fastai

fastai is a layered, developer-friendly deep-learning framework designed to democratize AI.
Built on top of PyTorch, it provides:

• High-level Components – concise APIs that hide boilerplate yet deliver cutting-edge results in vision, text, tabular data, collaborative filtering, and time-series.
• Low-level Building Blocks – modular pieces (optimizers, data blocks, callbacks, etc.) that researchers can mix-and-match to craft novel architectures without sacrificing performance.
• Modern Training Best-Practices – mixed-precision, progressive resizing, discriminative learning rates, and a powerful two-way callback system are available out-of-the-box.
• Colab-ready Notebooks & Courses – every documentation page doubles as an executable notebook; paired with the free “Practical Deep Learning for Coders” MOOC, newcomers can be productive in minutes.
• Ecosystem Integrations – seamless hooks for W&B, TensorBoard, Hugging Face Hub, and distributed or mixed-precision training, plus Docker images and Conda/Pip packages.

Created by the fast.ai research group (Jeremy Howard & Rachel Thomas) and released publicly in 2017, fastai continues to evolve (v2, nbdev-based docs, an accompanying 600-page book) while staying true to its mission: make deep learning easier, faster, and more accessible for everyone.