# libffcv/ffcv

FFCV: Fast Forward Computer Vision (and other ML workloads!)

Repository: https://github.com/libffcv/ffcv
Canonical: https://ross.abutalabs.com/products/ffcv
Homepage: https://ffcv.io
Language: Python
License: Apache-2.0
License Family: permissive
Topics: machine-learning, data-science, pytorch
Last push: 2024-06-16T15:59:22+00:00

## Health v2 (maintenance only)
Score: 23/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 0, release rhythm 8, longevity 100
- inputs: {"age_days": 1785, "days_push": 808, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2993, forks 181 (observed 2026-08-28T04:07:35.871348+00:00)

## What it is
FFCV is a fast data loading system for PyTorch that dramatically increases data throughput in model training by replacing standard data loaders with an optimized binary format and asynchronous pipeline. It enables training models like ImageNet in minutes on a single GPU with drop-in loader replacement.

## Use cases
- speed up pytorch model training data loading
- train imagenet fast on one gpu
- convert dataset to fast binary format for training
- replace torchvision dataloader with faster alternative
- accelerate image augmentation pipeline on gpu
- train cifar-10 in under a minute
- reduce cloud cost of model training

## When to choose
- training vision models where data loading is the bottleneck
- large image datasets with repeated training epochs
- cost-sensitive GPU training on a single machine

## When to avoid
- non-image or streaming datasets with complex formats
- Windows or macOS-first workflows (Linux is best supported)
- projects needing frequent dataset mutation after conversion

## Facets
- artifact type: library
- maturity: maintenance
- function: machine-learning, data-science, etl, image-processing
- domain: machine-learning, computer-vision, deep-learning
- platform: python, cross-platform
- tags: pytorch, data-loading, dataloader, image-decoding, training-acceleration, gpu-training, data-engineering, linux, gpu

## Member repositories
- libffcv/ffcv (main) score 23

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:07:35.871348+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 2026-08-30T07:30:51.705089+00:00, confidence not recorded.
  - readme: https://github.com/libffcv/ffcv (fetched 2026-08-28T04:07:35.871348+00:00, sha 97a41107f075)
  - homepage: https://ffcv.io (fetched 2026-08-29T09:46:10.069255+00:00, sha 34e3129a8efe)
  - site_page: https://docs.ffcv.io (fetched 2026-08-29T09:46:10.078375+00:00, sha 8f96973f536b)
  - site_page: https://docs.ffcv.io/quickstart.html (fetched 2026-08-29T09:46:10.080283+00:00, sha cd5811a0133a)
  - registry_pypi: https://pypi.org/pypi/ffcv/json (fetched 2026-08-29T09:46:10.081988+00:00, sha ff9344061cef)
- Data as of 2026-08-30T08:39:29.467469+00:00.