diff --git a/.github/configs/nvidia-master.yaml b/.github/configs/nvidia-master.yaml
index c8fe1ea97..0a770efa4 100644
--- a/.github/configs/nvidia-master.yaml
+++ b/.github/configs/nvidia-master.yaml
@@ -10312,6 +10312,175 @@ dsv4-fp4-gb300-dynamo-sglang-mtp:
           ep: 8
           dp-attn: true
 
+dsv4-fp4-gb300-dynamo-sglang-mtp-1k1k:
+  image: lmsysorg/sglang:nightly-dev-cu13-20260615-c127ba64
+  model: deepseek-ai/DeepSeek-V4-Pro
+  model-prefix: dsv4
+  runner: gb300-nv
+  precision: fp4
+  framework: dynamo-sglang
+  multinode: true
+  disagg: true
+  scenarios:
+    fixed-seq-len:
+    - isl: 1024
+      osl: 1024
+      search-space:
+      - spec-decoding: "mtp"
+        conc-list: [8192]
+        prefill:
+          num-worker: 1
+          tp: 4
+          ep: 4
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/sglang/deepseek-v4/1k1k/disagg-1p1d-dep8-conc8192-mtp.yaml"
+        decode:
+          num-worker: 1
+          tp: 8
+          ep: 8
+          dp-attn: true
+      - spec-decoding: "mtp"
+        conc-list: [8192]
+        prefill:
+          num-worker: 1
+          tp: 4
+          ep: 4
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/sglang/deepseek-v4/1k1k/disagg-1p1d-dep16-conc8192-mtp.yaml"
+        decode:
+          num-worker: 1
+          tp: 16
+          ep: 16
+          dp-attn: true
+      - spec-decoding: "mtp"
+        conc-list: [8192]
+        prefill:
+          num-worker: 2
+          tp: 4
+          ep: 4
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/sglang/deepseek-v4/1k1k/disagg-2p1d-dep16-conc8192-mtp.yaml"
+        decode:
+          num-worker: 1
+          tp: 16
+          ep: 16
+          dp-attn: true
+      - spec-decoding: "mtp"
+        conc-list: [64]
+        prefill:
+          num-worker: 1
+          tp: 4
+          ep: 4
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p1d-dep4-tp4-mtp.yaml"
+        decode:
+          num-worker: 1
+          tp: 4
+          ep: 1
+          dp-attn: false
+      - spec-decoding: "mtp"
+        conc-list: [64]
+        prefill:
+          num-worker: 1
+          tp: 4
+          ep: 4
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p2d-dep4-tp4-mtp.yaml"
+        decode:
+          num-worker: 2
+          tp: 4
+          ep: 1
+          dp-attn: false
+      - spec-decoding: "mtp"
+        conc-list: [64]
+        prefill:
+          num-worker: 1
+          tp: 4
+          ep: 4
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p4d-dep4-tp4-mtp.yaml"
+        decode:
+          num-worker: 4
+          tp: 4
+          ep: 1
+          dp-attn: false
+      - spec-decoding: "mtp"
+        conc-list: [64]
+        prefill:
+          num-worker: 1
+          tp: 4
+          ep: 4
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p6d-dep4-tp4-mtp.yaml"
+        decode:
+          num-worker: 6
+          tp: 4
+          ep: 1
+          dp-attn: false
+      - spec-decoding: "mtp"
+        conc-list: [2]
+        prefill:
+          num-worker: 1
+          tp: 4
+          ep: 1
+          dp-attn: false
+          additional-settings:
+          - "CONFIG_FILE=recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p1d-tp4-tp4-mtp.yaml"
+        decode:
+          num-worker: 1
+          tp: 4
+          ep: 1
+          dp-attn: false
+      - spec-decoding: "mtp"
+        conc-list: [2]
+        prefill:
+          num-worker: 1
+          tp: 4
+          ep: 1
+          dp-attn: false
+          additional-settings:
+          - "CONFIG_FILE=recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p2d-tp4-tp4-mtp.yaml"
+        decode:
+          num-worker: 2
+          tp: 4
+          ep: 1
+          dp-attn: false
+      - spec-decoding: "mtp"
+        conc-list: [2]
+        prefill:
+          num-worker: 1
+          tp: 4
+          ep: 1
+          dp-attn: false
+          additional-settings:
+          - "CONFIG_FILE=recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p4d-tp4-tp4-mtp.yaml"
+        decode:
+          num-worker: 4
+          tp: 4
+          ep: 1
+          dp-attn: false
+      - spec-decoding: "mtp"
+        conc-list: [2]
+        prefill:
+          num-worker: 1
+          tp: 4
+          ep: 1
+          dp-attn: false
+          additional-settings:
+          - "CONFIG_FILE=recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p6d-tp4-tp4-mtp.yaml"
+        decode:
+          num-worker: 6
+          tp: 4
+          ep: 1
+          dp-attn: false
+
 
 kimik2.5-int4-h100-vllm:
   image: vllm/vllm-openai:v0.20.2
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-1p1d-dep16-conc8192-mtp.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-1p1d-dep16-conc8192-mtp.yaml
new file mode 100644
index 000000000..fed845263
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-1p1d-dep16-conc8192-mtp.yaml
@@ -0,0 +1,160 @@
+name: "dsv4-pro-gb300-disagg-1k1k-mtp-1p1d-dep16-conc8192"
+
+slurm:
+  time_limit: "03:00:00"
+
+sbatch_directives:
+  cpus-per-task: '144'
+  mem: '0'
+
+dynamo:
+  hash: "34d55a596fb8d3d44daefe425ec1e303131f4d2c"
+  install: true
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: true
+  num_additional_frontends: 8
+
+model:
+  path: dsv4-pro
+  container: "lmsysorg/sglang:nightly-dev-cu13-20260615-c127ba64"
+  precision: fp4
+
+resources:
+  gpu_type: gb300
+  gpus_per_node: 4
+  prefill_nodes: 1
+  prefill_workers: 1
+  gpus_per_prefill: 4
+  decode_nodes: 4
+  decode_workers: 1
+  gpus_per_decode: 16
+
+backend:
+  type: sglang
+
+  prefill_environment:
+    PYTHONUNBUFFERED: '1'
+    SGLANG_RADIX_DISABLE_REUSE: '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_JIT_NORM: '1'
+    SGLANG_OPT_USE_JIT_INDEXER_METADATA: '1'
+    SGLANG_OPT_USE_TOPK_V2: '1'
+
+    SGLANG_OPT_SWA_EVICT_DROP_PAGE_MARGIN: '1'
+    SGLANG_OPT_USE_CUSTOM_ALL_REDUCE_V2: '1'
+    SGLANG_OPT_USE_FAST_MASK_EP: '1'
+    SGLANG_OPT_USE_DEEPGEMM_MEGA_MOE: '1'
+    SGLANG_OPT_FIX_HASH_MEGA_MOE: '1'
+    SGLANG_OPT_DEEPGEMM_MEGA_MOE_NUM_MAX_TOKENS_PER_RANK: '9216'
+    SGLANG_OPT_FIX_MEGA_MOE_MEMORY: '1'
+    SGLANG_OPT_FIX_NEXTN_MEGA_MOE: '1'
+    SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK: '0'
+
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    MC_FORCE_MNNVL: '1'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW: '1'
+
+  decode_environment:
+    PYTHONUNBUFFERED: '1'
+    SGLANG_RADIX_DISABLE_REUSE: '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_JIT_NORM: '1'
+    SGLANG_OPT_USE_JIT_INDEXER_METADATA: '1'
+    SGLANG_OPT_USE_TOPK_V2: '1'
+
+    SGLANG_OPT_SWA_EVICT_DROP_PAGE_MARGIN: '1'
+    SGLANG_OPT_USE_FAST_MASK_EP: '1'
+    SGLANG_OPT_USE_DEEPGEMM_MEGA_MOE: '1'
+    SGLANG_OPT_FIX_HASH_MEGA_MOE: '1'
+    SGLANG_OPT_DEEPGEMM_MEGA_MOE_NUM_MAX_TOKENS_PER_RANK: '4096'
+    SGLANG_OPT_FIX_MEGA_MOE_MEMORY: '1'
+    SGLANG_OPT_FIX_NEXTN_MEGA_MOE: '1'
+    SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK: '0'
+
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    MC_FORCE_MNNVL: '1'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW: '1'
+    SGLANG_OPT_USE_CUSTOM_ALL_REDUCE_V2: "0"  # CAR_V2 is single-node only.
