Don't Ignore the Tail: Decoupling top-K Probabilities for Efficient Language Model Distillation

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Dasgupta, Sayantan, Cohn, Trevor, Baldwin, Timothy
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866913100557451264
author Dasgupta, Sayantan
Cohn, Trevor
Baldwin, Timothy
author_facet Dasgupta, Sayantan
Cohn, Trevor
Baldwin, Timothy
contents The core learning signal used in language model distillation is the standard Kullback-Leibler (KL) divergence between the student and teacher distributions. Traditional KL divergence tends to be dominated by the next tokens with the highest probabilities, i.e., the teacher's modes, thereby diminishing the influence of less probable yet potentially informative components of the output distribution. We propose a new tail-aware divergence that decouples the contribution of the teacher model's top-K predicted probabilities from that of lower-probability predictions, while maintaining the same computational profile as the KL Divergence. Our decoupled approach reduces the impact of the teacher modes and, consequently, increases the contribution of the tail of the distribution. Experimental results demonstrate that our modified distillation method yields competitive performance in both pre-training and supervised distillation of decoder models across various datasets. Furthermore, the distillation process is efficient and can be performed with a modest academic budget for large datasets, eliminating the need for industry-scale computing.
format Preprint
id arxiv_https___arxiv_org_abs_2602_20816
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Don't Ignore the Tail: Decoupling top-K Probabilities for Efficient Language Model Distillation
Dasgupta, Sayantan
Cohn, Trevor
Baldwin, Timothy
Computation and Language
Machine Learning
The core learning signal used in language model distillation is the standard Kullback-Leibler (KL) divergence between the student and teacher distributions. Traditional KL divergence tends to be dominated by the next tokens with the highest probabilities, i.e., the teacher's modes, thereby diminishing the influence of less probable yet potentially informative components of the output distribution. We propose a new tail-aware divergence that decouples the contribution of the teacher model's top-K predicted probabilities from that of lower-probability predictions, while maintaining the same computational profile as the KL Divergence. Our decoupled approach reduces the impact of the teacher modes and, consequently, increases the contribution of the tail of the distribution. Experimental results demonstrate that our modified distillation method yields competitive performance in both pre-training and supervised distillation of decoder models across various datasets. Furthermore, the distillation process is efficient and can be performed with a modest academic budget for large datasets, eliminating the need for industry-scale computing.
title Don't Ignore the Tail: Decoupling top-K Probabilities for Efficient Language Model Distillation
topic Computation and Language
Machine Learning
url https://arxiv.org/abs/2602.20816