Data Parallel Extension for NumPy*ΒΆ

dpnp is a NumPy-compatible array library for data-parallel computing. It acts as a drop-in replacement for core NumPy* functions and numerical data types and provides implementations of selected SciPy* routines for data-parallel devices.

Overview

Learn about the Data Parallel Extension for NumPy*, its design principles, and what it provides.

Quick Start Guide

Get started with installation, setup, and your first dpnp program.

API Reference

Detailed documentation of all supported NumPy functions and classes in dpnp.

Tensor (dpnp.tensor)

The underlying Array API-compliant implementation based on data-parallel algorithms for accelerators.

Development information

C++ backend API reference and extension documentation for developers.