File size: 3,947 Bytes
ff5c1d3 75d9570 ff5c1d3 75d9570 ff5c1d3 75d9570 ff5c1d3 75d9570 ff5c1d3 5544af9 ff5c1d3 5544af9 ff5c1d3 5a66bdc ff5c1d3 75d9570 ff5c1d3 702b27b ff5c1d3 75d9570 | 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 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 | ---
base_model:
- inclusionAI/LLaDA2.0-mini
datasets:
- Zigeng/DMax-LLaDA-2.0-Mini-Math-Trajectories
license: apache-2.0
library_name: transformers
pipeline_tag: text-generation
---
<div align="center">
<h1>🚀 DMax: Aggressive Parallel Decoding for dLLMs</h1>
<div align="center">
<a href="https://github.com/czg1225/DMax/blob/main/LICENSE">
<img alt="Apache" src="https://img.shields.io/badge/License-Apache-4E94CE.svg">
</a>
<a href="https://arxiv.org/abs/2604.08302">
<img src="https://img.shields.io/badge/Paper-Arxiv-darkred.svg" alt="Paper">
</a>
<a href="https://github.com/czg1225/DMax">
<img src="https://img.shields.io/badge/GitHub-Code-blue.svg?logo=github&" alt="GitHub">
</a>
</div>
</div>
This repository contains the weights for **DMax-Math-16B**, presented in the paper [DMax: Aggressive Parallel Decoding for dLLMs](https://huggingface.co/papers/2604.08302).
DMax is a new paradigm for efficient diffusion language models (dLLMs) that mitigates error accumulation in parallel decoding, enabling aggressive decoding parallelism while preserving generation quality.
## 💪 Highlights
- **Aggressive Decoding Parallelism**: Achieves 6.0 TPF on math and reasoning tasks and 6.6 TPF on code tasks while preserving accuracy.
- **Self-Revising dLLM**: Extends a pretrained MDLM into a UDLM with an intrinsic ability to revise its own erroneous predictions during decoding.
- **Soft Parallel Decoding**: Uses interpolation between mask and token embeddings to propagate confidence priors from previous steps.
<div align="center">
<img src="assets/tradeoff.png" width="100%" />
<br>
<em>Superior Parallelism-Accuracy Trade-off, Increased TPF with Maintained Accuracy.</em>
</div>
## 💻 Model and Datasets
| Model | Description | Source Model | Link |
| --- | --- | --- | --- |
| 🤖 DMax-Math-16B | Highly parallel dLLM for math and reasoning. | LLaDA-2.0-mini | [HF](https://huggingface.co/Zigeng/DMax-Math-16B) |
| 🤖 DMax-Coder-16B | Highly parallel dLLM for code generation. | LLaDA-2.0-mini | [HF](https://huggingface.co/Zigeng/DMax-Coder-16B) |
| Dataset | Description | Link |
| --- | --- | --- |
| 📊 DMax-Math-Training-Data | math trajectories generated by LLaDA-2.0-mini | [HF](https://huggingface.co/datasets/Zigeng/DMax-LLaDA-2.0-Mini-Math-Trajectories) |
| 📊 DMax-Code-Training-Data | code trajectories generated by LLaDA-2.0-mini | [HF](https://huggingface.co/datasets/Zigeng/DMax-LLaDA-2.0-Mini-Code-Trajectories) |
## 🚀 Quick Start
```python
import torch
from transformers import AutoModelForCausalLM
from transformers import AutoTokenizer
model = AutoModelForCausalLM.from_pretrained(
"Zigeng/DMax-Math-16B", trust_remote_code=True, device_map="cuda:0"
)
model = model.to(torch.bfloat16)
model.eval()
tokenizer = AutoTokenizer.from_pretrained("Zigeng/DMax-Math-16B", trust_remote_code=True)
prompt = "A robe takes 2 bolts of blue fiber and half that much white fiber. How many bolts in total does it take?" + "
Let's think step by step
"
input_ids = tokenizer.apply_chat_template(
[{"role": "user", "content": prompt}],
add_generation_prompt=True,
tokenize=True,
return_tensors="pt",
)
nfe, generated_tokens = model.generate_spd(
inputs=input_ids,
gen_length=2048,
block_length=32,
threshold=0.5,
)
generated_answer = tokenizer.decode(
generated_tokens[0],
skip_special_tokens=True,
)
print(generated_answer)
print("nfe:",nfe,"token length",len(generated_tokens[0]))
```
## 📖 Experimental Results
## 📚 Citation
```bibtex
@misc{chen2026dmaxaggressiveparalleldecoding,
title={DMax: Aggressive Parallel Decoding for dLLMs},
author={Zigeng Chen and Gongfan Fang and Xinyin Ma and Ruonan Yu and Xinchao Wang},
year={2026},
eprint={2604.08302},
archivePrefix={arXiv},
primaryClass={cs.LG},
url={https://arxiv.org/abs/2604.08302},
}
``` |