ELAD: Explanation-Guided Large Language Models Active Distillation

TL;DR

An Explanation-Guided LLMs Active Distillation (ELAD) framework that employs an active learning strategy to optimize the balance between annotation costs and model performance and a customized LLM-annotated explanation revision technique where the teacher model detects and corrects flaws in the student model's reasoning.

Abstract

The deployment and application of Large Language Models (LLMs) is hindered by their memory inefficiency, computational demands, and the high costs of API inferences. Traditional distillation methods, which transfer the capabilities of LLMs to smaller models, often fail to determine whether the knowledge has been sufficiently transferred, potentially resulting in high costs or incomplete distillation. In this paper, we propose an Explanation-Guided LLMs Active Distillation (ELAD) framework that employs an active learning strategy to optimize the balance between annotation costs and model performance. To improve efficient sample selection, we introduce an explanation-guided sample selection method that identifies samples challenging its reasoning by exploiting uncertainties in explanation steps. Additionally, we present a customized LLM-annotated explanation revision technique where the teacher model detects and corrects flaws in the student model's reasoning. Our experiments across various reasoning datasets demonstrate that our framework significantly enhances the efficiency of LLM knowledge distillation.

Figures (4)

• Figure 1: Overview of the Explanation-Guided LLM Active Distillation (ELAD) framework: (a) illustrates the Explanation-Guided Sample Selection method, (b) depicts the Customized LLM-Annotated Explanation Revision technique, and (c) showcases the LLM Knowledge Distillation (small model fine-tuning) process.
• Figure 2: (a) illustrates reasoning not conditioned on the $i$-th reasoning step; (b) depicts reasoning conditioned on the $i$-th reasoning step.
• Figure 3: Customized LLM-Annotated Explanation Revision. (a) and (b) illustrate the process by which the LLM is prompted to revise the explanation and answer from the small model. (c) shows the DFS-based reasoning steps searching strategy.
• Figure 4: Performance curves of different sample selection methods for active learning. The y-axis denotes the accuracy for the question-answering task, and the x-axis represents the percentage of samples annotated by the LLM for small model fine-tuning. In this case, 100% denotes that all samples from the training set have been annotated.