feat: add DSV4 B300 Dynamo-SGLang MTP configuration / 新增 DSV4 B300 Dynamo-SGLang MTP 配置 (#2363)

* feat: add DSV4 FP4 B300 Dynamo-SGLang MTP config

中文：新增 DSV4 FP4 B300 Dynamo-SGLang MTP 配置，并接入启用聊天模板输入和 EAGLE 投机解码的共享 srt-slurm 配方矩阵。

* chore: link perf changelog to #2363

中文：将性能变更日志链接更新为 #2363。

* fix: use B300 DSV4 NVFP4 model path

中文：将 B300 DSV4 Dynamo-SGLang 配置指向服务器上预置的 DeepSeek-V4-Pro-NVFP4 模型路径。

* fix: pin srt-slurm recipe revision

固定 srt-slurm 配方版本，确保 DSV4 B300 Dynamo-SGLang 运行使用可复现的配置。

* fix(dsv4): pin UCX-enabled srt-slurm revision

修复（dsv4）：固定包含 UCX CUDA 传输配置的 srt-slurm 版本。

* fix: use checked-in DSV4 MTP recipes

Pin NVIDIA/srt-slurm main, overlay the checked-in recipes, and enable CUDA-aware UCX transport for every MTP topology.

修复：固定 NVIDIA/srt-slurm main 提交，覆盖仓库内置配方，并为所有 MTP 拓扑启用支持 CUDA 的 UCX 传输。

---------

Co-authored-by: Ankur-singh <ankusingh@nvidia.com>

Author: RohitNagraj

benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/8k1k/disagg-b300-1p1d-dep4-dep8-mtp.yaml

