apple/ml-fastvit

Health v2 · maintenance only

Flags: no_releases no_license

Adoption not part of the score

2039 stars · 130 forks observed · 2026-10-07

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Official PyTorch implementation of FastViT, a fast hybrid vision transformer architecture using structural reparameterization, published at ICCV 2023. It provides pretrained ImageNet-1K checkpoints and CoreML models optimized for low latency on mobile devices like iPhone.

Use cases

• run fast image classification on mobile devices
• fine-tune a pretrained vision transformer backbone
• use a fast backbone for detection or segmentation tasks
• export vision models to CoreML for iOS inference
• benchmark image classification latency on iPhone
• train a hybrid CNN-transformer model from scratch

When to choose

• you need a low-latency image classification model for mobile deployment
• you want a pretrained backbone for downstream vision tasks like detection or segmentation
• you target Apple devices and want CoreML-ready models
• you're researching structural reparameterization in vision transformers

When to avoid

• you need state-of-the-art accuracy regardless of latency
• you don't use PyTorch or Apple's ecosystem
• you need a general-purpose vision library rather than a specific model implementation
• you require frequent updates or long-term support

Facets

library · maturity maintenance

machine-learning deep-learning image-processing computer-vision deep-learning computer-vision image-processing machine-learning python cross-platform vision-transformer structural-reparameterization image-classification pytorch coreml mobile-inference iccv-2023 model-zoo pretrained-models gpu macos

Member repositories

For agents

markdown · JSON · MCP: product_card(name="apple-aiml-research/ml-fastvit")