+
+  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
+      disaggregation-transfer-backend: mooncake
+
+      tensor-parallel-size: 4
+      data-parallel-size: 4
+      expert-parallel-size: 4
+
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+
+      moe-a2a-backend: deepep
+      deepep-config: '{"normal_dispatch":{"num_sms":96},"normal_combine":{"num_sms":96}}'
+
+      mem-fraction-static: 0.9
+      max-running-requests: 1024
+      cuda-graph-max-bs: 1024
+      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
+      disaggregation-transfer-backend: mooncake
+
+      tensor-parallel-size: 16
+      data-parallel-size: 16
+      expert-parallel-size: 16
+
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+
+      moe-a2a-backend: deepep
+      deepep-config: '{"normal_dispatch":{"num_sms":96},"normal_combine":{"num_sms":96}}'
+
+      speculative-algo: "EAGLE"
+      speculative-num-steps: 3
+      speculative-eagle-topk: 1
+      speculative-num-draft-tokens: 4
+
+      mem-fraction-static: 0.9
+      max-running-requests: 18432
+      cuda-graph-max-bs: 1024
+      swa-full-tokens-ratio: 0.15
+      context-length: 16384
+      stream-interval: 60
+
+benchmark:
+  type: "sa-bench"
+  isl: 1024
+  osl: 1024
+  random_range_ratio: 0.8
+  concurrencies: "8192"
+  req_rate: "inf"
+  use_chat_template: true
+  custom_tokenizer: "sa_bench_tokenizers.sglang_deepseek_v4.SGLangDeepseekV4Tokenizer"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-1p1d-dep8-conc8192-mtp.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-1p1d-dep8-conc8192-mtp.yaml
new file mode 100644
index 000000000..65f4f4d0a
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-1p1d-dep8-conc8192-mtp.yaml
@@ -0,0 +1,161 @@
+name: "dsv4-pro-gb300-disagg-1k1k-mtp-1p1d-dep8-conc8192"
+
+slurm:
+  time_limit: "03:00:00"
+
+sbatch_directives:
+  cpus-per-task: '144'
+  mem: '0'
+
+dynamo:
+  hash: "34d55a596fb8d3d44daefe425ec1e303131f4d2c"
+  install: true
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: true
+  num_additional_frontends: 8
+
+model:
+  path: dsv4-pro
+  container: "lmsysorg/sglang:nightly-dev-cu13-20260615-c127ba64"
+  precision: fp4
+
+resources:
+  gpu_type: gb300
+  gpus_per_node: 4
+  prefill_nodes: 1
+  prefill_workers: 1
+  gpus_per_prefill: 4
+  decode_nodes: 2
+  decode_workers: 1
+  gpus_per_decode: 8
+
+backend:
+  type: sglang
+
+  prefill_environment:
+    PYTHONUNBUFFERED: '1'
+    SGLANG_RADIX_DISABLE_REUSE: '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_JIT_NORM: '1'
+    SGLANG_OPT_USE_JIT_INDEXER_METADATA: '1'
+    SGLANG_OPT_USE_TOPK_V2: '1'
+
+    SGLANG_OPT_SWA_EVICT_DROP_PAGE_MARGIN: '1'
+    SGLANG_OPT_USE_CUSTOM_ALL_REDUCE_V2: '1'
+    SGLANG_OPT_USE_FAST_MASK_EP: '1'
+    SGLANG_OPT_USE_DEEPGEMM_MEGA_MOE: '1'
+    SGLANG_OPT_FIX_HASH_MEGA_MOE: '1'
+    SGLANG_OPT_DEEPGEMM_MEGA_MOE_NUM_MAX_TOKENS_PER_RANK: '9216'
+    SGLANG_OPT_FIX_MEGA_MOE_MEMORY: '1'
+    SGLANG_OPT_FIX_NEXTN_MEGA_MOE: '1'
+    SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK: '0'
+
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    MC_FORCE_MNNVL: '1'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW: '1'
+
+  decode_environment:
+    PYTHONUNBUFFERED: '1'
+    SGLANG_RADIX_DISABLE_REUSE: '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_JIT_NORM: '1'
+    SGLANG_OPT_USE_JIT_INDEXER_METADATA: '1'
+    SGLANG_OPT_USE_TOPK_V2: '1'
+
+    SGLANG_OPT_SWA_EVICT_DROP_PAGE_MARGIN: '1'
+    SGLANG_OPT_USE_FAST_MASK_EP: '1'
+    SGLANG_OPT_USE_DEEPGEMM_MEGA_MOE: '1'
+    SGLANG_OPT_FIX_HASH_MEGA_MOE: '1'
+    SGLANG_OPT_DEEPGEMM_MEGA_MOE_NUM_MAX_TOKENS_PER_RANK: '2048'
+    SGLANG_OPT_FIX_MEGA_MOE_MEMORY: '1'
+    SGLANG_OPT_FIX_NEXTN_MEGA_MOE: '1'
+    SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK: '0'
+
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    MC_FORCE_MNNVL: '1'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW: '1'
+    SGLANG_OPT_USE_CUSTOM_ALL_REDUCE_V2: "0"  # CAR_V2 is single-node only.
