CharED: Character-wise Ensemble Decoding for Large Language Models

TL;DR

CharED introduces a novel character-wise ensemble decoding method that combines multiple large language models at inference time, improving performance across various domains without requiring shared vocabularies or fine-tuning.

Contribution

The paper presents CharED, a new inference-time ensembling algorithm that averages character distributions from multiple LLMs, overcoming limitations of traditional ensembling methods.

Findings

Improved performance on coding, math, and toxicity benchmarks.

Effective combination of models regardless of vocabulary or tokenization.

Enhances capabilities of LLM ensembles without additional training.

Abstract

Large language models (LLMs) have shown remarkable potential for problem solving, with open source models achieving increasingly impressive performance on benchmarks measuring areas from logical reasoning to mathematical ability. Ensembling models can further improve capabilities across a variety of domains. However, conventional methods of combining models at inference time such as shallow fusion necessitate a shared vocabulary and tokenization, and alternatives like fine-tuning for domain-specific performance are both time consuming and computationally expensive. We therefore present an inference-time ensembling algorithm aimed at "averaging" outputs from multiple LLMs and illustrate its improved performance across multiple domains compared to its constituent models alone. Character-wise ensemble decoding, CharED, finds the marginal distribution of each character for an individual model and performs a weighted average to generate an output, character by character. In coding, math, and toxicity benchmarks, we find our proposed model able to combine complimentary strengths of multiple LLMs, regardless of vocabulary, tokenization, or model size.