[FIX] Do not charge baseline trials against the optimization n_trials budget (+ test)#21
Open
VincentG1234 wants to merge 2 commits into
Open
[FIX] Do not charge baseline trials against the optimization n_trials budget (+ test)#21VincentG1234 wants to merge 2 commits into
n_trials budget (+ test)#21VincentG1234 wants to merge 2 commits into
Conversation
Signed-off-by: Vincent Gimenes <vincent.gimenes@gmail.com>
3 tasks
VincentG1234
added a commit
that referenced
this pull request
May 20, 2026
## Summary Add structured documentation for human maintainers and coding agents: `AGENTS.md`, `.ai/context/`, `.ai/skills/`, and a user-facing architecture guide with Mermaid diagrams. Link the new doc from README and quick start. Align the local example YAML with vLLM V1 constraints. ## Why The InseeFrLab fork needs a stable onboarding path for contributors and agents (local backend focus, safe commands, known issues/PRs) without reading the whole codebase. Architecture was previously only implicit in code; diagrams improve onboarding and reviews. ## What changed - `AGENTS.md` — entry point for agents (context files, skills, safe commands). - `.ai/context/` — repo map, execution flow, history, known issues, current work snapshot, external links. - `.ai/skills/` — pr-writer, pr-reviewer, test-writer, docs-writer, architecture-diagrams. - `docs/architecture.md` — end-to-end, layout, orchestration, trial lifecycle, outputs (Mermaid). - `README.md`, `docs/quick_start.md` — links to architecture doc. - `examples/study_config_local_exec.yaml` — disable `max_num_partial_prefills` (unsupported on V1; comment added). ## How tested - [x] `ruff check .` - [x] `pytest -v tests/` - [x] Manual E2E (maintainer): not required (docs-only PR) ## Risks / limitations - `.ai/context/current-work.md` is a point-in-time snapshot (open PRs #13, #17, #21, #22); it will drift until refreshed after merges. - Mermaid rendering depends on the viewer (GitHub, IDE); no runtime behavior change except the example YAML default. ## Links - (none — no issue closed) Signed-off-by: Vincent Gimenes <vincent.gimenes@gmail.com>
Contributor
There was a problem hiding this comment.
Pull request overview
Fixes premature stopping of optimization runs by ensuring baseline trials don’t consume the configured optimization n_trials budget. This aligns the synchronous baseline execution path with the async accounting logic that already counts only optimization trials.
Changes:
- Stop incrementing
StudyController.completed_trialsduring synchronous baseline execution so only optimization trials count towardn_trials. - Clarify
_run_baseline_trialsdocstring to staten_trialsapplies to optimization trials only. - Add a unit test with a fake backend to ensure baseline runs do not increment
completed_trials.
Reviewed changes
Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.
| File | Description |
|---|---|
| auto_tune_vllm/core/study_controller.py | Removes baseline increments against completed_trials and documents optimization-only budgeting. |
| tests/core/test_study_controller.py | Adds a regression test ensuring baseline runs don’t affect the optimization trial counter. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
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.
Summary
Baseline runs should not consume the configured optimization trial budget (
n_trials). The synchronous path in_run_baseline_trialsstill incrementedcompleted_trials, which caused the study to stop early (e.g. an incomplete grid). This now matches_collect_completed_trials, which already counts onlyoptimizationtrials.Changes
auto_tune_vllm/core/study_controller.pycompleted_trials += 1after each baseline finishes (success, failure, timeout, exception).n_trialsapplies to optimization trials only.tests/core/test_study_controller.py(generic module for futureStudyControllertests)concurrency_levels→completed_trialsstays0after_run_baseline_trials.Expected behavior after merge
n_trials(and grid cardinality in grid mode) without manually adding+1/+kfor baselines.concurrency_levelsstill mean multiple baseline Optuna trials, but they do not eat the optimization budget.How to test
pytest tests/core/test_study_controller.py