feat(agentx): add 5-minute warmup and 20-minute profiling preset / feat(agentx)：新增 5 分钟预热与 20 分钟性能采集预设 (#2388)

* feat(agentx): add fast feedback timing preset

Add an e2e AgentX fast-mode switch that uses a five-minute cache warmup and twenty-minute profile without changing AIPerf. Preserve the canonical one-hour profile and existing warmup override outside fast mode.

中文：新增 AgentX 端到端快速模式开关，采用 5 分钟缓存预热和 20 分钟性能采集，无需修改 AIPerf。快速模式之外继续保留标准的 1 小时性能采集以及现有预热时长覆盖能力。

* docs(agentx): document fast preflight dispatch

Document the agentx-fast e2e input, timing preset, duration precedence, and the requirement to return to canonical timing for official sweeps.

中文：记录 agentx-fast 端到端输入、时长预设、duration 优先级，以及正式扫描必须恢复标准时长的要求。

Author: cquil11

.github/workflows/benchmark-multinode-tmpl.yml

@@ -154,6 +154,11 @@ on:
         type: string
         required: false
         default: "3600"
+      agentx-fast:
+        description: "Use the AgentX 5-minute warmup and 20-minute profiling preset"
+        type: boolean
+        required: false
+        default: false
       kv-offloading:
         description: "KV offload mode for agentic scenarios (none/dram)"
         required: false
@@ -214,6 +219,7 @@ env:
   IS_AGENTIC: ${{ inputs.scenario-type == 'agentic-coding' && '1' || '0' }}
   CONC: ${{ inputs.conc }}
   DURATION: ${{ inputs.duration }}
+  AIPERF_EXPERIMENTAL_FAST: ${{ inputs.agentx-fast && '1' || '0' }}
   KV_OFFLOADING: ${{ inputs.kv-offloading }}
   KV_OFFLOAD_BACKEND: ${{ inputs.kv-offload-backend }}
   KV_OFFLOAD_BACKEND_METADATA: ${{ inputs.kv-offload-backend-metadata }}

.github/workflows/benchmark-tmpl.yml

@@ -129,6 +129,11 @@ on:
         required: false
         type: string
         default: '3600'
+      agentx-fast:
+        description: "Use the AgentX 5-minute warmup and 20-minute profiling preset"
+        required: false
+        type: boolean
+        default: false
       eval-limit:
         description: "Eval instance count: empty/full = whole split (default); N = first-N smoke slice"
         required: false
@@ -174,6 +179,7 @@ env:
   KV_P2P_TRANSFER: ${{ inputs.kv-p2p-transfer }}
   TOTAL_CPU_DRAM_GB: ${{ inputs.total-cpu-dram-gb }}
   DURATION: ${{ inputs.duration }}
+  AIPERF_EXPERIMENTAL_FAST: ${{ inputs.agentx-fast && '1' || '0' }}
   EVAL_LIMIT: ${{ inputs.eval-limit }}
   SWEBENCH_GEN_MODE: ${{ inputs.swebench-gen-mode }}
   AIPERF_FAILED_REQUEST_THRESHOLD: '0.10'

