d2Cache - GitLite

KO | EN

gitlite — search

#python #typescript #golang #rust #deep-learning #macos #java #awesome #cli #kubernetes #security #ios #mcp #openai

d2Cache

(★ 168)

[ICLR'26] Official code of paper "d2Cache: Accelerating Diffusion-based LLMs via Dual Adaptive Caching"

File Explorer

Download Latest Version (.zip)

- d2cache.png
- framework.png
- hooks.png
- logo.png
- - blockd.yaml
  - d2cache.yaml
  - dllm.yaml
  - prefix.yaml
- - ar.yaml
  - eb_sampler.yaml
  - klass.yaml
  - pc_sampler.yaml
  - vanilla.yaml
  - wino.yaml
- - dream-base.yaml
  - dream-common.yaml
  - dream-inst.yaml
  - llada-1.5.yaml
  - llada-base.yaml
  - llada-common.yaml
  - llada-inst.yaml
  - sdar-8b-chat.yaml
  - sdar-common.yaml
- default.yaml
- eval.yaml
- gen_args.py
- code_reading_guides.md
- customization.md
- decoding_strategies.md
- kv_caching.md
- common.txt
- fix_code_eval.sh
- run_eval.sh
- - __init__.py
  - base.py
  - blockd_cache.py
  - d2cache.py
  - dllm_cache.py
  - prefix_cache.py
- - __init__.py
  - ar.py
  - klass.py
  - utils.py
  - vanilla.py
  - wino.py
- - __init__.py
  - eval_mdlm.py
- - __init__.py
  - dream_corpus.json
  - llada_corpus.json
- - __init__.py
  - common.py
  - models.py
- __init__.py
- frame.py
- - humaneval.yaml
  - humaneval_64.yaml
  - humaneval_64_instruct.yaml
  - humaneval_instruct.yaml
  - humaneval_plus.yaml
  - README.md
  - utils.py
- - math-500.yaml
  - math_instruct.yaml
  - utils.py
.env.example
.gitignore
eval.py
LICENSE
pyproject.toml
README.md

# Installation Guide

1. Get the code

git clone https://github.com/Kamichanw/d2Cache

Downloads the entire project code from GitHub to your computer.

cd d2Cache

Moves into the project folder you just downloaded.

2. Python

Easy Recommended

Prerequisites

Git Needed to download the project code from GitHub.
Python 3 On Windows, be sure to check 'Add Python to PATH' during installation.

pip install -r requirements/common.txt

Installs the Python libraries listed in requirements.txt (or similar).

✅ If it runs without errors and prints output in the terminal, it worked.

Pulled directly from this repo's README.

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

<h1 align="center">	Research dLLM </h1>
<p align="center">
Codebase for Diffusion Language Models Research
</p>
<p align="center">
<img src="./assets/logo.png" width="400">
</p>

<div align="center">
<p align="center">
      <a href="https://github.com/Kamichanw/d2Cache/stargazers">
        <img alt="GitHub stars" src="https://img.shields.io/github/stars/Kamichanw/d2Cache?color=ccf" />
      </a>
      <a href="https://github.com/Kamichanw/d2Cache/issues">
        <img alt="Issues" src="https://img.shields.io/github/issues/Kamichanw/d2Cache?color=0088ff" />
      </a>
      <a href="https://github.com/Kamichanw/d2Cache/discussions">
        <img alt="Issues" src="https://img.shields.io/github/discussions/Kamichanw/d2Cache?color=0088ff" />
      </a>
      <a href="https://github.com/Kamichanw/d2Cache/pulls">
        <img alt="GitHub pull requests" src="https://img.shields.io/github/issues-pr/Kamichanw/d2Cache?color=0088ff" />
      </a>
</p>
</div>

<hr>

**Research dLLM** is a research-focused library for Diffusion Language Models (dLLMs), providing a comprehensive collection of baseline methods (primarily KV caching and decoding strategies) for reproducible experiments.

## Why Use This Codebase?

- **Unified Evaluation Framework**: Research dLLM provides a standardized testing environment that allows users to seamlessly switch between different baseline methods. 
- **Clean and Well-Documented Code**: Research dLLM is written with a strong emphasis on clarity and readability. 
- **Active and Ongoing Maintenance**: Research dLLM is actively maintained and continuously updated. More out-of-the-box baselines will be included in the future.

