perf(agentx): roll out deterministic warmup

Author: cquil11

perf-changelog.yaml

@@ -5230,3 +5230,10 @@
     - "Bump image to vllm/vllm-openai:nightly-4080263bb2c5d10deac17aaeb88e0823bc35bca9 and dynamo to 1.3.0.dev20260710; drop the minimax-m3-gb300-vllm-fixes.sh setup patch (shipped in the nightly image)"
     - "Collapse the 2xDEP8 decode into a single DEP16 worker (decode_workers 2->1, gpus_per_decode 8->16, data-parallel-size 8->16); recipe renamed 3p2d-dep2-dep8 -> 3p1d-dep2-dep16 (conc 512). The remaining decodes are 1xDEP8 or TEP and are unchanged."
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2340
+
+- config-keys:
+    - kimik3-fp4-b300-vllm-agentic
+  description:
+    - "Pin shared AIPerf to c57ad643 and replace duration-based warmup with 10 deterministic requests per lane."
+    - "Use one warmup request per lane with the 20-minute agentx-fast profile, and emit TTFT, ITL, and throughput metrics every 30 seconds."
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2415