1. Getting Started

This chapter walks you through the end-to-end workflow for deploying a deep learning model on a TI evaluation board (EVM) using TI TVM. By the end you will have a compiled model artifact running inference on the target device.

The workflow consists of four steps:

1. Install — Set up TI TVM and the required toolchain on your host machine.

2. Understand your model — Identify input/output names, shapes, and verify expected results before bringing TVM into the picture.

3. Compile — Use the TI TVM compiler to partition the model between TIDL-accelerated and TVM-generated code, and produce a deployable artifact.

4. Run inference — Load the compiled artifact on the EVM and run inference using the TVM runtime.

Follow the Quick Start to get something running end-to-end, then refer to Compiling Models and Running Inference for full reference documentation.

• 1.1. Installation Instructions
  • 1.1.1. Prerequisites
  • 1.1.2. Supported Devices
  • 1.1.3. Processor SDK Versions
  • 1.1.4. Host Installation (Compilation)
  • 1.1.5. Target Installation (Inference)
• 1.2. Preparing Your Model
  • 1.2.1. Verify the model before compilation
• 1.3. Quick Start
  • 1.3.1. Prepare calibration data
  • 1.3.2. Compile the model
  • 1.3.3. Copy artifacts to the EVM
  • 1.3.4. Run inference on the EVM