google-deepmind/dm-haiku

Health v2 · maintenance only

Adoption not part of the score

3275 stars · 298 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

Haiku is a simple neural network library for JAX from Google DeepMind, providing an object-oriented module abstraction (hk.Module) and a function transformation (hk.transform) that bridges OOP-style code with JAX's pure function transformations. It is the JAX counterpart of Sonnet and is now in maintenance mode, with DeepMind recommending Flax for new projects.

Use cases

• build neural networks in jax with object-oriented modules
• convert oop-style model code into pure jax functions
• port sonnet models from tensorflow to jax
• train deep learning models with gpu/tpu support in jax
• manage model parameters in jax research code

When to choose

• you maintain existing Haiku-based codebases
• you prefer Sonnet-style OOP module APIs in JAX
• you need a minimal, stable neural network library for JAX research

When to avoid

• you are starting a new project - DeepMind recommends Flax
• you need new features or an active community
• you want extensive end-to-end examples and documentation

Facets

library · maturity maintenance

machine-learning deep-learning machine-learning deep-learning python jax neural-networks sonnet google-deepmind object-oriented-api gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="google-deepmind/dm-haiku")