@@ -0,0 +1,134 @@
+name: "disagg-b300-1p1d-dep4-dep8-mtp"
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: true
+  num_additional_frontends: 8
+
+dynamo:
+  hash: "41882ae9b07232eed4850fb1daf8c958abb2556a"
+  install: true
+
+model:
+  path: "deepseek-v4-pro"
+  container: "lmsysorg/sglang:nightly-dev-cu13-20260721-8905cbd4"
+  precision: "fp4"
+
+sbatch_directives:
+  cpus-per-task: "144"
+  mem: "0"
+
+resources:
+  gpu_type: "b300"
+  gpus_per_node: 8
+  prefill_nodes: 1
+  prefill_workers: 1
+  gpus_per_prefill: 4
+  decode_nodes: 1
+  decode_workers: 1
+  gpus_per_decode: 8
+
+backend:
+  type: sglang
+
+  prefill_environment:
+    PYTHONUNBUFFERED: "1"
+    SGLANG_RADIX_FORCE_MISS: "1"
+    SGLANG_JIT_DEEPGEMM_FAST_WARMUP: "1"
+    SGLANG_DEFAULT_THINKING: "1"
+    SGLANG_DSV4_REASONING_EFFORT: "max"
+    SGLANG_OPT_SWA_SPLIT_LEAF_ON_INSERT: "1"
+    SGLANG_OPT_USE_CUSTOM_ALL_REDUCE_V2: "1"
+    SGLANG_OPT_DEEPGEMM_MEGA_MOE_USE_FP4_ACTS: "1"
+    SGLANG_OPT_DEEPGEMM_MEGA_MOE_USE_MXF4_KIND: "1"
+    SGLANG_OPT_DEEPGEMM_MEGA_MOE_NUM_MAX_TOKENS_PER_RANK: "9216"
+    NCCL_MNNVL_ENABLE: "1"
+    NCCL_CUMEM_ENABLE: "1"
+    MC_FORCE_MNNVL: "1"
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: "100000"
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: "100000"
+    SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW: "1"
+    UCX_TLS: "cuda_copy,rc"
+
+  decode_environment:
+    PYTHONUNBUFFERED: "1"
+    SGLANG_RADIX_FORCE_MISS: "1"
+    SGLANG_JIT_DEEPGEMM_FAST_WARMUP: "1"
+    SGLANG_DEFAULT_THINKING: "1"
+    SGLANG_DSV4_REASONING_EFFORT: "max"
+    SGLANG_OPT_SWA_SPLIT_LEAF_ON_INSERT: "1"
+    SGLANG_OPT_DEEPGEMM_MEGA_MOE_USE_FP4_ACTS: "1"
+    SGLANG_OPT_DEEPGEMM_MEGA_MOE_USE_MXF4_KIND: "1"
+    SGLANG_OPT_DEEPGEMM_MEGA_MOE_NUM_MAX_TOKENS_PER_RANK: "2048"
+    NCCL_MNNVL_ENABLE: "1"
+    NCCL_CUMEM_ENABLE: "1"
+    MC_FORCE_MNNVL: "1"
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: "100000"
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: "100000"
+    SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW: "1"
+    UCX_TLS: "cuda_copy,rc"
+    SGLANG_OPT_USE_CUSTOM_ALL_REDUCE_V2: "0"
+
+  sglang_config:
+    prefill:
+      served-model-name: "deepseek-ai/DeepSeek-V4-Pro"
+      model-path: "/model/"
+      trust-remote-code: true
+      tool-call-parser: deepseekv4
+
+      disaggregation-mode: "prefill"
+      watchdog-timeout: 86400
+      disaggregation-transfer-backend: nixl
+
+      tensor-parallel-size: 4
+      data-parallel-size: 4
+      expert-parallel-size: 4
+
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+      moe-a2a-backend: "megamoe"
+      mem-fraction-static: 0.9
+      max-running-requests: 256
+      cuda-graph-max-bs: 256
+      chunked-prefill-size: 32768
+      stream-interval: 60
+
+    decode:
+      served-model-name: "deepseek-ai/DeepSeek-V4-Pro"
+      model-path: "/model/"
+      trust-remote-code: true
+      tool-call-parser: deepseekv4
+
+      disaggregation-mode: "decode"
+      watchdog-timeout: 86400
+      disaggregation-transfer-backend: nixl
+
+      tensor-parallel-size: 8
+      data-parallel-size: 8
+      expert-parallel-size: 8
+
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+      moe-a2a-backend: "megamoe"
+
+      speculative-algo: "EAGLE"
+      speculative-num-steps: 3
+      speculative-eagle-topk: 1
+      speculative-num-draft-tokens: 4
+
+      mem-fraction-static: 0.94
+      max-running-requests: 3072
+      cuda-graph-max-bs: 256
+      swa-full-tokens-ratio: 0.15
+      context-length: 16384
+      stream-interval: 60
+
+benchmark:
+  type: "sa-bench"
+  isl: 8192
+  osl: 1024
+  random_range_ratio: 0.8
+  concurrencies: "256"
+  req_rate: "inf"
+  use_chat_template: true
+  custom_tokenizer: "sa_bench_tokenizers.sglang_deepseek_v4.SGLangDeepseekV4Tokenizer"

benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/8k1k/disagg-b300-1p1d-tp4-tp4-mtp.yaml

