This resource has been deprecated in favor of the nvidia-tao-deploy package in TAO. For further instruction on how to get this package, refer to the TAO documentation.
The models trained in TAO Toolkit are deployed to NVIDIA's Inference SDK's such as DeepStream, Riva etc
via TensorRT. NVIDIA TensorRT is an SDK for high-performance deep learning inference. TensorRT provides APIs and parsers
to import trained models from all major deep learning frameworks. It then generates optimized runtime engines
deployable in the datacenter as well as in automotive and embedded environments. To understand TensorRT and it's
capabilities better, refer to the official TensorRT documentation. While the conversational AI models trained using TAO Toolkit can be consumed via TensorRT only via Riva,
the computer vision models trained by TAO Toolkit can be consumed by TensorRT, via the tao-converter tool. The
TAO Converter parses the exported .etlt model file, and generates an optimized TensorRT engine. These engines
can be generated to support inference at low precision, such as FP16 or INT8.
While most of the TAO models support direct integration of the .etlt files to DeepStream 6.1, DeepStream can also
consume the optimized engine generated by the tao-converter.
The TensorRT engines generated by this tao-converter are specific to the GPU that it was generated on. So,
based on the platform that the model is being deployed to, you will need to download the specific version of
the tao-converter and generate the engine there.
The TAO models from TAO 3.0-22.05 have been verified to integrate with TensorRT version 7.0, 7.1, 7.2, 8.0, 8.2 and 8.4.
Even though TensorRT contains optimized implementations for several common operations used in Deep Neural Networks(DNNs),
with Deep Learning being such a quickly evolving discipline, TensorRT provides users a method to bring in new operations
via to the model graph via custom TensorRT Plugins. Several samples of these custom plug-ins are hosted on
GitHub under the repository called TensorRT OSS.
Instructions to build and install TensorRT OSS can be found in this repository.
The TAO applications that require TensorRT OSS are:
The TAO Converter is distributed as a separate binary for x86 and Jetson platforms. The following table lists the links where you can download the tao-converter.
| TensorRT | Platform |
|---|---|
| 7.2 | x86 |
| 7.1 | x86 |
| 8.0 | x86 |
| 8.2 | x86 |
| 8.4 | x86 |
| 8.0 | aarch64 |
| 8.2 | aarch64 |
| 8.4 | aarch64 |
Before installing the tao-converter, install the TensorRT OSS library by following the instructions
here.
For an x86 platform with discrete GPUs, the default TAO package includes the tao-converter
built for TensorRT 8.2.5.1 with CUDA 11.4 and CUDNN 8.2. However, for any other version of TensorRT,
you may download using the command below:
ngc registry resource download-version nvidia/tao/tao-converter:<latest_version> --dest /path/to/download/directory
Once the tao-converter is downloaded, follow the instructions below to generate a TensorRT engine.
Unzip the zip file on the target machine.
Install the OpenSSL package using the command:
sudo apt-get install libssl-dev
Export the following environment variables:
export TRT_LIB_PATH=”/usr/lib/x86_64-linux-gnu”
export TRT_INC_PATH=”/usr/include/x86_64-linux-gnu”
Run the tao-converter using the sample command below and generate the engine.
Instructions to build TensorRT OSS on Jetson can be found in the TensorRT OSS on x86 section above or in this GitHub repo`.
Make sure to follow the output node names as mentioned in
Exporting the Modelsection of the respective model in the TAO Toolkit documentation
Once the tao-converter is downloaded, follow the instructions below to generate a TensorRT engine.
Unzip the zip file on the target machine.
Install the OpenSSL package using the command:
sudo apt-get install libssl-dev
Export the following environment variables:
export TRT_LIB_PATH=”/usr/lib/aarch64-linux-gnu”
export TRT_INC_PATH=”/usr/include/aarch64-linux-gnu”
Run the tao-converter using the sample command below and generate the engine.
Instructions to build TensorRT OSS on Jetson can be found in the TensorRT OSS on aarch64 section above or in this GitHub repo`.
Due to changes in the TensorRT API between versions
8.x.xand7.2.x, the deployable models generated using theexporttask in TAO Toolkit 3.0-21.11+ can only be deployed in DeepStream version 6.1. Inorder to deploy the models compatible with DeepStream 5.1 from the table above with DeepStream 5.1, you will need to run the correspondingtao <model> exporttask using the TAO Toolkit 3.0-21.08 package to re-generate a deployable model and calibration cache file that is compatible with TensorRT 7.2.Similarly, if you have a model trained with TAO Toolkit 3.0-21.08 package and want to deploy to DeepStream 6.1, please regenerate the deployable
model.etltand int8 calibration file using the correspondingtao <model> exporttask in TAO Toolkit 3.0-21.11+TAO Toolkit 3.0-22.05 was built with TensorRT 8.2.5.1.
TAO Toolkit -> DeepStream version interoperability
To install the 3.0-21.08 or 3.0-21.11 package, please instantiate a new virtual environment by following the instructions in the :ref:
Quick Start Guide <installing_tao_toolkit>and run the following commandspip3 install nvidia-pyindex pip3 install nvidia-tao==0.1.19 # for 3.0-21.08 pip3 install nvidia-tao>=0.1.20 # for 3.0-21.11 pip3 install nvidia-tao>=0.1.23 # for 3.0-22.05
tao-converter [-h] -k <encryption_key>
-d <input_dimensions>
-o <comma separated output nodes>
[-c <path to calibration cache file>]
[-e <path to output engine>]
[-b <calibration batch size>]
[-m <maximum batch size of the TRT engine>]
[-t <engine datatype>]
[-w <maximum workspace size of the TRT Engine>]
[-i <input dimension ordering>]
[-p <optimization_profiles>]
[-s]
[-u <DLA_core>]
input_file
input_file: Path to the .etlt model exported using tao <model> export.-k: The key used to encode the .tlt model when doing the training.-d: Comma-separated list of input dimensions that should match the dimensions used for
tao <model> export.-o: Comma-separated list of output blob names that should match the output configuration
used for tao <model> export.-e: Path to save the engine to. (default: ./saved.engine)-t: Desired engine data type, generates calibration cache if in INT8 mode. The default
value is fp32. The options are {fp32, fp16, int8}.-w: Maximum workspace size for the TensorRT engine. The default value is 1073741824(1<<30).-i: Input dimension ordering, all other TAO commands use NCHW. The default value is
nchw. The options are {nchw, nhwc, nc}.-p: Optimization profiles for .etlt models with dynamic shape. Comma separated
list of optimization profile shapes in the format <input_name>,<min_shape>,<opt_shape>,<max_shape>,
where each shape has the format: <n>x<c>x<h>x<w>. Can be specified multiple times if there are
multiple input tensors for the model. This is only useful for new models introduced in TAO Toolkit 3.21.08.
This parameter is not required for models that are already existed in TAO Toolkit 2.0.-s: TensorRT strict type constraints. A Boolean to apply TensorRT strict type constraints
when building the TensorRT engine.-u: Use DLA core. Specifying DLA core index when building the TensorRT engine on Jetson devices.-c: Path to calibration cache file, only used in INT8 mode. The default value is
./cal.bin.-b: Batch size used during the export step for INT8 calibration cache generation.
(default: 8).-m: Maximum batch size for TensorRT engine.(default: 16). If meet with out-of-memory
issue, decrease the batch size accordingly. This parameter is not required for .etlt
models generated with dynamic shape. (This is only possible for new models introduced in TAO Toolkit 3.21.08.)The usage for each TAO Computer Vision is explained in the respective models chapter.