feat(agentx): port the MI355X DSV4 ATOM disagg LMCache config to srt-slurm

inferencex-e2e/benchmarks/multi_node/srt-slurm-recipes/dsv4/atom/mi355x-fp4/agentx/disagg-lmcache-dspark.yaml

@@ -0,0 +1,254 @@
+# DeepSeek-V4-Pro-0813 AgentX on MI355X: 1P1D ATOM disaggregation over Mooncake
+# RDMA behind AToMesh, with DSpark (three draft tokens). Ported from the legacy
+# amd_utils path (#3158; ATOM recipes/DeepSeek-V4-Agentic-PD-Max.md). Three tiers:
+# TP8 at concurrency 1-16, DP attention at 64-128, and DP attention plus ATOM's
+# in-process LMCache CPU offload on prefill at 256. srtctl generates the Mooncake
+# P/D connector; the offload tier adds lmcache_offload next to it
+# (runners/srt-slurm/patches/507-lmcache-server-atom-sglang.patch).
+base:
+  schema: 2
+  name: mi355x-dsv4-pro-0813-atom-agentx-lmcache
+  model:
+    path: DeepSeek-V4-Pro-0813
+    container: rocm/atom-dev:nightly_202609221542
+    precision: fp4
+  slurm:
+    time_limit: "08:00:00"
+  resources:
+    gpu_type: mi355x
+    gpus_per_node: 8
+  frontend:
+    type: atomesh
+    enable_multiple_frontends: false
+    args:
+      log-level: info
+      disable-health-check: true
+      disable-circuit-breaker: true
+      prometheus-port: 29100
+  engine: atom
+  roles:
+    prefill:
+      nodes: 1
+      workers: 1
+      gpus: 8
+      env: &environment
+        PYTHONUNBUFFERED: "1"
+        PYTHONDONTWRITEBYTECODE: "1"
+        SAFETENSORS_FAST_GPU: "1"
+        VLLM_LOG_LEVEL: WARNING
+        ATOM_LOG_LEVEL: WARNING
+        AITER_LOG_LEVEL: WARNING
+        LOG_LEVEL: WARNING
+        LOGLEVEL: WARNING
+        ATOM_MOE_GU_ITLV: "1"
+        AITER_BF16_FP8_MOE_BOUND: "0"
+        # Rail-isolated RDMA: matched rails replace dest-device affinity.
+        MC_GID_INDEX: "1"
+        ATOM_MOONCAKE_MATCHED_RAILS: auto
+        NCCL_IB_DISABLE: "1"
+        ATOM_DISABLE_MMAP: "true"
+        ATOM_NUMA_BIND: "1"
+        ATOM_DP_MASTER_PORT: "29510"
+        ATOM_DP_BASE_PORT: "29610"
+        ATOM_PREFIX_CACHE_POLICY: lru
+        ATOM_PREFIX_CACHE_PROTECTED_RATIO: "0.5"
+      args: &server
+        served-model-name: deepseek-ai/DeepSeek-V4-Pro-0813
+        trust-remote-code: true
+        method: dspark
+        num-speculative-tokens: 3
+        kv_cache_dtype: fp8
+        index-cache-dtype: fp4
+        # ATOM forces block size 256 for V4.
+        block-size: 256
+        gpu-memory-utilization: 0.9
+        max-model-len: 1000000
+        max-num-batched-tokens: 16384
+        attn-prefill-chunk-size: 16384
+        state-checkpoint-interval-tokens: 8192
+        level: 3
+        cudagraph-mode: FULL
+        enable-prefix-caching: true
+    decode:
+      nodes: 1
+      workers: 1
+      gpus: 8
+      env:
+        <<: *environment
+      args:
+        <<: *server
+  sbatch_directives:
+    cpus-per-task: "128"
+    mem: "0"
+  srun_options:
+    mem: "0"
+    container-writable: ""
+    container-remap-root: ""
+  health_check:
+    max_attempts: 720
+    interval_seconds: 5
+  benchmark:
+    type: custom
+    command: bash /infmax-workspace/benchmarks/srt_agentic.sh
+    env:
+      MODEL: deepseek-ai/DeepSeek-V4-Pro-0813
+      # The MI355X launcher collects results from the job workspace.
+      RESULT_DIR: /infmax-workspace/LOGS/agentic
+      AGENTIC_OUTPUT_DIR: /infmax-workspace
+      AIPERF_REQUIRED_SERVER_METRIC_PREFIX: "atom:"
+      AIPERF_DATASET_MMAP_CACHE_DIR: /aiperf_mmap_cache
+      HF_HUB_CACHE: /hf_hub_cache/hub
+      TOKENIZERS_PARALLELISM: "false"
+      TRANSFORMERS_VERBOSITY: error
+
+# One variant per point; each point runs in its own allocation. Admission is
+# 2x CONC on both roles. Decode captures every batch up to min(64, 2x CONC) on
+# TP, or CONC/4 per rank under DP attention. DP attention adds TBO on prefill.
+
+override_tp8_c1:
+  frontend:
+    args:
+      prefill-policy: round_robin
+      decode-policy: round_robin