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: "1800"
+
+  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
+      disaggregation-transfer-backend: mooncake
+
+      tensor-parallel-size: 4
+      data-parallel-size: 4
+      expert-parallel-size: 4
+
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+
+      moe-a2a-backend: deepep
+      deepep-config: '{"normal_dispatch":{"num_sms":96},"normal_combine":{"num_sms":96}}'
+
+      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
+      disaggregation-transfer-backend: mooncake
+
+      tensor-parallel-size: 8
+      data-parallel-size: 8
+      expert-parallel-size: 8
+
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+
+      moe-a2a-backend: deepep
+      deepep-config: '{"normal_dispatch":{"num_sms":96},"normal_combine":{"num_sms":96}}'
+
+      speculative-algo: "EAGLE"
+      speculative-num-steps: 3
+      speculative-eagle-topk: 1
+      speculative-num-draft-tokens: 4
+
+      mem-fraction-static: 0.94
+      max-running-requests: 1536
+      cuda-graph-max-bs: 256
+      swa-full-tokens-ratio: 0.15
+      context-length: 16384
+      stream-interval: 60
+
+benchmark:
+  type: "sa-bench"
+  isl: 1024
+  osl: 1024
+  random_range_ratio: 0.8
+  concurrencies: "8192"
+  req_rate: "inf"
+  use_chat_template: true
+  custom_tokenizer: "sa_bench_tokenizers.sglang_deepseek_v4.SGLangDeepseekV4Tokenizer"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-2p1d-dep16-conc8192-mtp.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-2p1d-dep16-conc8192-mtp.yaml
new file mode 100644
index 000000000..bcafd24bd
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-2p1d-dep16-conc8192-mtp.yaml
@@ -0,0 +1,160 @@
+name: "dsv4-pro-gb300-disagg-1k1k-mtp-2p1d-dep16-conc8192"
+
+slurm:
+  time_limit: "03:00:00"
+
+sbatch_directives:
+  cpus-per-task: '144'
+  mem: '0'
+
+dynamo:
+  hash: "34d55a596fb8d3d44daefe425ec1e303131f4d2c"
+  install: true
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: true
+  num_additional_frontends: 8
+
+model:
+  path: dsv4-pro
+  container: "lmsysorg/sglang:nightly-dev-cu13-20260615-c127ba64"
+  precision: fp4
+
+resources:
+  gpu_type: gb300
+  gpus_per_node: 4
+  prefill_nodes: 2
+  prefill_workers: 2
+  gpus_per_prefill: 4
+  decode_nodes: 4
+  decode_workers: 1
+  gpus_per_decode: 16
+
+backend:
+  type: sglang
+
+  prefill_environment:
+    PYTHONUNBUFFERED: '1'
+    SGLANG_RADIX_DISABLE_REUSE: '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_JIT_NORM: '1'
+    SGLANG_OPT_USE_JIT_INDEXER_METADATA: '1'
+    SGLANG_OPT_USE_TOPK_V2: '1'
+
+    SGLANG_OPT_SWA_EVICT_DROP_PAGE_MARGIN: '1'
+    SGLANG_OPT_USE_CUSTOM_ALL_REDUCE_V2: '1'
+    SGLANG_OPT_USE_FAST_MASK_EP: '1'
+    SGLANG_OPT_USE_DEEPGEMM_MEGA_MOE: '1'
+    SGLANG_OPT_FIX_HASH_MEGA_MOE: '1'
+    SGLANG_OPT_DEEPGEMM_MEGA_MOE_NUM_MAX_TOKENS_PER_RANK: '9216'
+    SGLANG_OPT_FIX_MEGA_MOE_MEMORY: '1'
+    SGLANG_OPT_FIX_NEXTN_MEGA_MOE: '1'
+    SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK: '0'
+
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    MC_FORCE_MNNVL: '1'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW: '1'
+
+  decode_environment:
+    PYTHONUNBUFFERED: '1'
+    SGLANG_RADIX_DISABLE_REUSE: '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_JIT_NORM: '1'
+    SGLANG_OPT_USE_JIT_INDEXER_METADATA: '1'
+    SGLANG_OPT_USE_TOPK_V2: '1'
+
+    SGLANG_OPT_SWA_EVICT_DROP_PAGE_MARGIN: '1'
+    SGLANG_OPT_USE_FAST_MASK_EP: '1'
+    SGLANG_OPT_USE_DEEPGEMM_MEGA_MOE: '1'
+    SGLANG_OPT_FIX_HASH_MEGA_MOE: '1'
+    SGLANG_OPT_DEEPGEMM_MEGA_MOE_NUM_MAX_TOKENS_PER_RANK: '4096'
+    SGLANG_OPT_FIX_MEGA_MOE_MEMORY: '1'
+    SGLANG_OPT_FIX_NEXTN_MEGA_MOE: '1'
+    SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK: '0'
+
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    MC_FORCE_MNNVL: '1'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW: '1'
+    SGLANG_OPT_USE_CUSTOM_ALL_REDUCE_V2: "0"  # CAR_V2 is single-node only.
+
+  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
+      disaggregation-transfer-backend: mooncake
+
+      tensor-parallel-size: 4
+      data-parallel-size: 4
+      expert-parallel-size: 4
+
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+
+      moe-a2a-backend: deepep
+      deepep-config: '{"normal_dispatch":{"num_sms":96},"normal_combine":{"num_sms":96}}'
+
+      mem-fraction-static: 0.9
+      max-running-requests: 1024
+      cuda-graph-max-bs: 1024
+      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
+      disaggregation-transfer-backend: mooncake
+
+      tensor-parallel-size: 16
+      data-parallel-size: 16
+      expert-parallel-size: 16
+
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+
+      moe-a2a-backend: deepep
+      deepep-config: '{"normal_dispatch":{"num_sms":96},"normal_combine":{"num_sms":96}}'
+
+      speculative-algo: "EAGLE"
+      speculative-num-steps: 3
+      speculative-eagle-topk: 1
+      speculative-num-draft-tokens: 4
+
+      mem-fraction-static: 0.9
+      max-running-requests: 18432
+      cuda-graph-max-bs: 1024
+      swa-full-tokens-ratio: 0.15
+      context-length: 16384
+      stream-interval: 60
+
+benchmark:
+  type: "sa-bench"
+  isl: 1024
+  osl: 1024
+  random_range_ratio: 0.8
+  concurrencies: "8192"
+  req_rate: "inf"
+  use_chat_template: true
+  custom_tokenizer: "sa_bench_tokenizers.sglang_deepseek_v4.SGLangDeepseekV4Tokenizer"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p1d-dep4-tp4-mtp.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p1d-dep4-tp4-mtp.yaml
new file mode 100644
index 000000000..126207601
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p1d-dep4-tp4-mtp.yaml
@@ -0,0 +1,134 @@
+name: "dsv4-pro-gb300-disagg-1k1k-mtp-low-latency-1p1d-dep4-tp4"
+
+frontend:
+  type: sglang
+  enable_multiple_frontends: false
+  args:
+    policy: "cache_aware"
+
+model:
+  path: "dsv4-pro"
+  container: "lmsysorg/sglang:nightly-dev-cu13-20260615-c127ba64"
+  precision: "mxfp4"
+
+resources:
+  gpu_type: "gb300"
+  gpus_per_node: 4
+  prefill_nodes: 1
+  prefill_workers: 1
+  decode_nodes: 1
+  decode_workers: 1
+
+backend:
+  type: sglang
+
+  prefill_environment:
+    PYTHONUNBUFFERED: "1"
+    SGLANG_RADIX_DISABLE_REUSE: "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_JIT_NORM: "1"
+    SGLANG_OPT_USE_JIT_INDEXER_METADATA: "1"
+    SGLANG_OPT_USE_TOPK_V2: "1"
+
+    SGLANG_OPT_SWA_EVICT_DROP_PAGE_MARGIN: "1"
+    SGLANG_OPT_USE_CUSTOM_ALL_REDUCE_V2: "1"
+    SGLANG_OPT_USE_FAST_MASK_EP: "1"
+    SGLANG_OPT_USE_DEEPGEMM_MEGA_MOE: "1"
+    SGLANG_OPT_FIX_HASH_MEGA_MOE: "1"
+    SGLANG_OPT_DEEPGEMM_MEGA_MOE_NUM_MAX_TOKENS_PER_RANK: "9216"
+    SGLANG_OPT_FIX_MEGA_MOE_MEMORY: "1"
+    SGLANG_OPT_FIX_NEXTN_MEGA_MOE: "1"
+    SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK: "0"
+
+    NCCL_MNNVL_ENABLE: "1"
+    NCCL_CUMEM_ENABLE: "1"
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: "True"
+    MC_FORCE_MNNVL: "1"
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: "100000"
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: "100000"
+    SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW: "1"
+
+  decode_environment:
+    PYTHONUNBUFFERED: "1"
+    SGLANG_RADIX_DISABLE_REUSE: "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_JIT_NORM: "1"
+    SGLANG_OPT_USE_JIT_INDEXER_METADATA: "1"
+    SGLANG_OPT_USE_TOPK_V2: "1"
+    NCCL_MNNVL_ENABLE: "1"
+    NCCL_CUMEM_ENABLE: "1"
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: "True"
+    MC_FORCE_MNNVL: "1"
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: "100000"
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: "100000"
+    SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW: "1"
+    # SGLANG_OPT_USE_CUSTOM_ALL_REDUCE_V2 intentionally NOT set: CAR_V2
+    # is single-node only and corrupts results in 2-node decode setups.
