fix(agentx): exclude auto-warmup baselines from profiling metrics

[cquil11]

Problem

AgentX auto-creates its warmup CreditPhaseConfig without phase_kind. PhasePublisher forwards None, and ServerMetricsManager interprets both boundary baselines as profiling. Phase-tagged exports consequently include warmup counter growth in profiling totals, despite declaring a profiling-only time range.

Observed in official InferenceX run 33524322076 while validating cached-token source metrics; the bug also affects existing server counters.

Fix

Set phase_kind="warmup" in the auto-warmup builder. No workload, timing, scheduling, or exporter semantics change.

Validation

• Both new start/end baseline regression cases failed before the fix and pass afterward.
• 2,104 server-metrics and timing unit tests passed.
• Pre-commit passed, including import and schema checks.

Supports InferenceX #2796 and #2797. Canonical GPU validation with the fixed client remains pending.

---

Note

Low Risk
Single metadata field on the agentic warmup config path; fixes export tagging without changing timing or scraping logic.

Overview
AgentX auto-warmup was missing phase_kind on its synthesized CreditPhaseConfig, so phase boundary baselines were treated as profiling and warmup counter growth leaked into JSON/CSV profiling totals.

The change sets phase_kind="warmup" in _build_agentic_warmup_config, aligning auto-warmup with user-declared warmup phases so ServerMetricsManager.collect_baseline tags start/end scrapes as warmup. Workload and timing behavior are unchanged.

Docs now state that AgentX auto-warmup boundary scrapes follow the same warmup exclusion as manual warmup. Regression tests cover START/END baselines for agentic auto-warmup via PhasePublisher → collect_baseline.

^{Reviewed by Cursor Bugbot for commit dcbd942. Bugbot is set up for automated code reviews on this repo. Configure here.}

[github-actions]

Try out this PR

Quick install:

pip install --upgrade --force-reinstall git+https://github.com/ai-dynamo/aiperf.git@dcbd942654ef87b357c7722837855dbccedc2353

Recommended with virtual environment (using uv):

uv venv --python 3.12 && source .venv/bin/activate
uv pip install --upgrade --force-reinstall git+https://github.com/ai-dynamo/aiperf.git@dcbd942654ef87b357c7722837855dbccedc2353

---

Last updated for commit: dcbd942 • Browse code

[github-actions]

Fern Docs Preview: generation failed — see the Actions log for details. This does not block merge; ask a maintainer to retry if needed.