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.

The models trained in TLT are deployed to NVIDIA’s Inference SDK’s such as DeepStream, Riva etc via TensorRT. While the conversational AI models trained using TLT can be consumed via TensorRT only via Riva, the computer vision models trained by TLT can be consumed by TensorRT, via the tlt-converter tool. The TLT 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 TLT models support direct integration of the .etlt files to DeepStream 5.1, DeepStream can also consume the optimized engine generated by the tlt-converter.

The TensorRT engines generated by this tlt-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 tlt-converter and generate the engine there.

The TLT models have been verified to integrate with TensorRT version 7.0, 7.1 and 7.2.

Eventhough 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 TLT applications that require TensorRT OSS are:

• FasterRCNN

• SSD

• DSSD

• YOLOv3

• YOLOv4

• RetinaNet

• MaskRCNN

The TLT Converter is distributed as a separate binary for x86 and Jetson platforms. The following table lists the links where you can download the tlt-converter.

Installing on an x86 platform

For an x86 platform with discrete GPUs, the default TLT package includes the tlt-converter built for TensorRT 7.2 with CUDA 11.1 and CUDNN 8.0. However, for any other version of CUDA and TensorRT, please refer to the overview section for download. Once the tlt-converter is downloaded, follow the instructions below to generate a TensorRT engine.

1. Unzip the zip file on the target machine.

2. Install the OpenSSL package using the command:

   Copy

   Copied!

               
   
               
   sudo apt-get install libssl-dev
           

3. 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”
        

4. Run the tlt-converter using the sample command below and generate the engine.

5. Instructions to build TensorRT OSS on Jetson can be found in the TensorRT OSS on x86 section above or in this GitHub repo.

Installing on an jetson platform

For the Jetson platform, the tlt-converter is available to download in the NVIDIA developer zone. You may choose the version you wish to download as listed in the overview section. Once the tlt-converter is downloaded, please follow the instructions below to generate a TensorRT engine.

1. Unzip the zip file on the target machine.

2. Install the OpenSSL package using the command:

   Copy

   Copied!

               
   
               
   sudo apt-get install libssl-dev
           

3. Export the following environment variables:

            

            
$ export TRT_LIB_PATH=”/usr/lib/aarch64-linux-gnu”
$ export TRT_INC_PATH=”/usr/include/aarch64-linux-gnu”
        

4. For Jetson devices, TensorRT 7.1 comes pre-installed with Jetpack. If you are using older JetPack, upgrade to JetPack 4.4 or JetPack 4.5.

5. Instructions to build TensorRT OSS on Jetson can be found in the TensorRT OSS on Jetson (ARM64) section above or in this GitHub repo.

6. Run the tlt-converter using the sample command below and generate the engine.

Using the tlt-converter

            

            
tlt-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
        

Required Arguments

• input_file: Path to the .etlt model exported using tlt <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 tlt <model> export.

• -o: Comma-separated list of output blob names that should match the output configuration used for tlt <model> export.

Optional Arguments

• -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 TLT 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 TLT 3.0. This parameter is not required for models that are already existed in TLT 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.

INT8 Mode Arguments

• -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 TLT 3.0.)

The usage for each TLT Computer Vision is explained in the respective models chapter.