@@ -0,0 +1,125 @@
+name: "disagg-b300-1p1d-tp4-tp4-mtp"
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: true
+  num_additional_frontends: 8
+
+dynamo:
+  hash: "41882ae9b07232eed4850fb1daf8c958abb2556a"
+  install: true
+
+model:
+  path: "deepseek-v4-pro"
+  container: "lmsysorg/sglang:nightly-dev-cu13-20260721-8905cbd4"
+  precision: "fp4"
+
+sbatch_directives:
+  cpus-per-task: "144"
+  mem: "0"
+
+resources:
+  gpu_type: "b300"
+  gpus_per_node: 8
+  prefill_nodes: 1
+  prefill_workers: 1
+  gpus_per_prefill: 4
+  decode_nodes: 1
+  decode_workers: 1
+  gpus_per_decode: 4
+
+backend:
+  type: sglang
+
+  prefill_environment:
+    PYTHONUNBUFFERED: "1"
+    SGLANG_RADIX_FORCE_MISS: "1"
+    SGLANG_JIT_DEEPGEMM_FAST_WARMUP: "1"
+    SGLANG_DEFAULT_THINKING: "1"
+    SGLANG_DSV4_REASONING_EFFORT: "max"
+    SGLANG_OPT_SWA_SPLIT_LEAF_ON_INSERT: "1"
+    NCCL_MNNVL_ENABLE: "1"
+    NCCL_CUMEM_ENABLE: "1"
+    MC_FORCE_MNNVL: "1"
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: "100000"
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: "100000"
+    SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW: "1"
+    UCX_TLS: "cuda_copy,rc"
+
+  decode_environment:
+    PYTHONUNBUFFERED: "1"
+    SGLANG_RADIX_FORCE_MISS: "1"
+    SGLANG_JIT_DEEPGEMM_FAST_WARMUP: "1"
+    SGLANG_DEFAULT_THINKING: "1"
+    SGLANG_DSV4_REASONING_EFFORT: "max"
+    SGLANG_OPT_SWA_SPLIT_LEAF_ON_INSERT: "1"
+    NCCL_MNNVL_ENABLE: "1"
+    NCCL_CUMEM_ENABLE: "1"
+    MC_FORCE_MNNVL: "1"
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: "100000"
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: "100000"
+    SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW: "1"
+    UCX_TLS: "cuda_copy,rc"
+
+  sglang_config:
+    prefill:
+      served-model-name: "deepseek-ai/DeepSeek-V4-Pro"
+      model-path: "/model/"
+      trust-remote-code: true
+      tool-call-parser: deepseekv4
+
+      disaggregation-mode: "prefill"
+      watchdog-timeout: 86400
+      disaggregation-transfer-backend: nixl
+
+      tensor-parallel-size: 4
+      data-parallel-size: 1
+      expert-parallel-size: 1
+
+      moe-runner-backend: "flashinfer_mxfp4"
+      disable-flashinfer-autotune: true
+
+      mem-fraction-static: 0.9
+      max-running-requests: 8
+      cuda-graph-max-bs: 8
+      chunked-prefill-size: 32768
+
+
+    decode:
+      served-model-name: "deepseek-ai/DeepSeek-V4-Pro"
+      model-path: "/model/"
+      trust-remote-code: true
+      tool-call-parser: deepseekv4
+
+      disaggregation-mode: "decode"
+      watchdog-timeout: 86400
+      disaggregation-transfer-backend: nixl
+
+      tensor-parallel-size: 4
+      data-parallel-size: 1
+      expert-parallel-size: 1
+
+      moe-runner-backend: "flashinfer_mxfp4"
+      disable-flashinfer-autotune: true
+
+      speculative-algo: "EAGLE"
+      speculative-num-steps: 3
+      speculative-eagle-topk: 1
+      speculative-num-draft-tokens: 4
+
+      mem-fraction-static: 0.9
+      max-running-requests: 8
+      cuda-graph-max-bs: 8
+      swa-full-tokens-ratio: 0.1
+      context-length: 16384
+
+
+benchmark:
+  type: "sa-bench"
+  isl: 8192
+  osl: 1024
+  random_range_ratio: 0.8
+  concurrencies: "1"
+  req_rate: "inf"
+  use_chat_template: true
+  custom_tokenizer: "sa_bench_tokenizers.sglang_deepseek_v4.SGLangDeepseekV4Tokenizer"

benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/8k1k/disagg-b300-1p2d-dep4-dep8-mtp.yaml

