3 accepted papers
The core learning signal used in language model distillation is the standard Kullback-Leibler (KL) divergence between the distribution of the student and the teacher. Traditional KL divergence tends to be dominated by the teacher’s highest-probability modes, thus diminishing the influence of less pr…
Hidden State Matching is shown to improve knowledge distillation of language models by encouraging similarity between a student and its teacher's hidden states, as demonstrated by DistilBERT and its successors. This typically uses a cosine loss, which restricts the dimensionality of the student to t…
Knowledge distillation (KD) involves training a small “student” model to replicate the strong performance of a high-capacity “teacher” model, enabling efficient deployment in resource-constrained settings. Top-performing methods tend to be task- or architecture-specific and lack generalizability. Se…