nvidia--tensorrt/samples/sampleIOFormats

Specifying I/O Formats

Table Of Contents

• Description
• How does this sample work?
• Running the sample
  • Sample --help options
• Preparing sample data
• Additional resources
• License
• Changelog
• Known issues

Description

This sample, sampleIOFormats, uses a Onnx model that was trained on the MNIST dataset and performs engine building and inference using TensorRT. The correctness of outputs is then compared to the golden reference. Specifically, it shows how to use APIs to explicitly specify input formats to TensorFormat::kLINEAR, TensorFormat::kHWC and TensorFormat::kCHW32 for Float32.

How does this sample work?

ITensor::setAllowedFormats is invoked to specify which format is expected to be supported.

```
bool SampleIOFormats::build(int dataWidth)
{
	...

	network->getInput(0)->setAllowedFormats(static_cast<TensorFormats>(1 << static_cast<int>(mTensorFormat)));
	...
}
```

Prerequisites

1. Preparing sample data See Preparing sample data in the main samples README.

Running the sample

1. Compile the sample by following build instructions in TensorRT README.

2. Run inference on the digit looping from 0 to 9:

   ./sample_io_formats --datadir=<path/to/data> --useDLACore=N
   

   For example:

   ./sample_io_formats --datadir $TRT_DATADIR/mnist
   

3. Verify that all 10 digits match correctly. If the sample runs successfully, you should see output similar to the following:

   &&&& RUNNING TensorRT.sample_io_formats # ./sample_io_formats
   [I] Build TRT engine with different IO data type and formats. Ensure that built engine abide by them
   [I] Testing datatype FP32 with format kLINEAR
   [I] Building and running a GPU inference engine with specified I/O formats.
   ... (omitted message)
   [I] Testing datatype FP32 with format kHWC
   [I] Building and running a GPU inference engine with specified I/O formats.
   ... (omitted message)
   [I] Testing datatype FP32 with format kCHW32
   [I] Building and running a GPU inference engine with specified I/O formats.
   ... (omitted message)
   &&&& PASSED TensorRT.sample_io_formats
   

   This output shows that the sample ran successfully; PASSED.

Sample --help options

To see the full list of available options and their descriptions, use the -h or --help command line option.

Additional resources

The following resources provide a deeper understanding about this sample:

Models

• MNIST

Documentation

• Introduction To NVIDIA’s TensorRT Samples
• Working With TensorRT Using The C++ API
• NVIDIA’s TensorRT Documentation Library

License

For terms and conditions for use, reproduction, and distribution, see the TensorRT Software License Agreement documentation.

Changelog

October 2025

• Migrate to strongly typed APIs.

August 2022

• Migrated code from parsing a caffe model to an onnx model.

Oct 2021

• Change names and topic from "reformat-free" to "I/O formats", because BuilderFlag::kSTRICT_TYPES is deprecated. "Reformat-free I/O" (see BuilderFlag::kDIRECT_IO) is generally counterproductive and fragile, since it constrains the optimizer from choosing the fastest implementation, and depends upon what kernels are available on a particular target.

June 2019

• This is the first release of the README.md file and sample.

Known issues

There are no known issues in this sample.