@@ -0,0 +1,135 @@
+name: "disagg-b300-1p2d-dep4-dep8-mtp"
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: true
+  num_additional_frontends: 8
+
+dynamo:
+  hash: "41882ae9b07232eed4850fb1daf8c958abb2556a"
+  install: true
+
+model:
+  path: "deepseek-v4-pro"
+  container: "lmsysorg/sglang:nightly-dev-cu13-20260721-8905cbd4"
+  precision: "fp4"
+
+sbatch_directives:
+  cpus-per-task: "144"
+  mem: "0"
+
+resources:
+  gpu_type: "b300"
+  gpus_per_node: 8
+  prefill_nodes: 1
+  prefill_workers: 1
+  gpus_per_prefill: 4
+  decode_nodes: 2
+  decode_workers: 2
+  gpus_per_decode: 8
+
+backend:
+  type: sglang
+
+  prefill_environment:
+    PYTHONUNBUFFERED: "1"
+    SGLANG_RADIX_FORCE_MISS: "1"
+    SGLANG_JIT_DEEPGEMM_FAST_WARMUP: "1"
+    SGLANG_DEFAULT_THINKING: "1"
+    SGLANG_DSV4_REASONING_EFFORT: "max"
+    SGLANG_OPT_SWA_SPLIT_LEAF_ON_INSERT: "1"
+    SGLANG_OPT_USE_CUSTOM_ALL_REDUCE_V2: "1"
+    SGLANG_OPT_DEEPGEMM_MEGA_MOE_USE_FP4_ACTS: "1"
+    SGLANG_OPT_DEEPGEMM_MEGA_MOE_USE_MXF4_KIND: "1"
+    SGLANG_OPT_DEEPGEMM_MEGA_MOE_NUM_MAX_TOKENS_PER_RANK: "9216"
+    NCCL_MNNVL_ENABLE: "1"
+    NCCL_CUMEM_ENABLE: "1"
+    MC_FORCE_MNNVL: "1"
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: "100000"
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: "100000"
+    SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW: "1"
+    UCX_TLS: "cuda_copy,rc"
+
+  decode_environment:
+    PYTHONUNBUFFERED: "1"
+    SGLANG_RADIX_FORCE_MISS: "1"
+    SGLANG_JIT_DEEPGEMM_FAST_WARMUP: "1"
+    SGLANG_DEFAULT_THINKING: "1"
+    SGLANG_DSV4_REASONING_EFFORT: "max"
+    SGLANG_OPT_SWA_SPLIT_LEAF_ON_INSERT: "1"
+    SGLANG_OPT_DEEPGEMM_MEGA_MOE_USE_FP4_ACTS: "1"
+    SGLANG_OPT_DEEPGEMM_MEGA_MOE_USE_MXF4_KIND: "1"
+    SGLANG_OPT_DEEPGEMM_MEGA_MOE_NUM_MAX_TOKENS_PER_RANK: "2048"
+    NCCL_MNNVL_ENABLE: "1"
+    NCCL_CUMEM_ENABLE: "1"
+    MC_FORCE_MNNVL: "1"
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: "100000"
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: "100000"
+    SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW: "1"
+    UCX_TLS: "cuda_copy,rc"
+    SGLANG_OPT_USE_CUSTOM_ALL_REDUCE_V2: "0"
+
+  sglang_config:
+    prefill:
+      served-model-name: "deepseek-ai/DeepSeek-V4-Pro"
+      model-path: "/model/"
+      trust-remote-code: true
+      tool-call-parser: deepseekv4
+
+      disaggregation-mode: "prefill"
+      watchdog-timeout: 86400
+      disaggregation-transfer-backend: nixl
+
+      tensor-parallel-size: 4
+      data-parallel-size: 4
+      expert-parallel-size: 4
+
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+      moe-a2a-backend: "megamoe"
+
+      mem-fraction-static: 0.9
+      max-running-requests: 256
+      cuda-graph-max-bs: 256
+      chunked-prefill-size: 32768
+      stream-interval: 60
+
+    decode:
+      served-model-name: "deepseek-ai/DeepSeek-V4-Pro"
+      model-path: "/model/"
+      trust-remote-code: true
+      tool-call-parser: deepseekv4
+
+      disaggregation-mode: "decode"
+      watchdog-timeout: 86400
+      disaggregation-transfer-backend: nixl
+
+      tensor-parallel-size: 8
+      data-parallel-size: 8
+      expert-parallel-size: 8
+
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+      moe-a2a-backend: "megamoe"
+
+      speculative-algo: "EAGLE"
+      speculative-num-steps: 3
+      speculative-eagle-topk: 1
+      speculative-num-draft-tokens: 4
+
+      mem-fraction-static: 0.94
+      max-running-requests: 3072
+      cuda-graph-max-bs: 256
+      swa-full-tokens-ratio: 0.15
+      context-length: 16384
+      stream-interval: 60
+
+benchmark:
+  type: "sa-bench"
+  isl: 8192
+  osl: 1024
+  random_range_ratio: 0.8
+  concurrencies: "256"
+  req_rate: "inf"
+  use_chat_template: true
+  custom_tokenizer: "sa_bench_tokenizers.sglang_deepseek_v4.SGLangDeepseekV4Tokenizer"