# Gen-Verse/MMaDA

MMaDA - Open-Sourced Multimodal Large Diffusion Language Models (dLLMs with block diffusion, mixed-CoT, unified RL)

Repository: https://github.com/Gen-Verse/MMaDA
Canonical: https://ross.abutalabs.com/products/mmada
Homepage: https://openreview.net/forum?id=wczmXLuLGd
Language: Python
License: MIT
License Family: permissive
Topics: diffusion-models, llm-reasoning, unified-multimodal-understanding-and-generation
Last push: 2026-02-14T05:50:05+00:00

## Health v2 (maintenance only)
Score: 49/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 67, release rhythm 35, longevity 33
- inputs: {"age_days": 469, "days_push": 200, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1668, forks 91 (observed 2026-08-28T04:05:19.830177+00:00)

## What it is
MMaDA is an open-source family of multimodal large diffusion language models that unify textual reasoning, multimodal understanding, and text-to-image generation under a single diffusion architecture. It includes training and inference code, mixed chain-of-thought fine-tuning, and the UniGRPO policy-gradient RL post-training algorithm.

## Use cases
- generate images from text prompts with a diffusion language model
- run a diffusion-based LLM for text reasoning tasks
- fine-tune a multimodal model with mixed chain-of-thought data
- post-train diffusion foundation models with reinforcement learning
- experiment with non-autoregressive multimodal generation
- research unified understanding and generation architectures

## When to choose
- you need a single model for both text reasoning and image generation
- you want to research or extend diffusion language models (dLLMs)
- you need an open-weights multimodal foundation model with RL training infrastructure

## When to avoid
- you need a production-ready autoregressive LLM with mature ecosystem support
- you lack GPU resources for 8B-parameter model inference or training
- you only need standard text-to-image generation without reasoning

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, llm-training, llm-inference, image-processing, nlp
- domain: large-language-models, deep-learning, artificial-intelligence, image-processing
- platform: python
- tags: diffusion-language-model, multimodal, text-to-image, block-diffusion, chain-of-thought, reinforcement-learning, unigrpo, foundation-model, natural-language-processing, gpu

## Member repositories
- Gen-Verse/MMaDA (main) score 49

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:19.830177+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:42:48.344301+00:00, confidence not recorded.
  - readme: https://github.com/Gen-Verse/MMaDA (fetched 2026-08-28T04:05:19.830177+00:00, sha e479a4a5f5ac)
  - homepage: https://openreview.net/forum?id=wczmXLuLGd (fetched 2026-08-29T11:16:03.431756+00:00, sha 77981cc9e3f2)
- Data as of 2026-08-30T08:39:29.467469+00:00.