+
+  sglang_config:
+    prefill:
+      served-model-name: "deepseek-ai/DeepSeek-V4-Pro"
+      model-path: "/model/"
+      trust-remote-code: true
+      disable-radix-cache: true
+
+      disaggregation-mode: "prefill"
+      disaggregation-transfer-backend: mooncake
+
+      tensor-parallel-size: 4
+      data-parallel-size: 4
+      expert-parallel-size: 4
+
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+
+      moe-a2a-backend: "deepep"
+      deepep-config: '{"normal_dispatch":{"num_sms":96},"normal_combine":{"num_sms":96}}'
+
+      mem-fraction-static: 0.90
+      max-running-requests: 512
+      cuda-graph-max-bs: 512
+      chunked-prefill-size: 4096
+
+    decode:
+      served-model-name: "deepseek-ai/DeepSeek-V4-Pro"
+      model-path: "/model/"
+      trust-remote-code: true
+      disable-radix-cache: true
+
+      disaggregation-mode: "decode"
+      disaggregation-transfer-backend: mooncake
+
+      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: 64
+      cuda-graph-max-bs:    64
+      swa-full-tokens-ratio: 0.1
+      context-length: 4096
+
+benchmark:
+  type: "sa-bench"
+  isl: 1024
+  osl: 1024
+  random_range_ratio: 0.8
+  concurrencies: "1x2x4x8x16x32x64"
+  req_rate: "inf"
+  use_chat_template: true
+  custom_tokenizer: "sa_bench_tokenizers.sglang_deepseek_v4.SGLangDeepseekV4Tokenizer"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p1d-tp4-tp4-mtp.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p1d-tp4-tp4-mtp.yaml
new file mode 100644
index 000000000..346933557
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p1d-tp4-tp4-mtp.yaml
@@ -0,0 +1,120 @@
+name: "dsv4-pro-gb300-disagg-1k1k-mtp-low-latency-1p1d-tp4-tp4"
+
+frontend:
+  type: sglang
+  enable_multiple_frontends: false
+  args:
+    policy: "cache_aware"
+
+model:
+  path: "dsv4-pro"
+  container: "lmsysorg/sglang:nightly-dev-cu13-20260615-c127ba64"
+  precision: "mxfp4"
+
+resources:
+  gpu_type: "gb300"
+  gpus_per_node: 4
+  prefill_nodes: 1
+  prefill_workers: 1
+  decode_nodes: 1
+  decode_workers: 1
+
+backend:
+  type: sglang
+
+  prefill_environment:
+    PYTHONUNBUFFERED: "1"
+    SGLANG_RADIX_DISABLE_REUSE: "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_JIT_NORM: "1"
+    SGLANG_OPT_USE_JIT_INDEXER_METADATA: "1"
+    SGLANG_OPT_USE_TOPK_V2: "1"
+    NCCL_MNNVL_ENABLE: "1"
+    NCCL_CUMEM_ENABLE: "1"
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: "True"
+    MC_FORCE_MNNVL: "1"
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: "100000"
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: "100000"
+    SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW: "1"
+
+  decode_environment:
+    PYTHONUNBUFFERED: "1"
+    SGLANG_RADIX_DISABLE_REUSE: "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_JIT_NORM: "1"
+    SGLANG_OPT_USE_JIT_INDEXER_METADATA: "1"
+    SGLANG_OPT_USE_TOPK_V2: "1"
+    NCCL_MNNVL_ENABLE: "1"
+    NCCL_CUMEM_ENABLE: "1"
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: "True"
+    MC_FORCE_MNNVL: "1"
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: "100000"
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: "100000"
+    SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW: "1"
+    # SGLANG_OPT_USE_CUSTOM_ALL_REDUCE_V2 intentionally NOT set: CAR_V2
+    # is single-node only and corrupts results in 2-node decode setups.
