# willisma/SiT

Official PyTorch Implementation of "SiT: Exploring Flow and Diffusion-based Generative Models with Scalable Interpolant Transformers"

Repository: https://github.com/willisma/SiT
Canonical: https://ross.abutalabs.com/products/sit
Homepage: https://scalable-interpolant.github.io/
Language: Python
License: MIT
License Family: permissive
Last push: 2025-12-22T02:48:45+00:00

## Health v2 (maintenance only)
Score: 53/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 58, release rhythm 35, longevity 72
- inputs: {"age_days": 1009, "days_push": 254, "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 1206, forks 81 (observed 2026-08-28T04:03:59.346726+00:00)

## What it is
Official PyTorch implementation of Scalable Interpolant Transformers (SiT), a family of generative models built on Diffusion Transformers that use flexible interpolant frameworks connecting distributions. Includes model definitions, pre-trained ImageNet 256x256 checkpoints, and DDP training/sampling scripts.

## Use cases
- generate images with a pretrained interpolant transformer
- train a flow-based generative model on ImageNet
- compare flow and diffusion generative model design choices
- sample class-conditional images from pretrained checkpoints
- research interpolant frameworks for generative modeling
- reproduce FID-50K 2.06 benchmark results

## When to choose
- you want state-of-the-art class-conditional ImageNet generation with a DiT-style backbone
- you need a flexible interpolant framework to study flow vs diffusion design choices
- you want pretrained checkpoints for image generation experiments

## When to avoid
- you need text-to-image generation rather than class-conditional ImageNet synthesis
- you lack GPU resources for training large transformer models
- you need a production-ready image generation service rather than research code

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, image-processing
- domain: deep-learning, machine-learning, image-processing, artificial-intelligence
- platform: python, cross-platform
- tags: diffusion-models, flow-matching, interpolant-transformers, generative-models, pytorch, image-generation, research-code, dit, gpu, linux

## Member repositories
- willisma/SiT (main) score 53

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:59.346726+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 2026-08-30T06:19:19.517414+00:00, confidence not recorded.
  - readme: https://github.com/willisma/SiT (fetched 2026-08-28T04:03:59.346726+00:00, sha 9dad873faff6)
  - homepage: https://scalable-interpolant.github.io/ (fetched 2026-08-29T12:27:02.933881+00:00, sha 89b264209d23)
- Data as of 2026-08-30T08:39:29.467469+00:00.