Effective Distillation to Hybrid xLSTM Architectures

TL;DR

This paper presents a novel distillation pipeline for xLSTM architectures that effectively compresses large language models, often matching or surpassing teacher performance on downstream tasks, promoting energy efficiency.

Contribution

Introduces an effective distillation process with a merging stage for xLSTM models, enabling lossless compression of large language models from popular families.

Findings

xLSTM students recover most teacher performance

Students outperform teachers on some tasks

Pipeline applicable to multiple model families

Abstract

There have been numerous attempts to distill quadratic attention-based large language models (LLMs) into sub-quadratic linearized architectures. However, despite extensive research, such distilled models often fail to match the performance of their teacher LLMs on various downstream tasks. We set out the goal of lossless distillation, which we define in terms of tolerance-corrected Win-and-Tie rates between student and teacher on sets of tasks. To this end, we introduce an effective distillation pipeline for xLSTM-based students. We propose an additional merging stage, where individually linearized experts are combined into a single model. We show the effectiveness of this pipeline by distilling base and instruction-tuned models from the Llama, Qwen, and Olmo families. In many settings, our xLSTM-based students recover most of the teacher's performance, and even exceed it on some downstream tasks. Our contributions are an important step towards more energy-efficient and cost-effective replacements for transformer-based LLMs.