Repository navigation
Conversation
`PhaseRunner._root_idle_gap_cap_seconds` armed the per-trajectory idle watchdog only for PROFILING. An accelerated cache-pressure warmup (--warmup-requests-per-lane / --agentic-cache-warmup-duration) replays the same recorded gaps through the same replay barrier, so with the cap off it waits out every multi-minute idle period, overrunning its budget, and fills the KV cache under a different arrival process than profiling measures. Arm the cap for WARMUP as well and document it. Signed-off-by: Francesco Di Natale <3429989+FrankD412@users.noreply.github.com>
Try out this PRQuick install: pip install --upgrade --force-reinstall git+https://github.com/ai-dynamo/aiperf.git@97a1f11f5655c48fbbb35fb5b13a4d4377baef39Recommended 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@97a1f11f5655c48fbbb35fb5b13a4d4377baef39Last updated for commit: |
|
Fern Docs Preview: generation failed — see the Actions log for details. This does not block merge; ask a maintainer to retry if needed. |
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Patch from Julien Lin (authorship preserved on the commit).
Why
--trace-idle-gap-cap-seconds Sadvances a trajectory's pending timers once its root and all descendants have had nothing in flight forSseconds.PhaseRunner._root_idle_gap_cap_secondsonly armed it for the profiling phase, so the agentic warmup ran uncapped.The accelerated cache-pressure warmup (
--warmup-requests-per-lane/--agentic-cache-warmup-duration) is synthesized as anAGENTIC_REPLAYphase (src/aiperf/timing/config.py) and replays the same recorded gaps through the sameReplayBarrierCoordinator. Without the cap it:What
src/aiperf/timing/phase/runner.py:_root_idle_gap_cap_secondsreturns the dataset'strace_idle_gap_cap_secondsforWARMUPas well asPROFILING. TheAGENTIC_REPLAYand run-config checks are unchanged.docs/benchmark-modes/semianalysis-agentx-faq.md,docs/tutorials/agentx-mvp.md: state that the cap is armed in both warmup and profiling.Behavior
SsSsDuration-based warmup sends more requests in the same wall time once long gaps are skipped, so warmup request counts differ from earlier runs with the same flags.
Not in scope
ai-dynamo/aiperfmainhas the same profiling-only gate; no upstream PR carries this change yet.Verification
tests/unit/timing/: 1298 passed.