+
+  sglang_config:
+    prefill:
+      served-model-name: "deepseek-ai/DeepSeek-V4-Pro"
+      model-path: "/model/"
+      trust-remote-code: true
+      disable-radix-cache: true
+
+      disaggregation-mode: "prefill"
+      disaggregation-transfer-backend: mooncake
+
+      tensor-parallel-size: 4
+      data-parallel-size:   1
+      expert-parallel-size: 1
+
+      moe-runner-backend: "flashinfer_mxfp4"
+      disable-flashinfer-autotune: true
+
+      mem-fraction-static: 0.90
+      max-running-requests: 8
+      cuda-graph-max-bs: 8
+      chunked-prefill-size: 4096
+
+    decode:
+      served-model-name: "deepseek-ai/DeepSeek-V4-Pro"
+      model-path: "/model/"
+      trust-remote-code: true
+      disable-radix-cache: true
+
+      disaggregation-mode: "decode"
+      disaggregation-transfer-backend: mooncake
+
+      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: 4096
+
+benchmark:
+  type: "sa-bench"
+  isl: 1024
+  osl: 1024
+  random_range_ratio: 0.8
+  concurrencies: "1x2"
+  req_rate: "inf"
+  use_chat_template: true
+  custom_tokenizer: "sa_bench_tokenizers.sglang_deepseek_v4.SGLangDeepseekV4Tokenizer"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p2d-dep4-tp4-mtp.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p2d-dep4-tp4-mtp.yaml
new file mode 100644
index 000000000..e7c712e50
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p2d-dep4-tp4-mtp.yaml
@@ -0,0 +1,134 @@
+name: "dsv4-pro-gb300-disagg-1k1k-mtp-low-latency-1p2d-dep4-tp4"
+
+frontend:
+  type: sglang
+  enable_multiple_frontends: false
+  args:
+    policy: "cache_aware"
+
+model:
+  path: "dsv4-pro"
+  container: "lmsysorg/sglang:nightly-dev-cu13-20260615-c127ba64"
+  precision: "mxfp4"
+
+resources:
+  gpu_type: "gb300"
+  gpus_per_node: 4
+  prefill_nodes: 1
+  prefill_workers: 1
+  decode_nodes: 2
+  decode_workers: 2
+
+backend:
+  type: sglang
+
+  prefill_environment:
+    PYTHONUNBUFFERED: "1"
+    SGLANG_RADIX_DISABLE_REUSE: "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_JIT_NORM: "1"
+    SGLANG_OPT_USE_JIT_INDEXER_METADATA: "1"
+    SGLANG_OPT_USE_TOPK_V2: "1"
+
+    SGLANG_OPT_SWA_EVICT_DROP_PAGE_MARGIN: "1"
+    SGLANG_OPT_USE_CUSTOM_ALL_REDUCE_V2: "1"
+    SGLANG_OPT_USE_FAST_MASK_EP: "1"
+    SGLANG_OPT_USE_DEEPGEMM_MEGA_MOE: "1"
+    SGLANG_OPT_FIX_HASH_MEGA_MOE: "1"
+    SGLANG_OPT_DEEPGEMM_MEGA_MOE_NUM_MAX_TOKENS_PER_RANK: "9216"
+    SGLANG_OPT_FIX_MEGA_MOE_MEMORY: "1"
+    SGLANG_OPT_FIX_NEXTN_MEGA_MOE: "1"
+    SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK: "0"
+
+    NCCL_MNNVL_ENABLE: "1"
+    NCCL_CUMEM_ENABLE: "1"
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: "True"
+    MC_FORCE_MNNVL: "1"
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: "100000"
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: "100000"
+    SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW: "1"
+
+  decode_environment:
+    PYTHONUNBUFFERED: "1"
+    SGLANG_RADIX_DISABLE_REUSE: "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_JIT_NORM: "1"
+    SGLANG_OPT_USE_JIT_INDEXER_METADATA: "1"
+    SGLANG_OPT_USE_TOPK_V2: "1"
+    NCCL_MNNVL_ENABLE: "1"
+    NCCL_CUMEM_ENABLE: "1"
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: "True"
+    MC_FORCE_MNNVL: "1"
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: "100000"
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: "100000"
+    SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW: "1"
+    # SGLANG_OPT_USE_CUSTOM_ALL_REDUCE_V2 intentionally NOT set: CAR_V2
+    # is single-node only and corrupts results in 2-node decode setups.
+
+  sglang_config:
+    prefill:
+      served-model-name: "deepseek-ai/DeepSeek-V4-Pro"
+      model-path: "/model/"
+      trust-remote-code: true
+      disable-radix-cache: true
+
+      disaggregation-mode: "prefill"
+      disaggregation-transfer-backend: mooncake
+
+      tensor-parallel-size: 4
+      data-parallel-size: 4
+      expert-parallel-size: 4
+
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+
+      moe-a2a-backend: "deepep"
+      deepep-config: '{"normal_dispatch":{"num_sms":96},"normal_combine":{"num_sms":96}}'
+
+      mem-fraction-static: 0.90
+      max-running-requests: 512
+      cuda-graph-max-bs: 512
+      chunked-prefill-size: 4096
+
+    decode:
+      served-model-name: "deepseek-ai/DeepSeek-V4-Pro"
+      model-path: "/model/"
+      trust-remote-code: true
+      disable-radix-cache: true
+
+      disaggregation-mode: "decode"
+      disaggregation-transfer-backend: mooncake
+
+      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: 64
+      cuda-graph-max-bs:    64
+      swa-full-tokens-ratio: 0.1
+      context-length: 4096
+
+benchmark:
+  type: "sa-bench"
+  isl: 1024
+  osl: 1024
+  random_range_ratio: 0.8
+  concurrencies: "1x2x4x8x16x32x64"
+  req_rate: "inf"
+  use_chat_template: true
+  custom_tokenizer: "sa_bench_tokenizers.sglang_deepseek_v4.SGLangDeepseekV4Tokenizer"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p2d-tp4-tp4-mtp.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p2d-tp4-tp4-mtp.yaml
new file mode 100644
index 000000000..134e08b34
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p2d-tp4-tp4-mtp.yaml
@@ -0,0 +1,120 @@
+name: "dsv4-pro-gb300-disagg-1k1k-mtp-low-latency-1p2d-tp4-tp4"
+
+frontend:
+  type: sglang
+  enable_multiple_frontends: false
+  args:
+    policy: "cache_aware"
+
+model:
+  path: "dsv4-pro"
+  container: "lmsysorg/sglang:nightly-dev-cu13-20260615-c127ba64"
+  precision: "mxfp4"
+
+resources:
+  gpu_type: "gb300"
+  gpus_per_node: 4
+  prefill_nodes: 1
+  prefill_workers: 1
+  decode_nodes: 2
+  decode_workers: 2
+
+backend:
+  type: sglang
+
+  prefill_environment:
+    PYTHONUNBUFFERED: "1"
+    SGLANG_RADIX_DISABLE_REUSE: "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_JIT_NORM: "1"
+    SGLANG_OPT_USE_JIT_INDEXER_METADATA: "1"
+    SGLANG_OPT_USE_TOPK_V2: "1"
+    NCCL_MNNVL_ENABLE: "1"
+    NCCL_CUMEM_ENABLE: "1"
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: "True"
+    MC_FORCE_MNNVL: "1"
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: "100000"
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: "100000"
+    SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW: "1"
+
+  decode_environment:
+    PYTHONUNBUFFERED: "1"
+    SGLANG_RADIX_DISABLE_REUSE: "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_JIT_NORM: "1"
+    SGLANG_OPT_USE_JIT_INDEXER_METADATA: "1"
+    SGLANG_OPT_USE_TOPK_V2: "1"
+    NCCL_MNNVL_ENABLE: "1"
+    NCCL_CUMEM_ENABLE: "1"
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: "True"
+    MC_FORCE_MNNVL: "1"
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: "100000"
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: "100000"
+    SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW: "1"
+    # SGLANG_OPT_USE_CUSTOM_ALL_REDUCE_V2 intentionally NOT set: CAR_V2
+    # is single-node only and corrupts results in 2-node decode setups.
