Principled Fine-tuning of LLMs from User-Edits: A Medley of Preference, Supervision, and Reward

TL;DR

This paper explores how to effectively fine-tune large language models using naturally generated user edits, unifying preferences, supervision, and reward signals, with theoretical bounds and an ensembling approach that improves adaptation and performance.

Contribution

It introduces a theoretical framework for learning from user edits, unifies multiple feedback types, and proposes an ensembling method that enhances LLM fine-tuning and robustness.

Findings

Ensembling from multiple feedback types outperforms individual methods.

Theoretical bounds reveal trade-offs depending on user and data characteristics.

Proposed method robustly adapts to different user-edit distributions.

Abstract

We study how to fine-tune LLMs using user-edit deployment data consisting of a set of context, an agent's response, and user edits. This deployment data is naturally generated by users in applications such as LLMs-based writing assistants and coding agents. The _natural_ origin of user edits makes it a desired source for adapting and personalizing LLMs. In this setup, there emerges a unification of various feedback types namely preferences, supervised labels, and cost that are typically studied separately in the literature. In this paper, we initiate the theoretical investigation of learning from user edits. We first derive bounds for learning algorithms that learn from each of these feedback types. We prove that these algorithms have different trade-offs depending upon the user, data distribution, and model class. We then propose a simple ensembling procedure to jointly learn from these feedback types. On two domains adapted from Gao et al. 2024, we show our ensembling procedure outperforms these methods that learn from individual feedback. Further, we show that our proposed procedure can robustly adapt to different user-edit distributions at test time.