# News
[25/9/30] We released [code reading guides](./docs/code_reading_guides.md), hoping this can help you to grasp our work :)

[25/10/13] Now, batch inference is supported!

[25/12/6] 🚀 We are looking for help to reproduce our experimental results using A100 GPUs. Please contact us if you can assist with the benchmarking.

[26/5/2] We now support SDAR series models, more block diffusion dLLMs will be added in the future!

# Supported Models & Methods
## Models
The following models are supported out-of-the-box:
| Model | Paper | Original Code Repo |
|:---|:---:|:---:|
| LLaDA-8B (`llada-base`, `llada-inst`) | <a href="https://arxiv.org/abs/2502.09992"><img alt="Static Badge" src="https://img.shields.io/badge/NIPS Oral-2502.09992-purple"></a> | <a href="https://github.com/ML-GSAI/LLaDA"><img alt="Static Badge" src="https://img.shields.io/github/stars/ML-GSAI/LLaDA?style=social&label=Stars"></a> |
| LLaDA-1.5 (`llada-1.5`) | <a href="https://arxiv.org/abs/2505.19223"><img alt="Static Badge" src="https://img.shields.io/badge/arXiv-2505.19223-red"></a> | <a href="https://github.com/ML-GSAI/LLaDA-1.5"><img alt="Static Badge" src="https://img.shields.io/github/stars/ML-GSAI/LLaDA-1.5?style=social&label=Stars"></a> |
| Dream-v0-7B (`dream-base`, `dream-inst`) | <a href="https://arxiv.org/abs/2508.15487"><img alt="Static Badge" src="https://img.shields.io/badge/arXiv-2508.15487-red"></a> | <a href="https://github.com/DreamLM/Dream"><img alt="Static Badge" src="https://img.shields.io/github/stars/DreamLM/Dream?style=social&label=Stars"></a> |
| SDAR-8B (`sdar-8b-chat`) | <a href="https://arxiv.org/abs/2510.06303"><img alt="Static Badge" src="https://img.shields.io/badge/arXiv-2510.06303-red"></a> | <a href="https://github.com/JetAstra/SDAR"><img alt="Static Badge" src="https://img.shields.io/github/stars/JetAstra/SDAR?style=social&label=Stars"></a> |

## KV Caching
The corresponding usages can be found [here](./docs/kv_caching.md).

| Method | Paper | Original Code Repo |
|:---|:---:|:---:|
| PrefixCache / DualCache | <a href="https://arxiv.org/abs/2505.22618 "> <img alt="Static Badge" src="https://img.shields.io/badge/ICLR-2505.22618-58C9ED"> </a> | <a href="https://github.com/NVLabs/Fast-dLLM"><img alt="Static Badge" src="https://img.shields.io/github/stars/NVLabs/Fast-dLLM?style=social&label=Stars"></a> |
| dLLM Cache | <a href="https://arxiv.org/abs/2506.06295"> <img alt="Static Badge" src="https://img.shields.io/badge/ICML-2506.06295-green"></a> | <a href="https://github.com/maomaocun/dLLM-Cache"><img alt="Static Badge" src="https://img.shields.io/github/stars/maomaocun/dLLM-Cache?style=social&label=Stars"></a> |
| d2Cache | <a href="https://arxiv.org/abs/2509.23094"> <img alt="Static Badge" src="https://img.shields.io/badge/ICLR-2509.23094-58C9ED"> </a> | This Repo |
| BlockDiffusion Cache | <a href="https://arxiv.org/abs/2510.06303"> <img alt="Static Badge" src="https://img.shields.io/badge/arXiv-2510.06303-red"></a> | <a href="https://github.com/JetAstra/SDAR"><img alt="Static Badge" src="https://img.shields.io/github/stars/JetAstra/SDAR?style=social&label=Stars"></a> |

