File size: 1,631 Bytes
f08a1ab e5b2077 f08a1ab e5b2077 f08a1ab |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 |
---
language:
- en
license: apache-2.0
pipeline_tag: text-generation
library_name: transformers
---
## 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!
<hr>
## Models
| Model | Checkpoint | Size |
|:--------------------|:------------------------------------------------------------------|:-----|
| StructureCoder-1.5B | 🤗 [HF Link](https://huggingface.co/SenseLLM/StructureCoder-1.5B) | 1.5B |
| StructureCoder-3B | 🤗 [HF Link](https://huggingface.co/SenseLLM/StructureCoder-3B) | 3B |
| StructureCoder-7B | 🤗 [HF Link](https://huggingface.co/SenseLLM/StructureCoder-7B) | 7B |
## Acknowledgments
We thank the following amazing projects that truly inspired us:
- [Qwen-Coder](https://github.com/QwenLM/Qwen3-Coder)
- [APPS](https://github.com/hendrycks/apps)
- [EvalPlus](https://github.com/evalplus/evalplus)
- [LiveCodeBench](https://github.com/LiveCodeBench/LiveCodeBench)
- [BigCodeBench](https://github.com/bigcode-project/bigcodebench) |