Tensor by Tensor

MODULE 03 · LESSON 01

What is ONNX Runtime?

A runtime loads a saved graph and runs its fixed calculations for inference.

PLAIN-LANGUAGE INTRODUCTION

What is this?

A runtime loads a saved graph and runs its fixed calculations for inference.

One simple example

A graph applies y=2x+1 to input [[1,2,3,4]]. It returns [[3,5,7,9]].

What goes in?

A model file and named arrays with expected shapes and data types.

What comes out?

Named prediction arrays from the graph.

Why does it matter?

An application can run a trained model without its training program.

What is it not?

A runtime does not train the saved model.

WORK THROUGH THE IDEA

See the idea in more detail

  1. A graph stores operations and learned constants. A session prepares that graph for execution.
  2. The example input has shape (1,4). Its four values are [1,2,3,4].
  3. The graph applies 2x+1 element by element. The result is [3,5,7,9].
  4. Input name, shape, and data type must match the model contract.
  5. Common mistake: a valid array with the wrong data type can still be rejected.
Open the detailed notes ↗