README.md

---
library_name: transformers
license: mit
pipeline_tag: text-generation
---

# LLaDA 1.5: Variance-Reduced Preference Optimization for Large Language Diffusion Models

We introduce LLaDA 1.5, a competitive large diffusion language model, trained by variance-reduced preference optimization (VRPO), as presented in the paper [LLaDA 1.5: Variance-Reduced Preference Optimization for Large Language Diffusion Models](https://huggingface.co/papers/2505.19223).

Compared with LLaDA-8B-Instruct, LLaDA 1.5 achieves better performance on a wide range of tasks, including Math, Code, and Alignment tasks.

[Project Page](https://ml-gsai.github.io/LLaDA-1.5-Demo/)

[Code](https://github.com/ML-GSAI/LLaDA-1.5)

<div style="display: flex; justify-content: center; align-items: center; width: 100%; margin: 0 auto;">
    <img src="https://github.com/ML-GSAI/LLaDA-1.5/raw/main/assets/llada_1_5.png" style="width: 50%; display: block; margin: 0 auto;" />
</div>

## Inference

The LLaDA 1.5 model is available on [Huggingface](https://huggingface.co/GSAI-ML/LLaDA-1.5). Please employ the [transformers](https://huggingface.co/docs/transformers/index) to load.

```python
from transformers import AutoModel, AutoTokenizer
import torch

tokenizer = AutoTokenizer.from_pretrained('GSAI-ML/LLaDA-1.5', trust_remote_code=True)
model = AutoModel.from_pretrained('GSAI-ML/LLaDA-1.5', trust_remote_code=True, torch_dtype=torch.bfloat16)
```

The model is based on LLaDA-8B-Instruct, you can use the code for [LLaDA-8B-Instruct](https://github.com/ML-GSAI/LLaDA/blob/main/generate.py) to inference.

## Citation

Please consider cite:

```bibtex
@article{zhu2025llada,
  title={LLaDA 1.5: Variance-Reduced Preference Optimization for Large Language Diffusion Models},
  author={Zhu, Fengqi and Wang, Rongzhen and Nie, Shen and Zhang, Xiaolu and Wu, Chunwei and Hu, Jun and Zhou, Jun and Chen, Jianfei and Lin, Yankai and Wen, Ji-Rong and others},
  journal={arXiv preprint arXiv:2505.19223},
  year={2025}
}
```