# Keras

> Keras is an open-source, high-level deep learning API that enables building, training, and deploying neural networks across JAX, TensorFlow, and PyTorch backends.

Keras is a human-centered deep learning API that emphasizes code clarity, fast debugging, and deployability across multiple backends. It provides high-level abstractions for layers and models (Sequential, Functional, and subclassing) while remaining extensible for advanced research. Keras supports multi-backend execution (JAX, TensorFlow, PyTorch), distributed and mixed-precision training, and includes KerasHub with pretrained model implementations and checkpoints.

- **Multi-backend support** — Use JAX, TensorFlow, or PyTorch backends to train and run models without changing Keras code; to get started, choose a backend and follow the backend configuration guides.  
- **Model APIs (Sequential, Functional, Subclassing)** — Build simple to complex architectures using intuitive APIs; begin with Sequential for linear stacks and the Functional API for arbitrary graphs.  
- **Training & callbacks** — Train with model.fit, evaluate, and use callback utilities (Checkpointing, EarlyStopping, TensorBoard) to manage experiments.  
- **KerasHub & pretrained models** — Access implementations and checkpoints for language, vision, and generative models for fine-tuning or inference.  
- **Data loading and utilities** — Built-in data loaders, preprocessing layers, and utilities streamline dataset handling and experiment management.  
- **Production & deployment features** — Model saving, serialization, mixed-precision, and distribution primitives support research-to-production workflows.

To get started, install Keras from source/package, follow the API guides, and run the provided examples and quickstarts to train or load pretrained models.

## Features
- Functional API and Sequential API
- Model.fit training and evaluation APIs
- Callbacks (ModelCheckpoint, EarlyStopping, TensorBoard)
- KerasHub pretrained models and checkpoints
- Multi-backend support (JAX, TensorFlow, PyTorch)
- Built-in datasets, preprocessing, and utilities
- Mixed-precision and multi-device distribution support
- Model saving and serialization utilities

## Integrations
JAX, TensorFlow, PyTorch, Kaggle, Hugging Face, Google Colab, TPU

## Platforms
DEVELOPER_SDK

## Pricing
Open Source

## Version
3.0

## Links
- Website: https://keras.io/
- Documentation: https://keras.io/api
- Repository: https://github.com/keras-team/keras
- EveryDev.ai: https://www.everydev.ai/tools/keras