# LTH14/mar

PyTorch implementation of MAR+DiffLoss https://arxiv.org/abs/2406.11838

Repository: https://github.com/LTH14/mar
Canonical: https://ross.abutalabs.com/products/mar
Language: Python
License: MIT
License Family: permissive
Last push: 2026-02-20T16:07:51+00:00

## Health v2 (maintenance only)
Score: 54/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 68, release rhythm 35, longevity 55
- inputs: {"age_days": 770, "days_push": 194, "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 1949, forks 124 (observed 2026-08-28T04:05:58.109114+00:00)

## What it is
Official PyTorch implementation of MAR (Masked Autoregressive) image generation with DiffLoss, from the NeurIPS 2024 paper 'Autoregressive Image Generation without Vector Quantization'. It includes pre-trained class-conditional models on ImageNet 256x256, training/evaluation scripts, and a Colab demo.

## Use cases
- generate images autoregressively without vector quantization
- run pre-trained MAR models on ImageNet class-conditional generation
- train MAR with DiffLoss using PyTorch DDP
- reproduce results from the MAR paper
- evaluate FID and Inception Score for image generation models
- try image generation in a Colab notebook

## When to choose
- you want to experiment with or reproduce state-of-the-art autoregressive image generation without VQ tokenizers
- you need pre-trained ImageNet 256x256 class-conditional generation models
- you want a simple PyTorch reference implementation of MAR and DiffLoss

## When to avoid
- you need text-to-image or general-purpose image generation rather than class-conditional ImageNet generation
- you lack GPU resources for training or sampling
- you need a production-ready image generation service rather than research code

## Facets
- artifact type: library
- maturity: stable
- function: machine-learning, deep-learning, image-processing
- domain: deep-learning, computer-vision, image-processing, artificial-intelligence
- platform: python
- tags: pytorch, image-generation, autoregressive-model, diffusion-loss, diffloss, imagenet, research-code, pretrained-models, gpu

## Member repositories
- LTH14/mar (main) score 54

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:58.109114+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 2026-08-30T03:06:53.960625+00:00, confidence not recorded.
  - readme: https://github.com/LTH14/mar (fetched 2026-08-28T04:05:58.109114+00:00, sha 3aa35ee85434)
- Data as of 2026-08-30T08:39:29.467469+00:00.