Fetching the paper…

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