+      atom-pd-rank-mapping-policy: none
+  roles:
+    prefill:
+      args:
+        max-num-seqs: 2
+    decode:
+      args:
+        max-num-seqs: 2
+        cudagraph-capture-sizes: '[1,2]'
+  benchmark:
+    env:
+      CONC: '1'
+
+override_tp8_c16:
+  frontend:
+    args:
+      prefill-policy: round_robin
+      decode-policy: round_robin
+      atom-pd-rank-mapping-policy: none
+  roles:
+    prefill:
+      args:
+        max-num-seqs: 32
+    decode:
+      args:
+        max-num-seqs: 32
+        cudagraph-capture-sizes: '[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32]'
+  benchmark:
+    env:
+      CONC: '16'
+
+override_dpa8_c64:
+  frontend:
+    args:
+      dp-aware: true
+      prefill-policy: cache_aware
+      decode-policy: cache_aware
+      cache-threshold: 0.8
+      balance-abs-threshold: 20
+      balance-rel-threshold: 2.0
+      eviction-interval: 300
+      atom-pd-rank-mapping-policy: none
+  roles:
+    prefill:
+      env:
+        ATOM_ENABLE_PREFILL_DELAYER: '0'
+        GPU_MAX_HW_QUEUES: '5'
+      args:
+        max-num-seqs: 128
+        enable-dp-attention: true
+        enable-tbo: true
+    decode:
+      env:
+        ATOM_ENABLE_PREFILL_DELAYER: '0'
+      args:
+        max-num-seqs: 128
+        cudagraph-capture-sizes: '[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16]'
+        enable-dp-attention: true
+  benchmark:
+    env:
+      CONC: '64'
+
+override_dpa8_c128:
+  frontend:
+    args:
+      dp-aware: true
+      prefill-policy: cache_aware
+      decode-policy: cache_aware
+      cache-threshold: 0.8
+      balance-abs-threshold: 20
+      balance-rel-threshold: 2.0
+      eviction-interval: 300
+      atom-pd-rank-mapping-policy: none
+  roles:
+    prefill:
+      env:
+        ATOM_ENABLE_PREFILL_DELAYER: '0'
+        GPU_MAX_HW_QUEUES: '5'
+      args:
+        max-num-seqs: 256
+        enable-dp-attention: true
+        enable-tbo: true
+    decode:
+      env:
+        ATOM_ENABLE_PREFILL_DELAYER: '0'
+      args:
+        max-num-seqs: 256
+        cudagraph-capture-sizes: '[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32]'
+        enable-dp-attention: true
+  benchmark:
+    env:
+      CONC: '128'
+
+override_dpa8_c256_lmcache:
+  frontend:
+    args:
+      dp-aware: true
+      prefill-policy: dp_sticky
+      decode-policy: dp_sticky
+      atom-pd-rank-mapping-policy: none
+  roles:
+    prefill:
+      env:
+        ATOM_ENABLE_PREFILL_DELAYER: '0'
+        GPU_MAX_HW_QUEUES: '5'
+        PYTHONHASHSEED: '0'
+        OFFLOAD_COPY_WORKERS: '1'
+        OFFLOAD_MIN_LOAD_TOKENS: '8192'
+        OFFLOAD_SLOT_STAGING_SLOTS: '4'
+      args:
+        max-num-seqs: 512
+        enable-dp-attention: true
+        enable-tbo: true
+        extra-kv-connectors:
+          - kv_connector: lmcache_offload
+            kv_role: offload
+            offload_layout: hybrid
+            max_pending_saves: 8
+            slot_sidecar_staging_slots: 4
+            lmcache.local_cpu: true
+            # total-cpu-dram-gb 1499 (dram-utilization 0.5) over the eight prefill ranks.
+            lmcache.max_local_cpu_size: 187
+            lmcache.local_disk: null
+            lmcache.max_local_disk_size: 0
+            lmcache.remote_url: null
+            lmcache.chunk_size: 256
+            lmcache.cache_policy: LRU
+            lmcache.lookup_server_worker_ids: []
+            lmcache.store_location: LocalCPUBackend
+            lmcache.retrieve_locations: [LocalCPUBackend]
+    decode:
+      env:
+        ATOM_ENABLE_PREFILL_DELAYER: '0'
+      args:
+        max-num-seqs: 512
+        cudagraph-capture-sizes: '[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64]'
+        enable-dp-attention: true
+  benchmark:
+    env:
+      CONC: '256'
+      # dp_sticky keys sessions on the correlation id.
+      AIPERF_HTTP_X_SESSION_ID_FROM_CORRELATION_ID: '1'

