FIRST: Teach A Reliable Large Language Model Through Efficient Trustworthy Distillation

TL;DR

This paper introduces FIRST, a new distillation method that improves the trustworthiness of large language models by reducing mis-calibration and enhancing accuracy efficiently using concentrated knowledge transfer.

Contribution

We propose a novel distillation approach called FIRST that leverages concentrated knowledge and trustworthy maximization to produce reliable, well-calibrated language models more efficiently.

Findings

Achieves +2.3% accuracy improvement

Reduces mis-calibration by 10%

Effective in both in-domain and out-of-domain scenarios

Abstract

Large language models (LLMs) have become increasingly prevalent in our daily lives, leading to an expectation for LLMs to be trustworthy -- - both accurate and well-calibrated (the prediction confidence should align with its ground truth correctness likelihood). Nowadays, fine-tuning has become the most popular method for adapting a model to practical usage by significantly increasing accuracy on downstream tasks. Despite the great accuracy it achieves, we found fine-tuning is still far away from satisfactory trustworthiness due to "tuning-induced mis-calibration". In this paper, we delve deeply into why and how mis-calibration exists in fine-tuned models, and how distillation can alleviate the issue. Then we further propose a brand new method named Efficient Trustworthy Distillation (FIRST), which utilizes a small portion of teacher's knowledge to obtain a reliable language model in a cost-efficient way. Specifically, we identify the "concentrated knowledge" phenomenon during distillation, which can significantly reduce the computational burden. Then we apply a "trustworthy maximization" process to optimize the utilization of this small portion of concentrated knowledge before transferring it to the student. Experimental results demonstrate the effectiveness of our method, where better accuracy (+2.3%) and less mis-calibration (-10%) are achieved on average across both in-domain and out-of-domain scenarios, indicating better trustworthiness.