Add nine Qwen3.5 NVFP4 GB300 Dynamo+SGLang AgentX points / 新增九个 Qwen3.5 NVFP4 GB300 Dynamo+SGLang AgentX 测试点

inferencex-e2e/benchmarks/multi_node/srt-slurm-recipes/qwen3.5/sglang/gb300-fp4/agentx/agg-src-variants.yaml

@@ -0,0 +1,201 @@
+schema: 2
+
+base:
+  slurm:
+    time_limit: '8:00:00'
+  health_check:
+    max_attempts: 1440
+    interval_seconds: 10
+  resources:
+    gpu_type: gb300
+    gpus_per_node: 2
+  frontend:
+    type: sglang-router
+    enable_multiple_frontends: false
+  engine: sglang
+  roles:
+    agg:
+      nodes: 1
+      workers: 1
+      gpus: 2
+      env:
+        SGLANG_TRTLLM_MHA_DECODE_SEQ_LEN_SPLITS: '1'
+        PYTORCH_CUDA_ALLOC_CONF: expandable_segments:True
+        NO_COLOR: '1'
+        PYTHONUNBUFFERED: '1'
+        PIP_BREAK_SYSTEM_PACKAGES: '1'
+        TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+        NCCL_MNNVL_ENABLE: '1'
+        NCCL_CUMEM_ENABLE: '1'
+        NCCL_NVLS_ENABLE: '0'
+        MC_FORCE_MNNVL: '1'
+        NVSHMEM_REMOTE_TRANSPORT: none
+        MC_TE_METRIC: 'true'
+        SGLANG_ENABLE_SPEC_V2: '1'
+        SGLANG_ENABLE_FLASHINFER_GEMM: 'true'
+        SGLANG_ENABLE_JIT_DEEPGEMM: 'true'
+        SGLANG_FLASHINFER_FP4_GEMM_BACKEND: cutlass
+        SGLANG_MOE_NVFP4_DISPATCH: '1'
+        SGLANG_CUTEDSL_MOE_NVFP4_DISPATCH: '1'
+        SGLANG_NCCL_ALL_GATHER_IN_OVERLAP_SCHEDULER_SYNC_BATCH: '1'
+        FLASHINFER_DISABLE_VERSION_CHECK: '1'
+        SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+        SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+        SGLANG_HEALTH_CHECK_TIMEOUT: '1800'
+        SGLANG_ENABLE_HEALTH_ENDPOINT_GENERATION: '0'
+        SGLANG_FLASHINFER_AUTOTUNE_CACHE: '1'
+        SGLANG_OPT_MAMBA_SKIP_DECODE_LOCK: '1'
+        SGLANG_SCHEDULER_SKIP_ALL_GATHER: '1'
+        SGLANG_Q35MF_GDN_VIEWS: '1'
+        SGLANG_Q35MF_GDN_LAZY_VALID: '1'
+        SGLANG_Q35MF_GDN_CONV_TILE: '1'
+        SGLANG_Q35MF_GDN_TILED_VERIFY: '1'
+        SGLANG_Q35MF_GDN_NORM_FP8: '1'
+        SGLANG_Q35MF_ATTN_FRONT: '1'
+        SGLANG_Q35MF_ATTN_GATE_QUANT: '1'
+        SGLANG_Q35MF_EDGE_SHM_SYNC: '1'
+        SGLANG_ENABLE_METADATA_GLUE_GRAPH: '1'
+        SGLANG_Q35MF_EDGE_POST_VERIFY_GRAPH: '1'
+        SGLANG_Q35MF_EDGE_CHAIN_TREE: '1'
+        SGLANG_Q35MF_DRAFT_ARGMAX_EXCHANGE: '1'
+        SGLANG_Q35MF_DRAFT_EXTEND_EXCHANGE: '1'
+        SGLANG_Q35MF_DRAFT_FC_SPLIT: '1'
+        SGLANG_Q35MF_EDGE_SPARSE_VERIFY_SAMPLING: '1'
+        SGLANG_Q35MF_EDGE_SIDE_STREAM_COMMIT: '1'
+        SGLANG_Q35MF_EDGE_PACKED_VERIFY_BCAST: '1'
+        SGLANG_Q35MF_ARRIVE_DRAFT_COPY: '1'
+        SGLANG_Q35MF_ARRIVE_D2H_RING: '1'
+        SGLANG_Q35MF_ARRIVE_SKIP_REGATHER: '1'
+        SGLANG_Q35MF_ARRIVE_DECODE_PREP: '1'
+        SGLANG_Q35MF_ARRIVE_DRAFT_PREP: '1'
+        SGLANG_Q35MF_ARRIVE_SEQ_LENS_SELECT: '1'
+        SGLANG_Q35MF_ARRIVE_COPY_PLAN: '1'
+        SGLANG_Q35MF_ROUTE_FUSED: '1'
+      args:
+        served-model-name: nvidia/Qwen3.5-397B-A17B-NVFP4
+        model-path: /model/
+        enable-cache-report: true
+        trust-remote-code: true
+        quantization: modelopt_fp4
+        kv-cache-dtype: fp8_e4m3
+        pipeline-parallel-size: 1
+        data-parallel-size: 1
+        expert-parallel-size: 1
+        enable-dp-attention: false
+        enable-dp-lm-head: false
+        moe-a2a-backend: none
+        load-balance-method: round_robin
+        mamba-radix-cache-strategy: extra_buffer
+        mamba-track-interval: 8192
+        mamba-max-states-per-path: 3
+        mamba-ssm-dtype: bfloat16
