patrick-kidger/jaxtyping

Type annotations and runtime checking for shape and dtype of JAX/NumPy/PyTorch/etc. arrays. https://docs.kidger.site/jaxtyping/

What it solves

It addresses the difficulty of tracking and verifying the shapes and data types of arrays and tensors across different machine learning frameworks. In complex models, mismatching dimensions often lead to runtime errors that are hard to debug.

How it works

The library provides specialized type annotations that allow developers to specify the expected shape and data type of a tensor (e.g., Float[Tensor, "dim1 dim2"]). These annotations are compatible with runtime type-checking packages like typeguard and beartype to enforce these constraints during execution.

Who it’s for

Researchers and engineers working with JAX, PyTorch, NumPy, TensorFlow, or MLX who want to make their code more readable, maintainable, and less prone to shape-related bugs.

Highlights

  • Support for multiple frameworks including JAX, PyTorch, NumPy, TensorFlow, and MLX.
  • Runtime type-checking capabilities when paired with compatible libraries.
  • Clear, descriptive annotations for array dimensions and axes.

Related

  • Project
    intel/neural-compressor

    An open-source Python library that provides model compression and quantization techniques for PyTorch, TensorFlow, and JAX models to improve inference efficiency.

  • Project
    lutzroeder/netron

    A visualizer for neural network and deep learning models that supports a wide variety of model formats including ONNX, TensorFlow, and PyTorch.

  • Project
    unifyai/ivy

    Ivy is a transpiler that allows users to convert machine learning models, tools, and libraries between frameworks like PyTorch, TensorFlow, JAX, and NumPy.

  • Project
    patrick-kidger/equinox

    A JAX library that provides PyTorch-like syntax for building neural networks and models by treating them as PyTrees, ensuring full compatibility with the JAX ecosystem.

  • Project
    microsoft/onnxruntime

    A cross-platform machine learning accelerator that optimizes inference and training for models from various frameworks like PyTorch and TensorFlow.