+
+  sglang_config:
+    prefill:
+      served-model-name: "deepseek-ai/DeepSeek-V4-Pro"
+      model-path: "/model/"
+      trust-remote-code: true
+      disable-radix-cache: true
+
+      disaggregation-mode: "prefill"
+      disaggregation-transfer-backend: mooncake
+
+      tensor-parallel-size: 4
+      data-parallel-size:   1
+      expert-parallel-size: 1
+
+      moe-runner-backend: "flashinfer_mxfp4"
+      disable-flashinfer-autotune: true
+
+      mem-fraction-static: 0.90
+      max-running-requests: 8
+      cuda-graph-max-bs: 8
+      chunked-prefill-size: 4096
+
+    decode:
+      served-model-name: "deepseek-ai/DeepSeek-V4-Pro"
+      model-path: "/model/"
+      trust-remote-code: true
+      disable-radix-cache: true
+
+      disaggregation-mode: "decode"
+      disaggregation-transfer-backend: mooncake
+
+      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: 4096
+
+benchmark:
+  type: "sa-bench"
+  isl: 1024
+  osl: 1024
+  random_range_ratio: 0.8
+  concurrencies: "1x2"
+  req_rate: "inf"
+  use_chat_template: true
+  custom_tokenizer: "sa_bench_tokenizers.sglang_deepseek_v4.SGLangDeepseekV4Tokenizer"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p4d-dep4-tp4-mtp.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p4d-dep4-tp4-mtp.yaml
new file mode 100644
index 000000000..97eff2d40
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p4d-dep4-tp4-mtp.yaml
@@ -0,0 +1,134 @@
+name: "dsv4-pro-gb300-disagg-1k1k-mtp-low-latency-1p4d-dep4-tp4"
+
+frontend:
+  type: sglang
+  enable_multiple_frontends: false
+  args:
+    policy: "cache_aware"
+
+model:
+  path: "dsv4-pro"
+  container: "lmsysorg/sglang:nightly-dev-cu13-20260615-c127ba64"
+  precision: "mxfp4"
+
+resources:
+  gpu_type: "gb300"
+  gpus_per_node: 4
+  prefill_nodes: 1
+  prefill_workers: 1
+  decode_nodes: 4
+  decode_workers: 4
+
+backend:
+  type: sglang
+
+  prefill_environment:
+    PYTHONUNBUFFERED: "1"
+    SGLANG_RADIX_DISABLE_REUSE: "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_JIT_NORM: "1"
+    SGLANG_OPT_USE_JIT_INDEXER_METADATA: "1"
+    SGLANG_OPT_USE_TOPK_V2: "1"
+
+    SGLANG_OPT_SWA_EVICT_DROP_PAGE_MARGIN: "1"
+    SGLANG_OPT_USE_CUSTOM_ALL_REDUCE_V2: "1"
+    SGLANG_OPT_USE_FAST_MASK_EP: "1"
+    SGLANG_OPT_USE_DEEPGEMM_MEGA_MOE: "1"
+    SGLANG_OPT_FIX_HASH_MEGA_MOE: "1"
+    SGLANG_OPT_DEEPGEMM_MEGA_MOE_NUM_MAX_TOKENS_PER_RANK: "9216"
+    SGLANG_OPT_FIX_MEGA_MOE_MEMORY: "1"
+    SGLANG_OPT_FIX_NEXTN_MEGA_MOE: "1"
+    SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK: "0"
+
+    NCCL_MNNVL_ENABLE: "1"
+    NCCL_CUMEM_ENABLE: "1"
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: "True"
+    MC_FORCE_MNNVL: "1"
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: "100000"
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: "100000"
+    SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW: "1"
+
+  decode_environment:
+    PYTHONUNBUFFERED: "1"
+    SGLANG_RADIX_DISABLE_REUSE: "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_JIT_NORM: "1"
+    SGLANG_OPT_USE_JIT_INDEXER_METADATA: "1"
+    SGLANG_OPT_USE_TOPK_V2: "1"
+    NCCL_MNNVL_ENABLE: "1"
+    NCCL_CUMEM_ENABLE: "1"
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: "True"
+    MC_FORCE_MNNVL: "1"
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: "100000"
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: "100000"
+    SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW: "1"
+    # SGLANG_OPT_USE_CUSTOM_ALL_REDUCE_V2 intentionally NOT set: CAR_V2
+    # is single-node only and corrupts results in 2-node decode setups.
+
+  sglang_config:
+    prefill:
+      served-model-name: "deepseek-ai/DeepSeek-V4-Pro"
+      model-path: "/model/"
+      trust-remote-code: true
+      disable-radix-cache: true
+
+      disaggregation-mode: "prefill"
+      disaggregation-transfer-backend: mooncake
+
+      tensor-parallel-size: 4
+      data-parallel-size: 4
+      expert-parallel-size: 4
+
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+
+      moe-a2a-backend: "deepep"
+      deepep-config: '{"normal_dispatch":{"num_sms":96},"normal_combine":{"num_sms":96}}'
+
+      mem-fraction-static: 0.90
+      max-running-requests: 512
+      cuda-graph-max-bs: 512
+      chunked-prefill-size: 4096
+
+    decode:
+      served-model-name: "deepseek-ai/DeepSeek-V4-Pro"
+      model-path: "/model/"
+      trust-remote-code: true
+      disable-radix-cache: true
+
+      disaggregation-mode: "decode"
+      disaggregation-transfer-backend: mooncake
+
+      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: 64
+      cuda-graph-max-bs:    64
+      swa-full-tokens-ratio: 0.1
+      context-length: 4096
+
+benchmark:
+  type: "sa-bench"
+  isl: 1024
+  osl: 1024
+  random_range_ratio: 0.8
+  concurrencies: "1x2x4x8x16x32x64"
+  req_rate: "inf"
+  use_chat_template: true
+  custom_tokenizer: "sa_bench_tokenizers.sglang_deepseek_v4.SGLangDeepseekV4Tokenizer"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p4d-tp4-tp4-mtp.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p4d-tp4-tp4-mtp.yaml
new file mode 100644
index 000000000..11e059294
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p4d-tp4-tp4-mtp.yaml
@@ -0,0 +1,120 @@
+name: "dsv4-pro-gb300-disagg-1k1k-mtp-low-latency-1p4d-tp4-tp4"
+
+frontend:
+  type: sglang
+  enable_multiple_frontends: false
+  args:
+    policy: "cache_aware"
+
+model:
+  path: "dsv4-pro"
+  container: "lmsysorg/sglang:nightly-dev-cu13-20260615-c127ba64"
+  precision: "mxfp4"
+
+resources:
+  gpu_type: "gb300"
+  gpus_per_node: 4
+  prefill_nodes: 1
+  prefill_workers: 1
+  decode_nodes: 4
+  decode_workers: 4
+
+backend:
+  type: sglang
+
+  prefill_environment:
+    PYTHONUNBUFFERED: "1"
+    SGLANG_RADIX_DISABLE_REUSE: "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_JIT_NORM: "1"
+    SGLANG_OPT_USE_JIT_INDEXER_METADATA: "1"
+    SGLANG_OPT_USE_TOPK_V2: "1"
+    NCCL_MNNVL_ENABLE: "1"
+    NCCL_CUMEM_ENABLE: "1"
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: "True"
+    MC_FORCE_MNNVL: "1"
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: "100000"
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: "100000"
+    SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW: "1"
+
+  decode_environment:
+    PYTHONUNBUFFERED: "1"
+    SGLANG_RADIX_DISABLE_REUSE: "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_JIT_NORM: "1"
+    SGLANG_OPT_USE_JIT_INDEXER_METADATA: "1"
+    SGLANG_OPT_USE_TOPK_V2: "1"
+    NCCL_MNNVL_ENABLE: "1"
+    NCCL_CUMEM_ENABLE: "1"
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: "True"
+    MC_FORCE_MNNVL: "1"
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: "100000"
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: "100000"
+    SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW: "1"
+    # SGLANG_OPT_USE_CUSTOM_ALL_REDUCE_V2 intentionally NOT set: CAR_V2
+    # is single-node only and corrupts results in 2-node decode setups.