+        max-mamba-cache-size: 1536
+        context-length: 262144
+        page-size: 64
+        attention-backend: trtllm_mha
+        moe-runner-backend: flashinfer_trtllm
+        linear-attn-decode-backend: triton
+        disable-shared-experts-fusion: true
+        speculative-algorithm: NEXTN
+        speculative-num-steps: 6
+        speculative-eagle-topk: 1
+        speculative-num-draft-tokens: 7
+        speculative-moe-a2a-backend: none
+        mem-fraction-static: 0.85
+        max-running-requests: 1
+        pp-max-micro-batch-size: 1
+        prefill-max-requests: 1
+        stream-interval: 20
+        decode-log-interval: 10
+        watchdog-timeout: 1000000
+        weight-loader-prefetch-checkpoints: true
+        weight-loader-prefetch-num-threads: 4
+        enable-hierarchical-cache: true
+        hicache-write-policy: write_back
+        hicache-io-backend: kernel
+        hicache-mem-layout: page_first_direct
+        hicache-size: 32
+        enable-metrics: true
+        enable-linear-replayssm-spec: true
+        disable-attn-tp-gather: true
+        cuda-graph-max-bs-decode: 1
+        cuda-graph-bs-decode:
+          - 1
+        flashinfer-allreduce-fusion-backend: cutedsl
+  sbatch_directives:
+    mem: '0'
+    cpus-per-task: '144'
+  srun_options:
+    mem: '0'
+    container-remap-root: ''
+  benchmark:
+    env:
+      AIPERF_HTTP_X_DYNAMO_SESSION_ID_FROM_CORRELATION_ID: 'true'
+      AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: '0'
+      AIPERF_REQUIRED_SERVER_METRIC_PREFIX: 'sglang:'
+
+override_tp2_c10:
+  name: agg-gb300-tp2-c10-b1-mtp-hicache-src
+  roles:
+    agg:
+      env:
+        SGLANG_Q35MF_GDN_DIRECT_SOLVE: '0'
+        SGLANG_Q35MF_GDN_EMPTY_HISTORY: '0'
+      args:
+        tensor-parallel-size: 2
+        moe-dense-tp-size: 2
+        speculative-moe-runner-backend: flashinfer_trtllm
+        chunked-prefill-size: 8192
+        max-prefill-tokens: 8192
+        prefill-decode-interval: 0
+
+override_tp8_c7:
+  name: agg-gb300-tp8-c7-b1-mtp-hicache-src
+  resources:
+    gpus_per_node: 4
+  roles:
+    agg:
+      nodes: 2
+      gpus: 8
+      env:
+        SGLANG_FLASHINFER_MOE_FUSED_FINALIZE: '1'
+        SGLANG_Q35MF_EDGE_SHM_SYNC_MULTIHOST: '1'
+        SGLANG_Q35MF_EDGE_MULTIHOST_TCP: '1'
+        SGLANG_Q35MF_DRAFT_MOE_GEMV: '1'
+        SGLANG_Q35MF_DRAFT_MOE_ROUTER: '1'
+        SGLANG_Q35MF_DRAFT_FC_GEMV: '1'
+        SGLANG_Q35MF_ATTN_SPLIT_KV: '1'
+        SGLANG_Q35MF_ROUTE_SHARED: '1'
+      args:
+        tensor-parallel-size: 8
+        moe-dense-tp-size: 8
+        speculative-moe-runner-backend: flashinfer_cutedsl
+        chunked-prefill-size: 16384
+        max-prefill-tokens: 16384
+        cuda-graph-backend-prefill: breakable
+        cuda-graph-max-bs-prefill: 16384
+        enable-symm-mem: true
+
+override_tp4_c7:
+  name: agg-gb300-tp4-c7-b1-mtp-hicache-src
+  resources:
+    gpus_per_node: 4
+  roles:
+    agg:
+      nodes: 1
+      gpus: 4
+      env:
+        SGLANG_FLASHINFER_MOE_FUSED_FINALIZE: '1'
+        SGLANG_Q35MF_GDN_PDL: '1'
+        SGLANG_Q35MF_DRAFT_MOE_GEMV: '1'
+        SGLANG_Q35MF_DRAFT_MOE_ROUTER: '1'
+        SGLANG_Q35MF_DRAFT_FC_GEMV: '1'
+      args:
+        tensor-parallel-size: 4
+        moe-dense-tp-size: 4
+        speculative-moe-runner-backend: flashinfer_cutedsl
+        chunked-prefill-size: 16384
+        max-prefill-tokens: 16384
+        cuda-graph-backend-prefill: breakable
+        cuda-graph-max-bs-prefill: 16384
+        enable-symm-mem: true