Keras Developer Guides
by Keras
The friendliest high-level deep-learning API — now multi-backend across JAX, TensorFlow and PyTorch.
Overview
keras.io/guides is the official, task-organised half of the Keras documentation — not an API reference but a set of about thirty end-to-end walkthroughs, each available as a runnable Colab notebook. The sequence starts with the Sequential model and the Functional API, then moves to making new layers and models via subclassing, training and evaluation with the built-in methods, serialization and saving, and writing your own callbacks. From there it splits by backend: customizing fit() and writing a custom training loop each have separate JAX, TensorFlow and PyTorch guides, as does distributed training, plus a backend-agnostic 'Distributed training with Keras 3' guide. Transfer learning and fine-tuning gets its own guide, which is the fastest path from a pretrained checkpoint to a working model on your own data. The newer material reflects where Keras 3 has gone since the multi-backend rewrite — a full quantization track (8-bit and 4-bit integer, GPTQ, AWQ, writing quantization-compatible layers, customizing quantization), Orbax checkpointing, the NNX backend, exporting to LiteRT for on-device inference, and defining custom TPU/GPU kernels. A 'Migrating Keras 2 code to Keras 3' guide exists for legacy codebases. Keras 3 (latest release 3.15.1, Python 3.11+) runs on TensorFlow 2.16.1+, JAX 0.4.20+, PyTorch 2.1.0+ and OpenVINO for inference only; the repository is Apache-2.0 with roughly 64k GitHub stars.
At a Glance
- Topic
- Frameworks
- Level
- All Levels
- Format
- Tutorial
- Cost
- Free
- Duration
- Self-paced — roughly 30 guides, each a runnable Colab notebook
- Provider
- Keras
- Hands-on
- Yes — code/exercises
- Certificate
- None
What You’ll Learn
- ✓Build models with the Sequential API and the more flexible Functional API
- ✓Write custom layers and models by subclassing Layer and Model
- ✓Customize fit() or write a full training loop in JAX, TensorFlow or PyTorch
- ✓Serialize, save and reload models, including custom objects and formats
- ✓Apply transfer learning and fine-tuning to a pretrained backbone
- ✓Scale training across devices using the per-backend distributed training guides
- ✓Quantize models to 8-bit or 4-bit, including GPTQ and AWQ
- ✓Migrate an existing Keras 2 codebase onto Keras 3
Highlights
- •Every guide is a runnable Colab notebook rather than prose
- •Backend-specific topics ship parallel JAX, TensorFlow and PyTorch versions
- •Unusually complete quantization track for framework docs: 8-bit, 4-bit, GPTQ, AWQ
- •Maintained alongside the Apache-2.0 repo (~64k stars, current release 3.15.1)
- •Includes an explicit Keras 2 to Keras 3 migration guide for legacy code
Who It’s For
Best For
- ✓Engineers who want a high-level API without committing to one backend
- ✓PyTorch or JAX users who prefer Keras' training loop over hand-rolling one
- ✓Teams maintaining Keras 2 code that has to move to Keras 3
- ✓Practitioners who need on-device (LiteRT) or quantized deployment paths
Prerequisites
- •Working Python including classes — the subclassing guides assume it
- •Basic neural network concepts: layers, loss, optimizer, backpropagation
- •One backend installed (TensorFlow, JAX or PyTorch); Keras does not ship one
FAQ
What is Keras Developer Guides?
The official Keras 3 guide collection: roughly thirty runnable walkthroughs covering the Sequential and Functional APIs, layer and model subclassing, custom training loops, serialization, callbacks, transfer learning, distributed training and quantization. Every backend-specific topic ships three parallel versions — JAX, TensorFlow and PyTorch — so the same high-level code runs on whichever framework your team already uses.
Is Keras Developer Guides free?
Keras Developer Guides is free to access.
What level is Keras Developer Guides for?
Keras Developer Guides is aimed at a all levels audience. Recommended background: Working Python including classes — the subclassing guides assume it, Basic neural network concepts: layers, loss, optimizer, backpropagation, One backend installed (TensorFlow, JAX or PyTorch); Keras does not ship one.
How long does Keras Developer Guides take?
Expect roughly Self-paced — roughly 30 guides, each a runnable Colab notebook. Most learners work through it at their own pace.
What will I learn from Keras Developer Guides?
You'll learn: Build models with the Sequential API and the more flexible Functional API; Write custom layers and models by subclassing Layer and Model; Customize fit() or write a full training loop in JAX, TensorFlow or PyTorch; Serialize, save and reload models, including custom objects and formats; Apply transfer learning and fine-tuning to a pretrained backbone; Scale training across devices using the per-backend distributed training guides; Quantize models to 8-bit or 4-bit, including GPTQ and AWQ; Migrate an existing Keras 2 codebase onto Keras 3.
Topics
Sources
This page was written from 3 sources, 2 on domains other than keras.io.