+
+  sglang_config:
+    prefill:
+      served-model-name: "deepseek-ai/DeepSeek-V4-Pro"
+      model-path: "/model/"
+      trust-remote-code: true
+      disable-radix-cache: true
+
+      disaggregation-mode: "prefill"
+      disaggregation-transfer-backend: mooncake
+
+      tensor-parallel-size: 4
+      data-parallel-size:   1
+      expert-parallel-size: 1
+
+      moe-runner-backend: "flashinfer_mxfp4"
+      disable-flashinfer-autotune: true
+
+      mem-fraction-static: 0.90
+      max-running-requests: 8
+      cuda-graph-max-bs: 8
+      chunked-prefill-size: 4096
+
+    decode:
+      served-model-name: "deepseek-ai/DeepSeek-V4-Pro"
+      model-path: "/model/"
+      trust-remote-code: true
+      disable-radix-cache: true
+
+      disaggregation-mode: "decode"
+      disaggregation-transfer-backend: mooncake
+
+      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: 4096
+
+benchmark:
+  type: "sa-bench"
+  isl: 1024
+  osl: 1024
+  random_range_ratio: 0.8
+  concurrencies: "1x2"
+  req_rate: "inf"
+  use_chat_template: true
+  custom_tokenizer: "sa_bench_tokenizers.sglang_deepseek_v4.SGLangDeepseekV4Tokenizer"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p6d-dep4-tp4-mtp.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p6d-dep4-tp4-mtp.yaml
new file mode 100644
index 000000000..b974668d2
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p6d-dep4-tp4-mtp.yaml
@@ -0,0 +1,134 @@
+name: "dsv4-pro-gb300-disagg-1k1k-mtp-low-latency-1p6d-dep4-tp4"
+
+frontend:
+  type: sglang
+  enable_multiple_frontends: false
+  args:
+    policy: "cache_aware"
+
+model:
+  path: "dsv4-pro"
+  container: "lmsysorg/sglang:nightly-dev-cu13-20260615-c127ba64"
+  precision: "mxfp4"
+
+resources:
+  gpu_type: "gb300"
+  gpus_per_node: 4
+  prefill_nodes: 1
+  prefill_workers: 1
+  decode_nodes: 6
+  decode_workers: 6
+
+backend:
+  type: sglang
+
+  prefill_environment:
+    PYTHONUNBUFFERED: "1"
+    SGLANG_RADIX_DISABLE_REUSE: "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_JIT_NORM: "1"
+    SGLANG_OPT_USE_JIT_INDEXER_METADATA: "1"
+    SGLANG_OPT_USE_TOPK_V2: "1"
+
+    SGLANG_OPT_SWA_EVICT_DROP_PAGE_MARGIN: "1"
+    SGLANG_OPT_USE_CUSTOM_ALL_REDUCE_V2: "1"
+    SGLANG_OPT_USE_FAST_MASK_EP: "1"
+    SGLANG_OPT_USE_DEEPGEMM_MEGA_MOE: "1"
+    SGLANG_OPT_FIX_HASH_MEGA_MOE: "1"
+    SGLANG_OPT_DEEPGEMM_MEGA_MOE_NUM_MAX_TOKENS_PER_RANK: "9216"
+    SGLANG_OPT_FIX_MEGA_MOE_MEMORY: "1"
+    SGLANG_OPT_FIX_NEXTN_MEGA_MOE: "1"
+    SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK: "0"
+
+    NCCL_MNNVL_ENABLE: "1"
+    NCCL_CUMEM_ENABLE: "1"
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: "True"
+    MC_FORCE_MNNVL: "1"
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: "100000"
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: "100000"
+    SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW: "1"
+
+  decode_environment:
+    PYTHONUNBUFFERED: "1"
+    SGLANG_RADIX_DISABLE_REUSE: "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_JIT_NORM: "1"
+    SGLANG_OPT_USE_JIT_INDEXER_METADATA: "1"
+    SGLANG_OPT_USE_TOPK_V2: "1"
+    NCCL_MNNVL_ENABLE: "1"
+    NCCL_CUMEM_ENABLE: "1"
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: "True"
+    MC_FORCE_MNNVL: "1"
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: "100000"
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: "100000"
+    SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW: "1"
+    # SGLANG_OPT_USE_CUSTOM_ALL_REDUCE_V2 intentionally NOT set: CAR_V2
+    # is single-node only and corrupts results in 2-node decode setups.
+
+  sglang_config:
+    prefill:
+      served-model-name: "deepseek-ai/DeepSeek-V4-Pro"
+      model-path: "/model/"
+      trust-remote-code: true
+      disable-radix-cache: true
+
+      disaggregation-mode: "prefill"
+      disaggregation-transfer-backend: mooncake
+
+      tensor-parallel-size: 4
+      data-parallel-size: 4
+      expert-parallel-size: 4
+
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+
+      moe-a2a-backend: "deepep"
+      deepep-config: '{"normal_dispatch":{"num_sms":96},"normal_combine":{"num_sms":96}}'
+
+      mem-fraction-static: 0.90
+      max-running-requests: 512
+      cuda-graph-max-bs: 512
+      chunked-prefill-size: 4096
+
+    decode:
+      served-model-name: "deepseek-ai/DeepSeek-V4-Pro"
+      model-path: "/model/"
+      trust-remote-code: true
+      disable-radix-cache: true
+
+      disaggregation-mode: "decode"
+      disaggregation-transfer-backend: mooncake
+
+      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: 64
+      cuda-graph-max-bs:    64
+      swa-full-tokens-ratio: 0.1
+      context-length: 4096
+
+benchmark:
+  type: "sa-bench"
+  isl: 1024
+  osl: 1024
+  random_range_ratio: 0.8
+  concurrencies: "1x2x4x8x16x32x64"
+  req_rate: "inf"
+  use_chat_template: true
+  custom_tokenizer: "sa_bench_tokenizers.sglang_deepseek_v4.SGLangDeepseekV4Tokenizer"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p6d-tp4-tp4-mtp.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p6d-tp4-tp4-mtp.yaml
new file mode 100644
index 000000000..b059593e8
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/disagg-low-latency-1p6d-tp4-tp4-mtp.yaml
@@ -0,0 +1,120 @@
+name: "dsv4-pro-gb300-disagg-1k1k-mtp-low-latency-1p6d-tp4-tp4"
+
+frontend:
+  type: sglang
+  enable_multiple_frontends: false
+  args:
+    policy: "cache_aware"
+
+model:
+  path: "dsv4-pro"
+  container: "lmsysorg/sglang:nightly-dev-cu13-20260615-c127ba64"
+  precision: "mxfp4"
+
+resources:
+  gpu_type: "gb300"
+  gpus_per_node: 4
+  prefill_nodes: 1
+  prefill_workers: 1
+  decode_nodes: 6
+  decode_workers: 6
+
+backend:
+  type: sglang
+
+  prefill_environment:
+    PYTHONUNBUFFERED: "1"
+    SGLANG_RADIX_DISABLE_REUSE: "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_JIT_NORM: "1"
+    SGLANG_OPT_USE_JIT_INDEXER_METADATA: "1"
+    SGLANG_OPT_USE_TOPK_V2: "1"
+    NCCL_MNNVL_ENABLE: "1"
+    NCCL_CUMEM_ENABLE: "1"
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: "True"
+    MC_FORCE_MNNVL: "1"
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: "100000"
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: "100000"
+    SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW: "1"
+
+  decode_environment:
+    PYTHONUNBUFFERED: "1"
+    SGLANG_RADIX_DISABLE_REUSE: "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_JIT_NORM: "1"
+    SGLANG_OPT_USE_JIT_INDEXER_METADATA: "1"
+    SGLANG_OPT_USE_TOPK_V2: "1"
+    NCCL_MNNVL_ENABLE: "1"
+    NCCL_CUMEM_ENABLE: "1"
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: "True"
+    MC_FORCE_MNNVL: "1"
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: "100000"
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: "100000"
+    SGLANG_OPT_SWA_RELEASE_LEAF_LOCK_AFTER_WINDOW: "1"
+    # SGLANG_OPT_USE_CUSTOM_ALL_REDUCE_V2 intentionally NOT set: CAR_V2
+    # is single-node only and corrupts results in 2-node decode setups.