## Decoding Strategies
The corresponding usages can be found [here](./docs/decoding_strategies.md).

| Method | Paper | Original Code Repo |
|:---|:---:|:---:|
| Auto-regressive | - | - |
| Vanilla / Semi-AR | <a href="https://arxiv.org/abs/2502.09992"><img alt="Static Badge" src="https://img.shields.io/badge/NIPS Oral-2502.09992-purple"></a> | <a href="https://github.com/ML-GSAI/LLaDA"><img alt="Static Badge" src="https://img.shields.io/github/stars/ML-GSAI/LLaDA?style=social&label=Stars"></a> |
| Parallel | <a href="https://arxiv.org/abs/2505.22618"> <img alt="Static Badge" src="https://img.shields.io/badge/ICLR-2505.22618-58C9ED"> </a> | <a href="https://github.com/NVlabs/Fast-dLLM"><img alt="Static Badge" src="https://img.shields.io/github/stars/NVLabs/Fast-dLLM?style=social&label=Stars"></a> |
| PC-Sampler | <a href="https://arxiv.org/abs/2508.13021"><img alt="Static Badge" src="https://img.shields.io/badge/ACL-2508.13021-E92529"></a> | <a href="https://github.com/NEUIR/PC-Sampler"><img alt="Static Badge" src="https://img.shields.io/github/stars/NEUIR/PC-Sampler?style=social&label=Stars"></a> |
| Certainty Prior Decoding | <a href="https://arxiv.org/abs/2509.23094"> <img alt="Static Badge" src="https://img.shields.io/badge/ICLR-2509.23094-58C9ED"> </a> | This Repo|
| KLASS | <a href="https://arxiv.org/abs/2511.05664"><img alt="Static Badge" src="https://img.shields.io/badge/NIPS Spotlight-2511.05664-purple"></a> | <a href="https://github.com/shkim0116/KLASS"><img alt="Static Badge" src="https://img.shields.io/github/stars/shkim0116/KLASS?style=social&label=Stars"></a> |
| EB-Sampler | <a href="https://arxiv.org/abs/2505.24857"><img alt="Static Badge" src="https://img.shields.io/badge/NIPS-2511.05664-purple"></a> | Not release |
| WINO | <a href="https://arxiv.org/abs/2507.18578"> <img alt="Static Badge" src="https://img.shields.io/badge/ICLR-2507.18578-58C9ED"> </a> | <a href="https://github.com/Feng-Hong/WINO-DLLM"><img alt="Static Badge" src="https://img.shields.io/github/stars/Feng-Hong/WINO-DLLM?style=social&label=Stars"></a> |
# Setup
```bash
# Create and activate the environment
conda create -n d2cache python=3.11 -y
conda activate d2cache

# Install dependencies
pip install -r requirements/common.txt

# Prepare dotenv file, and set model path manually 
cp .env.example .env
```

# Evaluation

Please check docs/ for detailed instructions on how to run evaluations with different methods.
A quick example is shown in [srcipts/run_eval.sh](./scripts/run_eval.sh).

Available models:
- llada-base: GSAI-ML/LLaDA-8B-Base
- llada-inst: GSAI-ML/LLaDA-8B-Instruct
- llada-1.5: GSAI-ML/LLaDA-1.5
- dream-base: Dream-org/Dream-v0-Base-7B
- dream-inst: Dream-org/Dream-v0-Instruct-7B
- sdar-8b-chat: JetAstra/SDAR-8B-Chat

Available datasets:
- gsm8k
- humaneval / humaneval_instruct
- math-500
- mbpp / mbpp_instruct
- ... (all tasks specified in `lm-eval` are available)

> [!IMPORTANT]
> To evaluate humaneval dataset on Dream-v0-Instruct-7B, please use its corresponding instruct variants, i.e., `humaneval_instruct`.

Additional general arguments can be specified in `configs/generation/*.yaml`. If `gen_args_script` is provided, dynamic defaults will be loaded from that script.

# Starchart

[![Star History Chart](https://api.star-history.com/svg?repos=Kamichanw/d2Cache&type=Date)](https://star-history.com/#Kamichanw/d2Cache&Date)

# Citation
If you find d²Cache or this repository useful for your research and applications, please cite using this BibTeX:
```bibtex
@article{jiang2025d2cache,
  title={d $\^{} 2$ Cache: Accelerating Diffusion-Based LLMs via Dual Adaptive Caching},
  author={Jiang, Yuchu and Cai, Yue and Luo, Xiangzhong and Fu, Jiale and Wang, Jiarui and Liu, Chonghan and Yang, Xu},
  journal={arXiv preprint arXiv:2509.23094},
  year={2025}
}
```
# Acknowledgment
We would like to thank the authors of all models and baseline methods for their excellent work and open-source contributions.

# License
This project is licensed under the Apache 2.0 License. See the [LICENSE](./LICENSE) file for details.