Flaming-hot Initiation with Regular Execution Sampling for Large Language Models

TL;DR

This paper introduces FIRE sampling, a novel method that improves the quality and diversity of large language model outputs during inference and training, especially for reasoning tasks, by efficiently sourcing high-quality responses.

Contribution

FIRE sampling is a simple, effective technique that enhances LLM response quality and diversity during inference and training, with analysis of its positional effects.

Findings

FIRE sampling improves inference quality and response diversity.

FIRE benefits training in the alignment stage.

Employing FIRE at different response positions impacts performance.

Abstract

Since the release of ChatGPT, large language models (LLMs) have demonstrated remarkable capabilities across various domains. A key challenge in developing these general capabilities is efficiently sourcing diverse, high-quality data. This becomes especially critical in reasoning-related tasks with sandbox checkers, such as math or code, where the goal is to generate correct solutions to specific problems with higher probability. In this work, we introduce Flaming-hot Initiation with Regular Execution (FIRE) sampling, a simple yet highly effective method to efficiently find good responses. Our empirical findings show that FIRE sampling enhances inference-time generation quality and also benefits training in the alignment stage. Furthermore, we explore how FIRE sampling improves performance by promoting diversity and analyze the impact of employing FIRE at different positions within a response.