inferencex-e2e/configs/amd-master.yaml

@@ -1215,52 +1215,74 @@ dsv4-fp4-mi355x-atom-disagg-agentic-lmcache-dspark:
   disagg: true
   scenarios:
     agentic-coding:
-    # MTP tiers with GPU-resident KV, unchanged from run 35643358891 apart from
-    # the image bump. Concurrency 1/16 run plain TP8; 64/128 add DP attention
-    # (c128 measured in run 35810485934, gsm8k eval in run 35851118405).
+    # srt-slurm recipe benchmarks/multi_node/srt-slurm-recipes/dsv4/atom/mi355x-fp4/agentx/disagg-lmcache-dspark.yaml
+    # (ATOM recipes/DeepSeek-V4-Agentic-PD-Max.md). Concurrency 1/16 run plain
+    # TP8; 64/128 add DP attention with prefill TBO and the cache-aware router.
     - dram-utilization: 0.80
       search-space:
       - spec-decoding: "draft_model"
-        conc-list: [ 1, 16 ]
+        conc-list: [ 1 ]
         kv-offloading: none
         prefill:
           num-worker: 1
           tp: 8
           ep: 1
           dp-attn: false
           additional-settings:
-          - "PREFILL_NODES=1"
+          - "CONFIG_FILE=recipes/dsv4/atom/mi355x-fp4/agentx/disagg-lmcache-dspark.yaml:override_tp8_c1"
         decode:
           num-worker: 1
           tp: 8
           ep: 1
           dp-attn: false
+      - spec-decoding: "draft_model"
+        conc-list: [ 16 ]
+        kv-offloading: none
+        prefill:
+          num-worker: 1
+          tp: 8
+          ep: 1
+          dp-attn: false
           additional-settings:
-          - "DECODE_NODES=1"
-          - "DECODE_MTP_SIZE=3"
-      # DP-attention tier (recipe: concurrency 64-128). TP8 + DP attention,
-      # prefill TBO only, cache-aware router, no CPU offload.
+          - "CONFIG_FILE=recipes/dsv4/atom/mi355x-fp4/agentx/disagg-lmcache-dspark.yaml:override_tp8_c16"
+        decode:
+          num-worker: 1
+          tp: 8
+          ep: 1
+          dp-attn: false
       - spec-decoding: "draft_model"
-        conc-list: [ 64, 128 ]
+        conc-list: [ 64 ]
         kv-offloading: none
         prefill:
           num-worker: 1
           tp: 8
           ep: 1
           dp-attn: true
           additional-settings:
-          - "PREFILL_NODES=1"
+          - "CONFIG_FILE=recipes/dsv4/atom/mi355x-fp4/agentx/disagg-lmcache-dspark.yaml:override_dpa8_c64"
         decode:
           num-worker: 1
           tp: 8
           ep: 1
           dp-attn: true
+      - spec-decoding: "draft_model"
+        conc-list: [ 128 ]
+        kv-offloading: none
+        prefill:
+          num-worker: 1
+          tp: 8
+          ep: 1
+          dp-attn: true
           additional-settings:
-          - "DECODE_NODES=1"
-          - "DECODE_MTP_SIZE=3"
-    # DP-attention + CPU offload tier (recipe: concurrency 256). Prefill adds
-    # the lmcache host KV offload tier; decode is plain Mooncake. DSpark draft
-    # model and dram-utilization 0.50 as measured in run 35825955863.
+          - "CONFIG_FILE=recipes/dsv4/atom/mi355x-fp4/agentx/disagg-lmcache-dspark.yaml:override_dpa8_c128"
+        decode:
+          num-worker: 1
+          tp: 8
+          ep: 1
+          dp-attn: true
+    # DP attention plus ATOM's in-process LMCache CPU offload on prefill
+    # (lmcache_offload); decode is plain Mooncake. dram-utilization 0.50 as
+    # measured in run 35825955863.
     - dram-utilization: 0.50
       search-space:
       - spec-decoding: "draft_model"
@@ -1273,15 +1295,12 @@ dsv4-fp4-mi355x-atom-disagg-agentic-lmcache-dspark:
           ep: 1
           dp-attn: true
           additional-settings:
-          - "PREFILL_NODES=1"
+          - "CONFIG_FILE=recipes/dsv4/atom/mi355x-fp4/agentx/disagg-lmcache-dspark.yaml:override_dpa8_c256_lmcache"
         decode:
           num-worker: 1
           tp: 8
           ep: 1
           dp-attn: true
-          additional-settings:
-          - "DECODE_NODES=1"
-          - "DECODE_MTP_SIZE=3"
 
