OpenNLI-Leaderboard / evaluate.py

Commit History

16 days ago

  1. evaluate.py: premise-only truncation (the hypothesis is never cut; longest-first only if the hypothesis alone cannot fit)

    Verified signature
    bd78ad9

22 days ago

  1. Add the calibration config (11,018 validation/dev rows, 32 benchmarks); evaluate.py: Skill Index = Brier skill at a temperature fitted on it

    Verified signature
    c2d2e87
  2. evaluate.py: report chance-corrected skill (random = 0, perfect = 100)

    Verified signature
    e294207

23 days ago

  1. Add evaluate.py (scoring script) and README scoring section

    Verified signature
    3cfaf50