README.md

---
license: unknown
base_model:
- apple/DiffuCoder-7B-Instruct
tags:
- code
- text-diffusion-model
- diffusion large language model
---

### DiffuCoder-7B-cpGRPO

The DiffuCoder-7B-cpGRPO variant further refines DiffuCoder-Instruct with reinforcement learning via Coupled-GRPO.

Training recipe:

- Initialized from DiffuCoder-7B-Instruct, post-training with coupled-GRPO on 21K code data (1 epoch).
- coupled-GRPO significantly improves DiffuCoder's performance on code generation benchmarks (+4.4\% on EvalPlus) and reduces reliance on AR bias during decoding.


#### More details and usage examples:

- Paper: [DiffuCoder: Understanding and Improving Masked Diffusion Models for Code Generation](https://arxiv.org/abs/2506.20639)

- GitHub: https://github.com/apple/ml-diffucoder

#### Acknowledgement
To power this HuggingFace model release, we reuse [Dream](https://huggingface.co/Dream-org/Dream-v0-Base-7B)'s modeling architecture and generation utils.