From Completion to Editing: Unlocking Context-Aware Code Infilling via Search-and-Replace Instruction Tuning

TL;DR

This paper introduces Search-and-Replace Infilling (SRI), a novel framework that enables context-aware code editing by integrating verification and editing into a single inference pass, improving performance and safety over traditional methods.

Contribution

The paper presents SRI, a unified approach for dynamic code editing that extends code completion paradigms, with a new dataset and fine-tuned models demonstrating superior performance.

Findings

SRI enables Chat models to outperform base models in code completion.

SRI maintains inference latency comparable to traditional fill-in-the-middle methods.

SRI preserves general coding skills while enhancing context-aware editing.

Abstract

The dominant Fill-in-the-Middle (FIM) paradigm for code completion is constrained by its rigid inability to correct contextual errors and reliance on unaligned, insecure Base models. While Chat LLMs offer safety and Agentic workflows provide flexibility, they suffer from performance degradation and prohibitive latency, respectively. To resolve this dilemma, we propose Search-and-Replace Infilling (SRI), a framework that internalizes the agentic verification-and-editing mechanism into a unified, single-pass inference process. By structurally grounding edits via an explicit search phase, SRI harmonizes completion tasks with the instruction-following priors of Chat LLMs, extending the paradigm from static infilling to dynamic context-aware editing. We synthesize a high-quality dataset, SRI-200K, and fine-tune the SRI-Coder series. Extensive evaluations demonstrate that with minimal data (20k samples), SRI-Coder enables Chat models to surpass the completion performance of their Base counterparts. Crucially, unlike FIM-style tuning, SRI preserves general coding competencies and maintains inference latency comparable to standard FIM. We empower the entire Qwen3-Coder series with SRI, encouraging the developer community to leverage this framework for advanced auto-completion and assisted development.