Implementation:Microsoft Onnxruntime Numpy Input Construction

Metadata

Field	Value
Implementation Name	Numpy_Input_Construction
Repository	Microsoft_Onnxruntime
Source Repository	https://github.com/microsoft/onnxruntime
Type	External Tool Doc
External Tool	numpy
Language	Python
Domain	ML_Inference, Model_Optimization
Last Updated	2026-02-10
Workflow	Python_Inference_Pipeline
Pair	4 of 6

Overview

External tool documentation for constructing properly typed and shaped numpy arrays as input data for ONNX Runtime inference. This implementation relies on the numpy library for array creation and type casting.

API Signature

numpy.random.random(shape).astype(numpy.float32)

Or for user-supplied data:

numpy.array(data).astype(numpy.float32)

Import

import numpy

Code Reference

Reference	Location
Usage example	docs/python/examples/plot_load_and_predict.py:L52-53

I/O Contract

Inputs

Parameter	Type	Required	Description
Model input metadata	Names, shapes, types from `session.get_inputs()`	Yes	Used to determine the correct array shape, dtype, and dictionary key.
Raw data	Any array-like	Yes	The actual data to be fed into the model (random, test data, or production data).

Outputs

Output	Type	Description
input_feed	`dict[str, numpy.ndarray]`	Dictionary mapping input tensor names to correctly-typed numpy arrays.

Usage Example

import numpy

# Generate random test input matching model shape
x = numpy.random.random((3, 4, 5)).astype(numpy.float32)
input_feed = {input_name: x}

From the source example at docs/python/examples/plot_load_and_predict.py:L52-53:

x = numpy.random.random((3, 4, 5))
x = x.astype(numpy.float32)

Common Type Conversions

# Float32 input (most common)
x = numpy.array(data).astype(numpy.float32)

# Int64 input (common for token IDs)
x = numpy.array(token_ids).astype(numpy.int64)

# Constructing multi-input feed
input_feed = {
    sess.get_inputs()[0].name: input_array_1,
    sess.get_inputs()[1].name: input_array_2,
}

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment