google research

It serves as the canonical source for implementations accompanying Google research papers, releasing the actual code used in experiments so that the broader research community can reproduce results, build on the work, and apply the techniques to their own problems.

The repository contains hundreds of projects across domains: computer vision architectures (ViT, EfficientNet), NLP models (T5, BERT variants, language understanding benchmarks), multimodal research (image-text models, audio-visual learning), reinforcement learning environments and algorithms, medical AI tools (retinal image analysis, dermatology classification), and mathematical and scientific computing tools.

Each project typically includes training code, pretrained model weights where applicable, and documentation sufficient to reproduce the paper's key results.

ML researchers building on Google's published architectures, practitioners implementing Google's techniques for production applications, and academic labs studying specific research problems use this repository as the authoritative source for Google's open-source research contributions.

The breadth of topics means the repository functions as a snapshot of one of the world's most productive AI research organizationsproviding visibility into the problems and approaches Google considers worth investing in, alongside the implementations needed to evaluate and extend the work.