Baby Llama: knowledge distillation from an ensemble of teachers trained on a small dataset with no performance penalty CoNLL

Inar Timiryasov | Jean-Loup Tastet |