CoditT5: Pretraining for Source Code and Natural Language Editing

TL;DR

CoditT5 introduces a novel pretraining approach explicitly modeling edits, enabling improved performance on software editing tasks like bug fixing and code review, outperforming standard models and achieving state-of-the-art results.

Contribution

The paper presents CoditT5, a large language model pretrained with an edit-specific objective for software editing tasks, a novel approach compared to traditional generation models.

Findings

Outperforms standard generation models on editing tasks

Achieves state-of-the-art results with reranking strategies

Demonstrates the effectiveness of edit-focused pretraining for code and language editing

Abstract

Pretrained language models have been shown to be effective in many software-related generation tasks; however, they are not well-suited for editing tasks as they are not designed to reason about edits. To address this, we propose a novel pretraining objective which explicitly models edits and use it to build CoditT5, a large language model for software-related editing tasks that is pretrained on large amounts of source code and natural language comments. We fine-tune it on various downstream editing tasks, including comment updating, bug fixing, and automated code review. By outperforming standard generation-based models, we demonstrate the generalizability of our approach and its suitability for editing tasks. We also show how a standard generation model and our edit-based model can complement one another through simple reranking strategies, with which we achieve state-of-the-art performance for the three downstream editing tasks.