google-research/big_vision

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

3528 stars · 226 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

Google Research's official Jax/Flax codebase for training large-scale vision models such as Vision Transformer, SigLIP, MLP-Mixer, and LiT on Cloud TPUs or GPUs. It provides reproducible training pipelines that scale from a single TPU core to distributed setups with up to 2048 TPU cores.

Use cases

• train a vision transformer from scratch on ImageNet
• reproduce SigLIP or LiT training results
• run large-scale vision experiments on Cloud TPUs
• fine-tune pretrained ViT models for image classification
• implement and benchmark new vision model architectures
• scale vision model training across many TPU cores

When to choose

• you need a battle-tested Jax codebase for large-scale vision model training
• you want to reproduce or build on Google's vision transformer research
• you have TPU or multi-GPU resources and need seamless scaling
• you want official implementations of ViT, SigLIP, MLP-Mixer, or LiT

When to avoid

• you need a production inference or deployment library
• you expect community-driven feature development or quick external contributions
• you work outside the Jax/Flax ecosystem, e.g. prefer PyTorch
• you need a simple transfer-only fine-tuning tool (the vision_transformer repo is a better fit)

Facets

library · maturity active

machine-learning deep-learning llm-training deep-learning computer-vision machine-learning python cloud jax flax vision-transformer tpu research-codebase siglip image-classification research gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="google-research/big_vision")