Skip to content

fix: map Harbor trial folders to eval case ids - #104

Merged
rng1995 merged 13 commits into
NVIDIA:mainfrom
mimran-khan:fix/refine-harbor-trial-folders
Sep 12, 2026
Merged

fix: map Harbor trial folders to eval case ids#104
rng1995 merged 13 commits into
NVIDIA:mainfrom
mimran-khan:fix/refine-harbor-trial-folders

Conversation

@mimran-khan

Copy link
Copy Markdown
Contributor

Summary

--refine keyed trajectories by the raw Harbor trial folder name. Collection writes case-001__Lmi47iy, and eval cases are case-001, so every lookup missed.

Discovery now splits off the __suffix and keys by the case id. A folder named case-001 still works. Fixes #93.

Verification

  • I am familiar with the Contributing Guidelines
  • Added or updated focused tests
  • Updated documentation for user-visible changes
  • Ran make lint
  • Ran make test
  • Ran make build
  • Did not add credentials, private datasets, or proprietary benchmark content

Release Impact

  • Updated CHANGELOG.md

--refine keyed trajectories by the raw trial directory name, but Harbor
writes folders like case-001__Lmi47iy. Split off the suffix so
trajectories.get(case id) actually finds them.

Fixes NVIDIA#93

Signed-off-by: mimran-khan <mohammed_imran.khan@outlook.com>

@rng1995 rng1995 left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Reviewed the exact current head against #93 and the Harbor persisted-trial naming path. Suffixed folders map back to case ids and legacy unsuffixed folders remain unchanged. Focused tests pass (18), with Ruff and diff checks clean. No actionable findings; approved. The red Gitleaks job is unrelated branch history.

@rng1995

rng1995 commented Aug 28, 2026

Copy link
Copy Markdown
Collaborator

@mimran-khan - Please resolve merge conflicts

@mimran-khan

Copy link
Copy Markdown
Contributor Author

Merged main in. Changelog conflict is resolved.

@mimran-khan

Copy link
Copy Markdown
Contributor Author

Merged latest main (including #102) and resolved the CHANGELOG conflict. Ready for another look.

@rng1995

rng1995 commented Aug 31, 2026

Copy link
Copy Markdown
Collaborator

Please resolve merge conflict in CHANGELOG.md so that I can merge it. Thanks for your contribution and patience @mimran-khan

@rng1995
rng1995 enabled auto-merge (squash) August 31, 2026 15:55
auto-merge was automatically disabled August 31, 2026 17:06

Head branch was pushed to by a user without write access

Comment thread src/skillevaluator/tier3/generate_dataset.py Outdated
Merge main and prefer reward.json entry_id when mapping persisted trial
folders to eval case ids. Use folder-name parsing only for unambiguous
Harbor suffix tails; keep full folder names when metadata is missing.
@mimran-khan

Copy link
Copy Markdown
Contributor Author

Merged main and addressed the reward entry_id feedback at 5cb7de6. Ready for re-review.

Comment thread src/skillevaluator/tier3/generate_dataset.py Outdated
mimran-khan and others added 5 commits September 3, 2026 15:51
Signed-off-by: mimran-khan <mohammed_imran.khan@outlook.com>
Stop guessing Harbor suffix shape from folder names. Prefer reward
entry_id, then result.json task metadata, and keep the full directory
name when neither is present.

Signed-off-by: mimran-khan <mohammed_imran.khan@outlook.com>
Signed-off-by: mimran-khan <mohammed_imran.khan@outlook.com>

@rng1995 rng1995 left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Re-reviewed the exact 6043dc2 head. The canonical case ID now comes from persisted reward metadata, then Harbor result metadata, without unsafe suffix guessing; supported double-underscore IDs and ShortUUID trial folders are covered. The focused suite passes (24 tests), with Ruff and diff checks clean. Approved on code merits. GitHub currently reports a conflict with main, so that still needs to be resolved and the resulting head should complete CI before merge.

@rng1995

rng1995 commented Sep 12, 2026

Copy link
Copy Markdown
Collaborator

@mimran-khan - Please resolve merge conflicts.

@rng1995
rng1995 enabled auto-merge (squash) September 12, 2026 07:35
Signed-off-by: mimran-khan <mohammed_imran.khan@outlook.com>
auto-merge was automatically disabled September 12, 2026 10:29

Head branch was pushed to by a user without write access

@rng1995
rng1995 enabled auto-merge (squash) September 12, 2026 14:33
@rng1995
rng1995 merged commit d979722 into NVIDIA:main Sep 12, 2026
17 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[BUG]: create-eval-dataset --refine cannot match Harbor trial folders to case ids

3 participants