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

@@ -1,4 +1,3 @@
-
 schema: 2
 
 base:
@@ -38,7 +37,6 @@ base:
         SGLANG_FLASHINFER_FP4_GEMM_BACKEND: cutlass
         SGLANG_MOE_NVFP4_DISPATCH: '1'
         SGLANG_CUTEDSL_MOE_NVFP4_DISPATCH: '1'
-        SGLANG_NVFP4_CKPT_FP8_NEXTN_MOE: '1'
         SGLANG_NCCL_ALL_GATHER_IN_OVERLAP_SCHEDULER_SYNC_BATCH: '1'
         FLASHINFER_DISABLE_VERSION_CHECK: '1'
         SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
@@ -48,18 +46,41 @@ base:
         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
-        enable-cache-report: true
         model-path: /model/
+        enable-cache-report: true
         trust-remote-code: true
         quantization: modelopt_fp4
         kv-cache-dtype: fp8_e4m3
-        tensor-parallel-size: 2
         pipeline-parallel-size: 1
         data-parallel-size: 1
         expert-parallel-size: 1
-        moe-dense-tp-size: 2
         enable-dp-attention: false
         enable-dp-lm-head: false
         moe-a2a-backend: none
@@ -79,10 +100,7 @@ base:
         speculative-num-steps: 6
         speculative-eagle-topk: 1
         speculative-num-draft-tokens: 7
-        speculative-moe-runner-backend: flashinfer_trtllm
         speculative-moe-a2a-backend: none
-        chunked-prefill-size: 8192
-        max-prefill-tokens: 8192
         mem-fraction-static: 0.85
         max-running-requests: 1
         pp-max-micro-batch-size: 1
@@ -103,6 +121,7 @@ base:
         cuda-graph-max-bs-decode: 1
         cuda-graph-bs-decode:
           - 1
+        flashinfer-allreduce-fusion-backend: cutedsl
   sbatch_directives:
     mem: '0'
     cpus-per-task: '144'
@@ -115,51 +134,68 @@ base:
       AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: '0'
       AIPERF_REQUIRED_SERVER_METRIC_PREFIX: 'sglang:'
 
-override_tp2_c44_b1:
-  name: agg-gb300-tp2-c44-b1-mtp-hicache-nightly-20260831
+override_tp2_c10:
+  name: agg-gb300-tp2-c10-b1-mtp-hicache-src
   roles:
     agg:
       env:
-        SGLANG_DG_CACHE_DIR: /tmp/agentx-pareto-v2/main09ec-pr36248-20260829-tp2-fi-trtllm-c44-mrr1/deepgemm-cache
-        FLASHINFER_WORKSPACE_BASE: /tmp/agentx-pareto-v2/main09ec-pr36248-20260829-tp2-fi-trtllm-c44-mrr1/flashinfer-cache
-        SGLANG_CACHE_DIR: /tmp/agentx-pareto-v2/main09ec-pr36248-20260829-tp2-fi-trtllm-c44-mrr1/sglang-cache
+        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_tp2_c44_b2:
-  name: agg-gb300-tp2-c44-b2-mtp-hicache-nightly-20260831
+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_DG_CACHE_DIR: /tmp/agentx-pareto-v2/main09ec-pr36248-20260829-tp2-fi-trtllm-hicache128-c44-mrr2/deepgemm-cache
-        FLASHINFER_WORKSPACE_BASE: /tmp/agentx-pareto-v2/main09ec-pr36248-20260829-tp2-fi-trtllm-hicache128-c44-mrr2/flashinfer-cache
-        SGLANG_CACHE_DIR: /tmp/agentx-pareto-v2/main09ec-pr36248-20260829-tp2-fi-trtllm-hicache128-c44-mrr2/sglang-cache
+        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:
-        prefill-decode-interval: 0
-        max-running-requests: 2
-        pp-max-micro-batch-size: 2
-        prefill-max-requests: 2
-        cuda-graph-max-bs-decode: 2
-        cuda-graph-bs-decode:
-          - 1
-          - 2
-        hicache-size: 128
+        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_tp8_c7_b1:
-  name: agg-gb300-tp8-c7-b1-mtp-hicache-nightly-20260831
+override_tp4_c7:
+  name: agg-gb300-tp4-c7-b1-mtp-hicache-src
   resources:
     gpus_per_node: 4
   roles:
     agg:
-      nodes: 2
-      gpus: 8
+      nodes: 1
+      gpus: 4
       env:
-        SGLANG_DG_CACHE_DIR: /tmp/agentx-pareto-v2/c7-mrr1/deepgemm-cache
-        FLASHINFER_WORKSPACE_BASE: /tmp/agentx-pareto-v2/c7-mrr1/flashinfer-cache
-        SGLANG_CACHE_DIR: /tmp/agentx-pareto-v2/c7-mrr1/sglang-cache
+        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: 8
-        moe-dense-tp-size: 8
-        moe-runner-backend: flashinfer_cutedsl
+        tensor-parallel-size: 4
+        moe-dense-tp-size: 4
         speculative-moe-runner-backend: flashinfer_cutedsl
-        disable-prefill-cuda-graph: true
+        chunked-prefill-size: 16384
+        max-prefill-tokens: 16384
+        cuda-graph-backend-prefill: breakable
+        cuda-graph-max-bs-prefill: 16384
+        enable-symm-mem: true