[AgentX] GB300 DSV4.1 Flash SGLang: 20261014 nightly, fastokens, DEP2 bounded replay and c160, TP4 c1 single admission / [AgentX] GB300 DSV4.1 Flash SGLang：升级 20261014 nightly、启用 fastokens、DEP2 开 bounded replay 并新增 c160、TP4 c1 单请求准入

inferencex-e2e/benchmarks/single_node/srt-slurm-recipes/dsv41flash/sglang/gb300-fp4-mtp/agentic.yaml

@@ -29,6 +29,7 @@ base:
         cuda-graph-max-bs-decode: 64
         reasoning-parser: auto
         tool-call-parser: auto
+        tokenizer-backend: fastokens
         # Draft passes under long-context load outlast the 1800 s default watchdog.
         watchdog-timeout: 3600
         enable-metrics: true
@@ -56,15 +57,6 @@ base:
       AIPERF_HTTP_TCP_USER_TIMEOUT: '900000'
       AIPERF_REQUIRED_SERVER_METRIC_PREFIX: 'sglang:'
 
-# One variant per point. Admission is 2x CONC, capped at the graph batch. SWA
-# prefix tails are 64x CONC up to 4096. TP2 keeps row-sharded Engram tables in
-# host DRAM and runs 16K chunks with decoder SWA bounded replay and a longer
-# prefill-decode interval. c64 and c128 run DEP2 (DP attention with MegaMoE)
-# behind the SGLang router with shortest-prefill-first scheduling. TP4 uses
-# static ragged verify and 16K chunks; c1 keeps Engram in HBM with no
-# prefill-decode interval. TP4/EP1 c4 and c16 keep Engram in host DRAM with 4096
-# SWA tails and a 128 graph batch; c16 adds a 192-step interval. Saturation
-# points get a longer warmup drain.
 override_tp2_c16:
   roles:
     agg:
@@ -137,6 +129,7 @@ override_dep2_c64:
           tokenizer-worker-num: 4
           chunked-prefill-size: 16384
           swa-prefix-tails: 4096
+          enable-decoder-swa-bounded-replay: true
         prefill-decode-interval: 24
         max-running-requests: 128
       env: &dep2_env
@@ -169,6 +162,24 @@ override_dep2_c128:
       <<: *dep2_client
       CONC: '128'
 
+override_dep2_c160:
+  frontend: *dep2_frontend
+  roles:
+    agg:
+      gpus: 2
+      args:
+        <<: *dep2_args
+        prefill-decode-interval: 8
+        mem-fraction-static: 0.85
+        max-running-requests: 256
+        cuda-graph-max-bs-decode: 128
+        tokenizer-backend: huggingface
+      env: *dep2_env
+  benchmark:
+    env:
+      <<: *dep2_client
+      CONC: '160'
+
 override_tp4_c1:
   roles:
     agg:
@@ -178,7 +189,9 @@ override_tp4_c1:
         expert-parallel-size: 1
         chunked-prefill-size: 16384
         prefill-decode-interval: 0
-        max-running-requests: 2
+        max-running-requests: 1
+        enable-decoder-swa-bounded-replay: true
+        cuda-graph-backend-prefill: disabled
       env:
         SGLANG_ENABLE_DSV41_ENGRAM_HOST_TABLE: '0'
         SGLANG_RAGGED_VERIFY_MODE: static

inferencex-e2e/configs/nvidia-master.yaml

@@ -8946,7 +8946,7 @@ dsv41flash-fp4-b300-sglang-agentic-dspark:
 
 # Official SGLang nightly; TP4 C1 uses GPU Engram, other points use host DRAM.
 dsv41flash-fp4-gb300-sglang-agentic-dspark:
-  image: lmsysorg/sglang:nightly-dev-cu13-20261012-e6072935@sha256:b5c488694d74c13587af6057b90220498c4869571c8e1e3c43dd6f43ae721306
+  image: lmsysorg/sglang:nightly-dev-cu13-20261014-684fea8a@sha256:ddae0e399d3d94f7827f16183301cf8dac4cf0eae0ef25590b7dac5881d713d8
   model: deepseek-ai/DeepSeek-V4.1-Flash
   model-prefix: dsv41flash
   runner: cluster:gb300-nv
@@ -8959,7 +8959,7 @@ dsv41flash-fp4-gb300-sglang-agentic-dspark:
     - dram-utilization: 0.80
       search-space:
       - { tp: 2, ep: 2, kv-offloading: none, spec-decoding: mtp, conc-list: [16, 32], srt-recipe: agentic.yaml }
-      - { tp: 2, ep: 2, dp-attn: true, kv-offloading: none, spec-decoding: mtp, conc-list: [64, 128], router: { name: sglang-router, version: "0.3.2" }, srt-recipe: agentic.yaml }
+      - { tp: 2, ep: 2, dp-attn: true, kv-offloading: none, spec-decoding: mtp, conc-list: [64, 128, 160], router: { name: sglang-router, version: "0.3.2" }, srt-recipe: agentic.yaml }
       - { tp: 4, ep: 1, kv-offloading: none, spec-decoding: mtp, conc-list: [1, 4, 16], srt-recipe: agentic.yaml }
 dsv41flash-fp4-gb300-dynamo-sglang-agentic-agg:
   image: lmsysorg/sglang:nightly-dev-20260928-81f27fb3@sha256:d9e4917808cfaa4b3be033a0c85a3a73d71eb93c17accbfa2ac3e96f551f3a33

inferencex-e2e/perf-changelog.yaml

@@ -10022,3 +10022,15 @@
     - "Append GB300 DeepSeek-V4.1-Flash Dynamo + vLLM AgentX c5120 (8P1D) and c6656 (10P1D) points using the agentx-harness 1.0.10 release already pinned on main."
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/4126
   append-only: true
+
+- config-keys:
+    - dsv41flash-fp4-gb300-sglang-agentic-dspark
+  scenario-type:
+    - agentic-coding
+  description:
+    - "Bump the image to lmsysorg/sglang:nightly-dev-cu13-20261014-684fea8a, which ships the fastokens tokenizer and decoder SWA bounded replay with DP attention."
+    - "Use tokenizer-backend fastokens on every point except DEP2 c160, which keeps the HuggingFace tokenizer."
+    - "Enable decoder SWA bounded replay on DEP2 c64 and c128."
+    - "TP4 c1 admits one request at a time, enables decoder SWA bounded replay and disables the prefill CUDA graph."
+    - "Add DEP2 c160 with the c128 settings, a prefill-decode interval of 8, mem-fraction-static 0.85, 256 max running requests and a 128 decode graph batch."
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/4033