+
+  sglang_config:
+    prefill:
+      served-model-name: "deepseek-ai/DeepSeek-V4-Pro"
+      model-path: "/model/"
+      trust-remote-code: true
+      disable-radix-cache: true
+
+      disaggregation-mode: "prefill"
+      disaggregation-transfer-backend: mooncake
+
+      tensor-parallel-size: 4
+      data-parallel-size:   1
+      expert-parallel-size: 1
+
+      moe-runner-backend: "flashinfer_mxfp4"
+      disable-flashinfer-autotune: true
+
+      mem-fraction-static: 0.90
+      max-running-requests: 8
+      cuda-graph-max-bs: 8
+      chunked-prefill-size: 4096
+
+    decode:
+      served-model-name: "deepseek-ai/DeepSeek-V4-Pro"
+      model-path: "/model/"
+      trust-remote-code: true
+      disable-radix-cache: true
+
+      disaggregation-mode: "decode"
+      disaggregation-transfer-backend: mooncake
+
+      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: 4096
+
+benchmark:
+  type: "sa-bench"
+  isl: 1024
+  osl: 1024
+  random_range_ratio: 0.8
+  concurrencies: "1x2"
+  req_rate: "inf"
+  use_chat_template: true
+  custom_tokenizer: "sa_bench_tokenizers.sglang_deepseek_v4.SGLangDeepseekV4Tokenizer"
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 67a91d5d2..744e0f971 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -3887,3 +3887,13 @@
     - "Align MiniMax-M3 B300 vLLM fixed-sequence serving with MiniMax-M2.5 FP8 settings by setting VLLM_FLOAT32_MATMUL_PRECISION=high and restoring max cudagraph capture size 2048."
     - "Add TP4+EP4 coverage for MiniMax-M3 B300: DP-attention rows for 1k1k/8k1k and the missing non-DP-attention row for 8k1k."
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1781
+
+- config-keys:
+    - dsv4-fp4-gb300-dynamo-sglang-mtp-1k1k
+  description:
+    - "Add DeepSeek-V4-Pro FP4 GB300 disaggregated SGLang MTP 1k/1k coverage using lmsysorg/sglang:nightly-dev-cu13-20260615-c127ba64"
+    - "New top-level config (sibling of dsv4-fp4-gb300-dynamo-sglang-mtp which stays on the older 8k/1k image nightly-dev-20260527-14f81a67); separate entry is required because the launcher builds one squashfs per top-level image and the 1k/1k recipes pin a newer container string"
+    - "Wires 11 disagg recipes adapted from NVIDIA/srt-slurm PR #177 (recipes/dsv4-pro/sglang/gb300-fp4/1k1k/disagg/mtp/), staged locally under benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/1k1k/: 3 high-throughput conc=8192 recipes (1p1d-dep8, 1p1d-dep16, 2p1d-dep16), 4 flat low-latency dep4 recipes (1p1d/1p2d/1p4d/1p6d-dep4-tp4 at conc=64), and 4 flat low-latency tp4 recipes (1p1d/1p2d/1p4d/1p6d-tp4-tp4 at conc=2)"
+    - "Flat dep4 and tp4 splits replace the upstream disagg-low-latency-dep4-mtp.yaml and disagg-low-latency-tp4-mtp.yaml templates (each base + zip_override_1k1k_lowlat with decode_nodes [1,2,4,6]). The templated form silently broke under the gb300-cw launcher: srtctl fans the single recipe into 4 Slurm submissions per `srtctl apply`, but launch_gb300-cw.sh:282 captures the job IDs with `grep -oP '✅ Job \\K[0-9]+'` into a scalar JOB_ID, jamming 4 newline-separated IDs into one variable. Downstream uses (LOGS_DIR path interpolation at :293-294, squeue polls at :313-314, error message at :315) all corrupt, producing the observed `ERROR: Job 7079\\n7082 failed before creating log file` and orphaning the real Slurm jobs. Single-config files match the 8k1k convention (one recipe per decode-worker count) and force srtctl to emit exactly 1 job per launcher invocation"
+    - "Decode side runs EAGLE (num-steps=3, eagle-topk=1, num-draft-tokens=4) on all 11 recipes; all recipes pin lmsysorg/sglang:nightly-dev-cu13-20260615-c127ba64"
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1697
diff --git a/runners/launch_gb300-nv.sh b/runners/launch_gb300-nv.sh
index d21c91d10..aec68d7ee 100644
--- a/runners/launch_gb300-nv.sh
+++ b/runners/launch_gb300-nv.sh
@@ -143,6 +143,12 @@ elif [[ $FRAMEWORK == "dynamo-sglang" && $MODEL_PREFIX == "glm5" ]]; then
     git checkout sa-submission-q2-2026
     mkdir -p recipes/sglang/glm5
     cp -rT "$GITHUB_WORKSPACE/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5" recipes/sglang/glm5
+elif [[ $FRAMEWORK == "dynamo-sglang" && $MODEL_PREFIX == "dsv4" ]]; then
+    git clone https://github.com/NVIDIA/srt-slurm.git "$SRT_REPO_DIR"
+    cd "$SRT_REPO_DIR"
+    git checkout main
+    mkdir -p recipes/sglang/deepseek-v4
+    cp -rT "$GITHUB_WORKSPACE/benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4" recipes/sglang/deepseek-v4
 elif [[ $FRAMEWORK == "dynamo-vllm" && $MODEL_PREFIX == "minimaxm2.5" && $PRECISION == "fp8" ]]; then
     git clone https://github.com/NVIDIA/srt-slurm.git "$SRT_REPO_DIR"
     cd "$SRT_REPO_DIR"