This repository is a comprehensive, curated collection of resources for Diffusion Language Models (DLMs), supporting the survey paper "A Survey on Diffusion Language Models." It targets researchers and practitioners seeking to navigate the evolving DLM landscape, offering a structured overview of key advancements, models, and applications. Its primary benefit is providing a centralized, organized entry point into this complex research area.

How It Works

The repository categorizes DLM research into foundational concepts, discrete, continuous, and multimodal DLMs, alongside training strategies, inference optimization, and applications. A timeline visualizes DLM development, highlighting the increasing prominence of discrete models. It functions as an annotated bibliography and pointer to significant papers, code, and experimental playgrounds.

Quick Start & Requirements

This repository is a curated list of research papers and projects, not a software package. Therefore, there are no installation commands or specific prerequisites for the repository itself.

Highlighted Details

• Categorization: Comprehensive breakdown of DLMs into discrete, continuous, and multimodal types, with sections for foundations, training, inference, and applications.
• Timeline: Visualizes DLM development, noting trends like the rise of discrete models.
• Key Areas: Covers training strategies, inference optimization (e.g., KV cache, parallel decoding), and applications like code generation and protein design.
• Resource Pointers: Features links to "Playground" projects and "Must-Read" papers for exploration and deep dives.

Maintenance & Community

The repository is linked to an academic survey paper. Several related "Awesome" lists and project repositories are cited, indicating community interest. No direct community channels (Discord, Slack) are provided.

Licensing & Compatibility

The license for the VILA-Lab/Awesome-DLMs repository itself is not specified. Users must consult the licenses of individual linked papers, code, and projects for usage rights.

Limitations & Caveats

As a curated list, this repository offers no runnable code or direct tooling. Its value lies in its organization as a reference. The rapid evolution of DLMs necessitates frequent updates to maintain currency.