Parses ONNX models for execution with TensorRT.
See also the TensorRT documentation.
For the list of recent changes, see the changelog.
For a list of commonly seen issues and questions, see the FAQ.
For business inquiries, please contact researchinquiries@nvidia.com
For press and other inquiries, please contact Hector Marinez at hmarinez@nvidia.com
Development on the this branch is for the latest version of TensorRT 10.7 with full-dimensions and dynamic shape support.
For previous versions of TensorRT, refer to their respective branches.
Current supported ONNX operators are found in the operator support matrix.
- Protobuf >= 3.0.x
- TensorRT 10.7
- [TensorRT 10.7 open source libaries] (https://github.com/NVIDIA/TensorRT/)
For building within docker, we recommend using and setting up the docker containers as instructed in the main TensorRT repository to build the onnx-tensorrt library.
Once you have cloned the repository, you can build the parser libraries and executables by running:
cd onnx-tensorrt
mkdir build && cd build
cmake .. -DTENSORRT_ROOT=<path_to_trt> && make -j
# Ensure that you update your LD_LIBRARY_PATH to pick up the location of the newly built library:
export LD_LIBRARY_PATH=$PWD:$LD_LIBRARY_PATH
Note that this project has a dependency on CUDA. By default the build will look in /usr/local/cuda for the CUDA toolkit installation. If your CUDA path is different, overwrite the default path by providing -DCUDA_TOOLKIT_ROOT_DIR=<path_to_cuda_install> in the CMake command.
To build with protobuf-lite support, add -DUSE_ONNX_LITE_PROTO=1 to the end of the cmake command.
There are two implementations of InstanceNormalization that may perform differently depending on various parameters. By default, the parser will use the native TensorRT implementation of InstanceNorm. Users that want to benchmark using the plugin implementation of InstanceNorm can unset the parser flag kNATIVE_INSTANCENORM prior to parsing the model. Note that the plugin implementation cannot be used for building version compatible or hardware compatible engines, and attempting to do so will result in an error.
C++ Example:
// Unset the kNATIVE_INSTANCENORM flag to use the plugin implementation.
parser->unsetFlag(nvonnxparser::OnnxParserFlag::kNATIVE_INSTANCENORM);
Python Example:
// Unset the NATIVE_INSTANCENORM flag to use the plugin implementation.
parser.clear_flag(trt.OnnxParserFlag.NATIVE_INSTANCENORM)
There are currently two officially supported tools for users to quickly check if an ONNX model can parse and build into a TensorRT engine from an ONNX file.
For C++ users, there is the trtexec binary that is typically found in the <tensorrt_root_dir>/bin directory. The basic command of running an ONNX model is:
trtexec --onnx=model.onnx
Refer to the link or run trtexec -h for more information on CLI options.
For Python users, there is the polygraphy tool. The basic command for running an onnx model is:
polygraphy run model.onnx --trt
Refer to the link or run polygraphy run -h for more information on CLI options.
Python bindings for the ONNX-TensorRT parser are packaged in the shipped .whl files.
TensorRT 10.7 supports ONNX release 1.17.0. Install it with:
python3 -m pip install onnx==1.17.0
The ONNX-TensorRT backend can be installed by running:
python3 setup.py install
The TensorRT backend for ONNX can be used in Python as follows:
import onnx
import onnx_tensorrt.backend as backend
import numpy as np
model = onnx.load("/path/to/model.onnx")
engine = backend.prepare(model, device='CUDA:1')
input_data = np.random.random(size=(32, 3, 224, 224)).astype(np.float32)
output_data = engine.run(input_data)[0]
print(output_data)
print(output_data.shape)The model parser library, libnvonnxparser.so, has its C++ API declared in this header:
NvOnnxParser.h
After installation (or inside the Docker container), ONNX backend tests can be run as follows:
Real model tests only:
python onnx_backend_test.py OnnxBackendRealModelTest
All tests:
python onnx_backend_test.py
You can use -v flag to make output more verbose.
Pre-trained models in ONNX format can be found at the ONNX Model Zoo