README.md

---
license: apache-2.0
language:
- en
tags:
- reinforcement-learning
- code-generation
- dllm
- bgpo
- llada
size_categories:
- 8B
---

# LLaDA-8B-BGPO-code

[![Paper](https://img.shields.io/badge/Paper-arXiv:2510.11683-red)](https://arxiv.org/abs/2510.11683)
[![Code](https://img.shields.io/badge/Code-GitHub-blue)](https://github.com/THU-KEG/BGPO)

## Model Description

**LLaDA-8B-BGPO-code** is an 8-billion parameter diffusion large language model (dLLM) that was trained on LLaDA-8B-Instruct using Boundary-Guided Policy Optimization (BGPO) for enhanced code generation capabilities.

## Model Details

- **Model Type**: Diffusion Large Language Model (dLLM)
- **Parameters**: 8 billion
- **Training Method**: Boundary-Guided Policy Optimization (BGPO)
- **Base Model**: LLaDA-8B-Instruct
- **Task**: Code generation
- **Language**: English

## Training Details

- **Training Epochs**: 5 epochs (112 steps per epoch)
- **Total Steps**: 560 steps
- **Response Length**: 512 tokens
- **Train Diffusion Steps**: 512
- **Eval Diffusion Steps**: 512
- **Block Size**: 32
- **Monte Carlo Sample Size ($n_t$)**: 16
- **Learning Rate**: 5e-7
- **Batch Size**: 16
- **Framework**: Built on VeRL (Volcengine Reinforcement Learning)

## Usage & Limitations

- Primarily designed for code generation tasks.
- Performance may vary on other tasks.
- Requires appropriate computational resources for inference.