Predicting Rewards Alongside Tokens: Non-disruptive Parameter Insertion for Efficient Inference Intervention in Large Language Model

TL;DR

This paper introduces Otter, a method that inserts extra parameters into large language models to predict calibration signals, improving safety and reliability during inference with minimal overhead and easy integration.

Contribution

Otter is a novel non-disruptive parameter insertion technique that enhances LLM inference by predicting calibration signals without significant computational or memory costs.

Findings

Achieves state-of-the-art performance on multiple tasks.

Reduces space overhead by up to 86.5%.

Reduces inference time by up to 98.5%.

Abstract

Transformer-based large language models (LLMs) exhibit limitations such as generating unsafe responses, unreliable reasoning, etc. Existing inference intervention approaches attempt to mitigate these issues by finetuning additional models to produce calibration signals (such as rewards) that guide the LLM's decoding process. However, this solution introduces substantial time and space overhead due to the separate models required. This work proposes Non-disruptive parameters insertion (Otter), inserting extra parameters into the transformer architecture to predict calibration signals along with the original LLM output. Otter offers state-of-the-art performance on multiple demanding tasks while saving up to 86.5\% extra space and 98.5\% extra time. Furthermore, Otter seamlessly integrates with existing inference engines, requiring only a one-line code change, and the original model response remains accessible after the parameter insertion. Our code is publicly available at \url{https://github.com/chenhan97/Otter}