A deep learning library built on PyTorch that provides high-level components for rapid results and low-level components for research flexibility.
fastai is a deep learning library built on PyTorch that provides high-level components for practitioners to quickly achieve state-of-the-art results in domains like computer vision, NLP, and tabular data. It also offers low-level, composable building blocks for researchers to develop novel approaches without sacrificing ease of use or performance.
Deep learning practitioners seeking rapid productivity and researchers needing flexible, hackable components for experimental work, both leveraging PyTorch.
Developers choose fastai for its layered API that balances high-level simplicity with low-level configurability, enabling quick results with minimal code while allowing deep customization through decoupled abstractions and modern best practices.
The fastai deep learning library
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Enables building state-of-the-art models like image classifiers and text analyzers with just a few lines of code, as shown in the Quick Start.
Provides a hierarchy of APIs from high-level productivity to low-level configurability, allowing deep customization without learning the lowest level, based on the layered design.
Features a novel two-way callback system that accesses and modifies data, model, or optimizer behavior at any point during training, offering fine-grained control.
Includes an extensible computer vision library written in pure Python but optimized for GPU performance, as highlighted in the key features.
Automatically resets num_workers to 0 on Windows with Jupyter, slowing computer vision tasks, as admitted in the README's Windows support section.
Tightly integrated with PyTorch, making migration to or use with other frameworks like TensorFlow difficult without significant code changes.
Requires editable installs and dependencies like fastcore for contributing, which is more involved than simpler libraries, as noted in the installing instructions.