BiLD: Bi-directional Logits Difference Loss for Large Language Model Distillation

TL;DR

This paper introduces BiLD, a novel loss function for large language model distillation that filters long-tail noise and leverages internal ranking, significantly improving performance over existing methods across multiple datasets.

Contribution

The paper proposes the BiLD loss, which effectively utilizes top-k logits and internal ranking information to enhance LLM distillation performance.

Findings

BiLD outperforms existing distillation methods on 13 datasets.

Filtering long-tail noise improves distillation quality.

Using internal ranking information enhances logits utilization.

Abstract

In recent years, large language models (LLMs) have shown exceptional capabilities across various natural language processing (NLP) tasks. However, such impressive performance often comes with the trade-off of an increased parameter size, posing significant challenges for widespread deployment. Knowledge distillation (KD) provides a solution by transferring knowledge from a large teacher model to a smaller student model. In this paper, we explore the task-specific distillation of LLMs at the logit level. Our investigation reveals that the logits of fine-tuned LLMs exhibit a more extreme long-tail distribution than those from vision models, with hidden "noise" in the long tail affecting distillation performance. Furthermore, existing logits distillation methods often struggle to effectively utilize the internal ranking information from the logits. To address these, we propose the Bi-directional Logits Difference (BiLD) loss. The BiLD loss filters out the long-tail noise by utilizing only top-$k$ teacher and student logits, and leverages the internal logits ranking information by constructing logits differences. To evaluate BiLD loss, we conduct comprehensive experiments on 13 datasets using two types of LLMs. Our results show that the BiLD loss, with only the top-8 logits, outperforms supervised fine-tuning (SFT), vanilla KL loss, and five other distillation methods from both NLP and CV fields.