.github/workflows/e2e-tests.yml

@@ -21,6 +21,11 @@ on:
                 required: false
                 type: string
                 default: ""
+            agentx-fast:
+                description: "AgentX fast feedback: 5-minute cache warmup and 20-minute profile."
+                required: false
+                type: boolean
+                default: false
             eval-limit:
                 description: "Eval instance count: empty/full = whole split (default); N = first-N smoke slice"
                 required: false
@@ -50,6 +55,11 @@ on:
                 required: false
                 type: string
                 default: ""
+            agentx-fast:
+                description: "AgentX fast feedback: 5-minute cache warmup and 20-minute profile."
+                required: false
+                type: boolean
+                default: false
             eval-limit:
                 description: "Eval instance count: empty/full = whole split (default); N = first-N smoke slice"
                 required: false
@@ -264,7 +274,8 @@ jobs:
             kv-offload-backend: ${{ matrix.config['kv-offload-backend'].name }}
             kv-offload-backend-metadata: ${{ matrix.config['kv-offload-backend'] && toJson(matrix.config['kv-offload-backend']) || '' }}
             total-cpu-dram-gb: ${{ matrix.config.total-cpu-dram-gb }}
-            duration: ${{ inputs.duration-override != '' && inputs.duration-override || matrix.config.duration }}
+            duration: ${{ inputs.agentx-fast && '1200' || (inputs.duration-override != '' && inputs.duration-override || matrix.config.duration) }}
+            agentx-fast: ${{ inputs.agentx-fast }}
             isl: '0'
             osl: '0'
             max-model-len: '0'
@@ -301,7 +312,8 @@ jobs:
             kv-offloading: ${{ matrix.config.kv-offloading }}
             kv-offload-backend: ${{ matrix.config.kv-offload-backend }}
             total-cpu-dram-gb: ${{ matrix.config.total-cpu-dram-gb }}
-            duration: ${{ inputs.duration-override != '' && inputs.duration-override || matrix.config.duration }}
+            duration: ${{ inputs.agentx-fast && '1200' || (inputs.duration-override != '' && inputs.duration-override || matrix.config.duration) }}
+            agentx-fast: ${{ inputs.agentx-fast }}
             isl: '0'
             osl: '0'
             max-model-len: '0'
@@ -364,7 +376,8 @@ jobs:
             kv-offloading: ${{ matrix.config.kv-offloading }}
             kv-offload-backend: ${{ matrix.config['kv-offload-backend'].name }}
             kv-offload-backend-metadata: ${{ matrix.config['kv-offload-backend'] && toJson(matrix.config['kv-offload-backend']) || '' }}
-            duration: ${{ inputs.duration-override != '' && inputs.duration-override || matrix.config.duration }}
+            duration: ${{ inputs.agentx-fast && '1200' || (inputs.duration-override != '' && inputs.duration-override || matrix.config.duration) }}
+            agentx-fast: ${{ inputs.agentx-fast }}
             run-eval: false
             scenario-type: agentic-coding
             ref: ${{ inputs.ref }}

AGENTS.md

@@ -119,6 +119,8 @@ gh api -X POST \
 
 Inputs: top-level `ref` (required) is the workflow ref to dispatch from, almost always `main`. `inputs[ref]` is the repo ref under test (defaults to the dispatch ref's `github.sha`). `inputs[generate-cli-command]` (required) is passed verbatim to `generate_sweep_configs.py` - test locally first. `inputs[test-name]` is the display name in the Actions UI. `inputs[duration-override]` overrides per-config duration (seconds); empty = use matrix value.
 
+For an AgentX preflight, add `-F 'inputs[agentx-fast]=true'` to the dispatch command. This selects a 5-minute cache warmup and 20-minute profile for every AgentX job in that e2e dispatch and takes precedence over `duration-override`. Use it to get faster signal while debugging, then run the official sweep without `agentx-fast`; canonical AgentX timing remains a 10-minute cache warmup and 1-hour profile.
+
 The POST returns no body and no run ID - find the run with `gh run list` below.
 
 ### Monitoring jobs

benchmarks/benchmark_lib.sh

@@ -1744,6 +1744,15 @@ build_replay_cmd() {
     # utils/aiperf/docs/tutorials/agentx-mvp.md.
     local result_dir="$1"
     local duration="$DURATION"
+    local cache_warmup_duration="${AIPERF_AGENTIC_CACHE_WARMUP_DURATION:-600}"
+
+    # Fast mode is an e2e-only feedback preset used before canonical one-hour
+    # sweeps. AIPerf already exposes both controls, so no AIPerf patch is
+    # required.
+    if [[ "${AIPERF_EXPERIMENTAL_FAST:-0}" == "1" ]]; then
+        duration=1200
+        cache_warmup_duration=300
+    fi
 
     export AIPERF_DATASET_WEKA_LIVE_ASSISTANT_RESPONSES="${AIPERF_DATASET_WEKA_LIVE_ASSISTANT_RESPONSES:-0}"
     # Dataset configuration (load + reconstruct + inputs.json + mmap)
@@ -1776,9 +1785,9 @@ build_replay_cmd() {
     REPLAY_CMD+=" --trajectory-start-min-ratio 0.25"
     REPLAY_CMD+=" --trajectory-start-max-ratio 0.75"
     # After the normal t* snapshot warmup, continue those exact trajectories
-    # with one-token outputs and no idle delays for 10 minutes. Profiling begins
-    # only after those requests drain and resumes from the resulting live state.
-    REPLAY_CMD+=" --agentic-cache-warmup-duration 600"
+    # with one-token outputs and no idle delays. Profiling begins only after
+    # those requests drain and resumes from the resulting live state.
+    REPLAY_CMD+=" --agentic-cache-warmup-duration $cache_warmup_duration"
     # Give long-context warmup requests up to 30 minutes to drain before
     # declaring warmup failed. Recipes whose saturation arms carry a larger
     # in-flight working set may override via AGENTIC_WARMUP_GRACE_PERIOD