fix(benchmark): an accuracy run that produces no number must fail - #454
Open
leopck wants to merge 1 commit into
Open
fix(benchmark): an accuracy run that produces no number must fail#454leopck wants to merge 1 commit into
leopck wants to merge 1 commit into
Conversation
|
MLCommons CLA bot All contributors have signed the MLCommons CLA ✍️ ✅ |
github-actions
Bot
added
the
size/normal
leopck
force-pushed
the
fix/acc-no-score-is-failure
branch
from
August 19, 2026 13:53
e3ee97c to
8ba6509
Compare
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #454 +/- ##
=======================================
Coverage ? 81.37%
=======================================
Files ? 151
Lines ? 20475
Branches ? 0
=======================================
Hits ? 16662
Misses ? 3813
Partials ? 0 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
leopck
force-pushed
the
fix/acc-no-score-is-failure
branch
from
August 20, 2026 16:12
8ba6509 to
63e1a38
Compare
4 tasks
An accuracy run could complete every unit of work, exit 0, and report `N/A`. `finalize_benchmark()` scored, wrote `accuracy_results.json` with `score: null`, printed the summary, and returned -- so `main.py` exited 0 and every wrapper downstream read the run as a pass. This is not hypothetical. A distributed SWE-bench run drove all 20 of its units to terminal records; 17 were abandoned to infrastructure failures, the merge gate correctly refused, `score()` returned None -- and the sbatch wrapper wrote `disposition=run completed (driver rc=0, results=20/20)` over a run with no accuracy number at all. rc=0, all work "done", no number is the failure shape that costs whole GPU allocations, so make it loud where it originates instead of asking each caller to notice. `_require_accuracy_numbers()` runs last, after every artifact is on disk, so the failure never costs the evidence needed to diagnose it. A real number flagged `complete=False` (a partial headline) still passes: the number exists and the entry already says it is partial. A PERF-mode run owes nothing for an externally-scored dataset it never dispatched. Also count what was evaluated, not what was loaded, in the accuracy-only summary line: SWE-bench Verified loads all 500 rows and scores `num_instances` of them, so the run above printed "500 samples evaluated" directly beneath its own "unit=200" headline.
leopck
force-pushed
the
fix/acc-no-score-is-failure
branch
from
August 26, 2026 20:42
63e1a38 to
cee851b
Compare
arekay-nv
approved these changes
Aug 26, 2026
| entry["dataset_name"] | ||
| for entry in accuracy_scores | ||
| if entry.get("dataset_type") == DatasetType.ACCURACY.value | ||
| and entry.get("score") is not None |
Collaborator
There was a problem hiding this comment.
This is checking for None which is not the same as being a number - suggest to either harden the check or make the distinction clear in the method description/documentation.
roborluo
self-requested a review
August 27, 2026 22:24
roborluo
reviewed
Aug 27, 2026
| for ec in ctx.eval_configs: | ||
| if ec.dataset_type != DatasetType.ACCURACY: | ||
| continue | ||
| external = effective_external_sample_count(ec) |
Collaborator
There was a problem hiding this comment.
[P2]:Only to make this consistant with the notes, suggst to add
if ec.scorer.SKIP_ENDPOINT_PHASE:
before external = effective_external_sample_count(ec).
roborluo
approved these changes
Aug 27, 2026
roborluo
reviewed
Aug 28, 2026
| assert results["accuracy_scores"][0]["complete"] is False | ||
|
|
||
| @pytest.mark.unit | ||
| def test_perf_only_run_owes_no_accuracy_number(self, tmp_path): |
Collaborator
There was a problem hiding this comment.
Sugguest to delete this test, this will never happen since we already have
if not (test_mode == TestMode.PERF and scorer_cls.SKIP_ENDPOINT_PHASE):
...
eval_configs.append(...)
Maybe you want to test such scorer_cls will not be added into eval_configs?
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
4 participants
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.
An accuracy run could complete every unit of work, exit 0, and report
N/A.finalize_benchmark()scored, wroteaccuracy_results.jsonwithscore: null, printed the summary, and returned -- somain.pyexited 0 and every wrapper downstream read the run as a pass.This is not hypothetical. A distributed SWE-bench run drove all 20 of its units to terminal records; 17 were abandoned to infrastructure failures, the merge gate correctly refused,
score()returned None -- and the sbatch wrapper wrotedisposition=run completed (driver rc=0, results=20/20)over a run with no accuracy number at all. rc=0, all work "done", no number is the failure shape that costs whole GPU allocations, so make it loud where it originates instead of asking each caller to notice._require_accuracy_numbers()runs last, after every artifact is on disk, so the failure never costs the evidence needed to diagnose it. A real number flaggedcomplete=False(a partial headline) still passes: the number exists and the entry already says it is partial. A PERF-mode run owes nothing for an externally-scored dataset it never dispatched.Also count what was evaluated, not what was loaded, in the accuracy-only summary line: SWE-bench Verified loads all 500 rows and scores
num_instancesof them, so the run above printed "500 samples evaluated" directly beneath its own "unit=200" headline.What does this PR do?
Type of change