SenseLLM/StructureCoder-7B

Alignment with Fill-In-the-Middle for Enhancing Code Generation

<p align="center"> <a href="https://arxiv.org/abs/2508.19532">📄 Paper</a> • <a href="https://github.com/SenseLLM/StructureCoder">🏠 Repo</a> • <a href="https://huggingface.co/SenseLLM/StructureCoder-7B">🤖 Models</a> </p>

Introduction

Structure splits code snippets into smaller, granular blocks, creatingmore diverse DPO pairs from the same testcases. Additionally, we introduce the Abstract Syntax Tree (AST) splitting and curriculum training method to enhance the DPO training. Please refer to our paper for more details!

[image]

<hr>

Models

Acknowledgments

We thank the following amazing projects that truly inspired us:

• —Qwen-Coder
• —APPS
• —EvalPlus
• —LiveCodeBench
• —BigCodeBench