RoeiG/Laya-Hebrew-V2.1
Zero-Shot Classification • 0.4B • Updated • 20 • 1
Calibrated Hebrew–English decision models: choice, yes/no and scale questions in one fast forward pass.
Note Latest. Relevance as a scale with your own rubric; he_bench 0.659, claim-form relevance 0.833. Laya v0.3.20.
Note First release (Laya 0.3.7), kept unchanged for anyone who depends on it.
Note The Hebrew–English encoder both releases are built on (Dicta).
Note The teacher that wrote and labeled the Hebrew training data (Apache-2.0).