Introduction to PyBuda

PyBuda ™ is a compute framework used to develop, run, and analyze ML workloads on Tenstorrent hardware.

For a quick start, follow First 5 Things for TT-Buda.

Table of Contents

• Introduction to PyBuda

• User Guide

  • Framework Support

  • PyBuda Introduction

  • Saving and Loading Models

  • Pybuda Automatic Mixed Precision

  • Multiple Devices

  • Pybuda Multi-Model Support (Embedded Applications Only)

  • TT-SMI

  • Examples of PyBuda use cases

• API Reference

  • Python Runtime API

  • C++ Runtime API

  • Configuration and Placement

  • Operations

  • Module Types

  • Device Types

  • Miscellaneous

• Terminology

  • Grayskull

  • Wormhole

  • Tensix

  • NOC

  • L1

  • FPU

  • SFPU

  • Unpacker

  • Packer

  • Op

  • Epoch

  • Buffer

  • Pipe

  • Tile

  • Streaming op

  • Blocking op

  • Block

  • TM op

  • Unicast

  • Multicast

  • Gather

  • HLK

  • LLK

  • HLKC

• Advanced User Guide

  • Software Stack Overview

  • Compiler Configuration

  • Tools & Debug

  • Comparison To GPU Programming Model

• Hardware Overview

• Data Formats and Math Fidelity

  • Data Formats

  • Block Floating Point

  • Math Fidelity

• Buda Workload Overview

  • Anatomy of a Buda op

• Developer Reference

  • Low Level Kernels

Index

• Index