Optimized Multi-Token Joint Decoding with Auxiliary Model for LLM Inference

TL;DR

This paper introduces MTAD, a novel decoding framework that accelerates multi-token joint decoding in large language models, reducing perplexity, improving performance, and saving energy by using an auxiliary model for efficient approximation.

Contribution

The paper proposes MTAD, a new auxiliary-model-based framework that enhances multi-token joint decoding efficiency and effectiveness in large language models.

Findings

MTAD reduces perplexity by 21.2%.

MTAD achieves a 1.42x speed-up over conventional methods.

MTAD consumes 1.54x less energy than traditional speculative decoding.

Abstract

Large language models (LLMs) have achieved remarkable success across diverse tasks, yet their inference processes are hindered by substantial time and energy demands due to single-token generation at each decoding step. While previous methods such as speculative decoding mitigate these inefficiencies by producing multiple tokens per step, each token is still generated by its single-token distribution, thereby enhancing speed without improving effectiveness. In contrast, our work simultaneously enhances inference speed and improves the output effectiveness. We consider multi-token joint decoding (MTJD), which generates multiple tokens from their joint distribution at each iteration, theoretically reducing perplexity and enhancing task performance. However, MTJD suffers from the high cost of sampling from the joint distribution of multiple tokens. Inspired by speculative decoding, we introduce multi-token assisted decoding (MTAD), a novel framework designed to accelerate MTJD. MTAD leverages a smaller auxiliary model to approximate the joint distribution of a larger model, incorporating a verification mechanism that not only ensures the accuracy of this approximation, but also improves the decoding efficiency over conventional speculative decoding. Theoretically, we demonstrate that MTAD closely approximates exact MTJD with bounded error. Empirical evaluations using Llama-2 and OPT models ranging from 13B to 70B parameters across various tasks reveal that MTAD reduces perplexity by 21.2% and improves downstream performance compared to standard single-token sampling. Furthermore, MTAD achieves a 1.42x speed-up and consumes 1.54x less energy than conventional speculative decoding methods. These results highlight MTAD's ability to make multi-token joint decoding both effective and efficient, promoting more sustainable and high-performance deployment of LLMs.