MoDification: Mixture of Depths Made Easy

TL;DR

MoDification introduces a method to efficiently transform existing large language models into mixture of depths models, significantly reducing latency and memory usage in long-context tasks without extensive retraining.

Contribution

The paper proposes MoDification, a novel approach that enables efficient conversion of existing LLMs into mixture of depths models through architectural and data refinements.

Findings

Achieves up to 1.2x speedup in latency

Reduces memory usage by up to 1.8x

Effective across models from 3B to 70B scale

Abstract

Long-context efficiency has recently become a trending topic in serving large language models (LLMs). And mixture of depths (MoD) is proposed as a perfect fit to bring down both latency and memory. In this paper, however, we discover that MoD can barely transform existing LLMs without costly training over an extensive number of tokens. To enable the transformations from any LLMs to MoD ones, we showcase top-k operator in MoD should be promoted to threshold-p operator, and refinement to architecture and data should also be crafted along. All these designs form our method termed MoDification. Through a comprehensive set of experiments covering model scales from 3B to 70B, we exhibit MoDification strikes an excellent balance between efficiency and effectiveness. MoDification can achieve up to ~1.2x speedup in latency and ~1.8x reduction in memory compared to original LLMs especially in long-context applications.