# bytetriper/RAE

Official PyTorch Implementation of "Diffusion Transformers with Representation Autoencoders"

Repository: https://github.com/bytetriper/RAE
Canonical: https://ross.abutalabs.com/products/rae
Language: Python
License: MIT
License Family: permissive
Last push: 2026-02-25T23:07:26+00:00

## Health v2 (maintenance only)
Score: 48/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 69, release rhythm 35, longevity 24
- inputs: {"age_days": 339, "days_push": 189, "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 2001, forks 87 (observed 2026-08-28T04:06:04.360640+00:00)

## What it is
Official PyTorch implementation of 'Diffusion Transformers with Representation Autoencoders' (RAE), a two-stage image generation pipeline using frozen pretrained encoders (DINOv2, SigLIP2) with trained ViT decoders. Includes pretrained RAE decoders, LightningDiT/DiT-DH diffusion transformers, and training/sampling scripts for GPU and TPU.

## Use cases
- generate high-fidelity images with diffusion transformers
- train a diffusion model on representation autoencoder latent space
- reproduce RAE paper results in PyTorch
- run image generation sampling on TPU with TorchXLA
- download pretrained RAE and DiT checkpoints for image synthesis
- train two-stage RAE plus DiT pipeline on ImageNet

## When to choose
- you need a research-grade implementation of representation autoencoders for diffusion models
- you want pretrained DiT-based image generation weights
- you need GPU or TPU training and sampling scripts for latent diffusion transformers

## When to avoid
- you need a production-ready image generation API or service
- you want a general-purpose diffusion library rather than a specific paper's codebase
- you cannot work with research code and changing APIs

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, image-processing, llm-training
- domain: machine-learning, deep-learning, image-processing, artificial-intelligence
- platform: python
- tags: diffusion-transformers, autoencoders, pytorch, image-generation, tpu, research-code, dinov2, siglip2, dit, latent-space, gpu, linux, docker

## Member repositories
- bytetriper/RAE (main) score 48

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:04.360640+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 2026-08-30T03:02:04.632974+00:00, confidence not recorded.
  - readme: https://github.com/bytetriper/RAE (fetched 2026-08-28T04:06:04.360640+00:00, sha 5c545d4bc405)
- Data as of 2026-08-30T08:39:29.467469+00:00.