Humanizr/Humanizer benchmark card
Humanizr/Humanizer: issues, models and the leaderboard with intervals, and a README badge.
1 public benchmark of this repository is listed. Together they hold 12 real issues from its git history, each proved by the repository's own tests. 8 runs of them are published.
A coding agent gets the broken code and the issue report, has two tries per issue, and the tests of the repository decide whether the issue is fixed.
Repository: Humanizr/Humanizer benchmark card