BabyLlama-2: Ensemble-Distilled Models Consistently Outperform Teachers With Limited Data

Re-assign community