 # GLM-5.2 FP4 agentic-coding benchmark on MI355X via SGLang with MTP speculative
 # decoding. Two arms:

inferencex-e2e/infx/srt_slurm/synthetic_acceptance.py

@@ -28,6 +28,7 @@
     "trt": "trtllm",
     "dynamo-trt": "trtllm",
     "atom": "atom",
+    "atom-disagg": "atom",
 }
 SGLANG_VARIABLES = (
     "SGLANG_SIMULATE_ACC_LEN",

inferencex-e2e/perf-changelog.yaml

@@ -8995,3 +8995,12 @@
     - "Validate MiniMax-M3 AgentX on H200 with vLLM FP8 after moving the end-to-end project into inferencex-e2e/. Preserve the existing image, recipe, and benchmark settings."
     - "将端到端项目迁入 inferencex-e2e/ 后，验证 H200 上的 vLLM FP8 MiniMax-M3 AgentX 运行路径，沿用现有镜像、配方和基准测试设置。"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/3525
+
+- config-keys:
+    - dsv4-fp4-mi355x-atom-disagg-agentic-lmcache-dspark
+  scenario-type:
+    - agentic-coding
+  description:
+    - "Port the DeepSeek-V4-Pro-0813 MI355X ATOM 1P1D AgentX config from the legacy amd_utils path to a native srt-slurm recipe (benchmarks/multi_node/srt-slurm-recipes/dsv4/atom/mi355x-fp4/agentx/disagg-lmcache-dspark.yaml), one override variant per point. Server flags, env, AToMesh routing policies and decode CUDA-graph capture sizes match the legacy server_atom.sh for every tier: TP8 at concurrency 1 and 16, DP attention with prefill TBO at 64 and 128, and DP attention with ATOM's in-process LMCache CPU offload (lmcache_offload, 187 GB per prefill rank) at 256. Golden acceptance 3.01 for DSpark with three draft tokens is now injected by the srt-slurm path, the same value the legacy models_atom.yaml hardcoded."
+    - "srt-slurm had no way to add LMCache next to its generated Mooncake connector, so runners/srt-slurm/patches/507-lmcache-server-atom-sglang.patch applies SemiAnalysisAI/srt-slurm#32 (which includes NVIDIA/srt-slurm#507) to the pinned submodule until it lands upstream."
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/XXXX