23343b"
+    revision: "b5968e9190ef611bbf34a7229255be88a0e937c1"
   container:
-    image: "vllm/vllm-openai:v0.20.0-ubuntu2404"
+    image: "vllm/vllm-openai:v0.25.1"
   frameworks:
-    dynamo: "1.2.0.dev20260426"
-    vllm: "0.20.0"
+    dynamo: "1.3.0.dev1"
+    vllm: "0.25.1"

benchmarks/multi_node/srt-slurm-recipes/vllm/deepseek-v4/8k1k/disagg-gb200-mid-curve-megamoe.yaml

@@ -8,24 +8,22 @@ name: "svf-vllm-disagg-gb200-mid-curve-megamoe"
 # 256/512/1024 with no CPU/NVMe offload.
 #
 # Local deltas vs upstream:
-#   * model.path alias renamed deepseekv4-fp4 -> deepseek-v4-pro to match
-#     SRT_SLURM_MODEL_PREFIX in runners/launch_gb200-nv.sh.
-#   * model.container set to vllm/vllm-openai:v0.20.0-ubuntu2404 to
+#   * model.path uses the deepseek-v4-pro-mxfp4 alias from
+#     runners/launch_gb200-nv.sh.
+#   * model.container set to vllm/vllm-openai:v0.25.1 to
 #     match nvidia-master.yaml image (which the launch script registers as
 #     the alias key in srtslurm.yaml). Upstream variants ship either the
 #     non-dynamo floating tag or a sha256 pin.
 #   * slurm.time_limit + health_check set to 8h / 1440 attempts to
 #     absorb cold-cache /mnt/numa1 model loads.
 model:
-  path: "deepseek-v4-pro"
-  container: "vllm/vllm-openai:v0.20.0-ubuntu2404"
+  path: "deepseek-v4-pro-mxfp4"
+  container: "vllm/vllm-openai:v0.25.1"
   precision: "fp4"
 
 dynamo:
   install: true
-  wheel: "1.2.0.dev20260426"
-
-setup_script: vllm-container-deps.sh
+  wheel: "1.3.0.dev1"
 
 slurm:
   time_limit: "8:00:00"
@@ -48,11 +46,14 @@ infra:
 
 frontend:
   type: dynamo
-  enable_multiple_frontends: false
+  enable_multiple_frontends: true
+  env:
+    DYN_REQUEST_PLANE_CODEC: "msgpack"
 backend:
   type: vllm
   connector: null
   prefill_environment:
+    DYN_REQUEST_PLANE_CODEC: "msgpack"
     TILELANG_CLEANUP_TEMP_FILES: "1"
     VLLM_USE_NCCL_SYMM_MEM: "1"
     TORCH_SYMMMEM: "NVSHMEM"
@@ -70,6 +71,7 @@ backend:
     UCX_CUDA_IPC_ENABLE_MNNVL: "y"
     NCCL_P2P_LEVEL: NVL
   decode_environment:
+    DYN_REQUEST_PLANE_CODEC: "msgpack"
     TILELANG_CLEANUP_TEMP_FILES: "1"
     VLLM_USE_NCCL_SYMM_MEM: "1"
     TORCH_SYMMMEM: "NVSHMEM"
@@ -147,9 +149,9 @@ benchmark:
 identity:
   model:
     repo: "deepseek-ai/DeepSeek-V4-Pro"
-    revision: "0366e4e064385807ea86b088a5c6c878ff23343b"
+    revision: "b5968e9190ef611bbf34a7229255be88a0e937c1"
   container:
-    image: "vllm/vllm-openai:v0.20.0-ubuntu2404"
+    image: "vllm/vllm-openai:v0.25.1"
   frameworks:
-    dynamo: "1.2.0.dev20260426"
-    vllm: "0.20.0"
+    dynamo: "1.3.0.dev1"
+    vllm: "0.25.1"

configs/nvidia-master.yaml

@@ -4982,7 +4982,7 @@ dsv4-fp4-b200-dynamo-vllm:
           dp-attn: true
 
 dsv4-fp4-gb200-dynamo-vllm:
-  image: vllm/vllm-openai:v0.20.0-ubuntu2404
+  image: vllm/vllm-openai:v0.25.1
   model: deepseek-ai/DeepSeek-V4-Pro
   model-prefix: dsv4
   runner: gb200

perf-changelog.yaml

@@ -5273,3 +5273,11 @@
     - "Add 8 checked-in srt-slurm recipes with EAGLE speculative decoding and chat-template inputs"
     - "Image: lmsysorg/sglang:nightly-dev-cu13-20260721-8905cbd4"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2363
+
+- config-keys:
+    - dsv4-fp4-gb200-dynamo-vllm
+  description:
+    - "Update the GB200 DeepSeek-V4-Pro Dynamo-vLLM stack to Dynamo 1.3.0.dev1, vLLM 0.25.1, and the MXFP4 checkpoint"
+    - "Enable multiple frontends and MessagePack request-plane encoding, and reduce 1P4D decode GPU memory utilization to 0.85"
+    - "Drop the recipe setup_script entirely (removes the DeepGEMM NVLink barrier timeout patch); required container deps now ship in vLLM 0.25.1, on NVIDIA/srt-slurm v1.0.31"
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2139