# apple/ml-4m

4M: Massively Multimodal Masked Modeling

Repository: https://github.com/apple/ml-4m
Canonical: https://ross.abutalabs.com/products/ml-4m
Homepage: https://4m.epfl.ch
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2025-06-02T22:35:20+00:00

## Health v2 (maintenance only)
Score: 35/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 24, release rhythm 35, longevity 62
- inputs: {"age_days": 877, "days_push": 457, "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 1808, forks 114 (observed 2026-08-28T04:05:39.286411+00:00)

## What it is
4M is a framework from Apple and EPFL for training any-to-any multimodal foundation models using masked modeling over discrete tokens across tens of modalities. It includes official implementations, training code, tokenizers, and pre-trained models (4M-7, 4M-21) for versatile vision and multimodal generation tasks.

## Use cases
- train any-to-any multimodal foundation models
- run a single vision model across tens of tasks and modalities
- generate images conditioned on text or other modalities
- perform multimodal image editing and in-painting
- fine-tune a generalist vision model for downstream tasks
- tokenize images, text, geometry, and semantics into discrete tokens
- download pretrained multimodal transformer checkpoints

## When to choose
- you need one unified model handling many vision modalities and tasks
- you want to experiment with multimodal masked modeling research
- you need steerable multimodal generation or editing capabilities
- you want pretrained any-to-any vision models with open weights

## When to avoid
- you need a lightweight single-task vision model
- you lack GPU resources for large transformer training or inference
- you need production-ready multimodal APIs rather than a research framework
- your use case is text-only NLP

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, image-processing, nlp, data-science
- domain: machine-learning, computer-vision, deep-learning, artificial-intelligence, image-processing
- platform: python
- tags: multimodal, masked-modeling, foundation-models, transformer, any-to-any, tokenization, pretrained-models, vision-models, generative-models, pytorch, gpu, linux

## Member repositories
- apple/ml-4m (main) score 35

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:39.286411+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 2026-08-30T03:21:21.858574+00:00, confidence not recorded.
  - readme: https://github.com/apple/ml-4m (fetched 2026-08-28T04:05:39.286411+00:00, sha cd7298de3e48)
  - homepage: https://4m.epfl.ch (fetched 2026-08-29T11:00:20.261548+00:00, sha c8386effc6c8)
- Data as of 2026-08-30T08:39:29.467469+00:00.