Welcome to Hidet’s Documentation

Hidet is an open-source DNN inference framework, it features

• Ease of Use: Support end to end inference for PyTorch and ONNX models.

• High Performance: Graph-level optimizations and operator-level kernel tuning.

• Extensibility: Easy to add new operators, and fusion patterns.

• Python Oriented: All core components are written in Python.

Getting Started

• Installation
• Quick Start

Tutorials

• Optimize PyTorch Model
• Optimize ONNX Model

Hidet Script

• Introduction
• Examples
• Reference

Developer Guide

• Add New Operator
• Add Operator Resolve Rule
• Add Sub-Graph Rewrite Rule
• Contributing

Notes

• Operator Cache
• Visualize Flow Graph

Reference

• Python API
• Index