README.md

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

# Introduction to TraDo

[Paper](https://huggingface.co/papers/2509.06949) | [Code](https://github.com/Gen-Verse/dLLM-RL) | [Project Page](https://huggingface.co/collections/Gen-Verse/trado-series-68beb6cd6a26c27cde9fe3af)

We introduce **TraDo**, SOTA diffusion language model, trained with **TraceRL**.

*   **TraDo-4B-Instruct** and **TraDo-8B-Instruct** outperform similarly sized strong AR models across math reasoning tasks.
*   **TraDo-8B-Thinking** is the first Long-CoT diffusion language model.

<p align="center">
  <img src="https://github.com/yinjjiew/Data/raw/main/dllm-rl/figure1.png" width="100%"/>
</p>


<p align="center">
  <img src="https://github.com/yinjjiew/Data/raw/main/dllm-rl/maintable.png" width="100%"/>
</p>

## Sample Usage

You can download and try our model:
```python
from transformers import AutoModelForCausalLM, AutoTokenizer
from generate import block_diffusion_generate

model_name = "Gen-Verse/TraDo-8B-Instruct"

model = AutoModelForCausalLM.from_pretrained(
    model_name, trust_remote_code=True, torch_dtype="float16", device_map="cuda"
)
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)

prompt = "What's the solution of x^2 - 2x + 1 = 0\
Please reason step by step, and put your final answer within \\\\boxed{}.\
"
messages = [{"role": "user", "content": prompt}]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)

tokens = tokenizer.batch_encode_plus([text], return_tensors='pt', padding=True, truncation=True, max_length=200)
tokens = {k: v.to(model.device) for k, v in tokens.items()}

output_ids = block_diffusion_generate(
    model,
    prompt=tokens,
    mask_id=151669,
    gen_length=200,
    block_length=4, denoising_steps=4,
    temperature=1.0, top_k=0, top_p=1.0,
    remasking_strategy="low_confidence_dynamic",
    confidence_threshold=0.9
)

output_text = tokenizer.decode(output_ids[0], skip_special_tokens=False)
cleaned_text = output_text.replace('<|MASK|>', '').replace('<|endoftext|>', '')
print(cleaned_text)
```

# Citation

```
@article{wang2025trado,
  title={Revolutionizing Reinforcement Learning Framework for Diffusion Large Language Models},
  author={Wang, Yinjie and Yang, Ling and Li, Bowen and Tian, Ye and Shen, Ke and Wang, Mengdi},
  journal={arXiv preprint arXiv:2509.06949},
  year={2025}
}
```

## Acknowledgement

This work is heavily built on the following open-source models:

[SDAR](https://github.com/JetAstra/SDAR), [Dream](https://github.com/DreamLM/Dream), [LLaDA](https://github.com/ML-GSAI/LLaDA), [MMaDA](https://github.com/Gen-Verse/MMaDA/tree/main), and [Diffu-coder](https://github.com/apple/ml-diffuCoder). 

these acceleration methods (engines):

[Fast-dllm](https://github.com/NVlabs/Fast-dLLM/tree/main), [jetengine](https://github.com/Labman42/JetEngine/tree/0ddc55ad3fb712b6374515b78d656f420e1a7243),

and theoretical foundations:

[MDLM](https://arxiv.org/pdf/2406.07524), [DiffuLLaMA](https://arxiv.org/abs/2410.17891), [Block Diffusion](https://arxiv.org/abs/2503.09573).