Table of Contents
  • Docs >
  • Torch-TensorRT Getting Started - ResNet 50
Shortcuts 
[1]:

# Copyright 2019 NVIDIA Corporation. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================

b3fd7c0e57b74b66bc0d32c2be7cdcb4

Torch-TensorRT Getting Started - ResNet 50

Overview

In the practice of developing machine learning models, there are few tools as approachable as PyTorch for developing and experimenting in designing machine learning models. The power of PyTorch comes from its deep integration into Python, its flexibility and its approach to automatic differentiation and execution (eager execution). However, when moving from research into production, the requirements change and we may no longer want that deep Python integration and we want optimization to get the best performance we can on our deployment platform. In PyTorch 1.0, TorchScript was introduced as a method to separate your PyTorch model from Python, make it portable and optimizable. TorchScript uses PyTorch’s JIT compiler to transform your normal PyTorch code which gets interpreted by the Python interpreter to an intermediate representation (IR) which can have optimizations run on it and at runtime can get interpreted by the PyTorch JIT interpreter. For PyTorch this has opened up a whole new world of possibilities, including deployment in other languages like C++. It also introduces a structured graph based format that we can use to do down to the kernel level optimization of models for inference.

When deploying on NVIDIA GPUs TensorRT, NVIDIA’s Deep Learning Optimization SDK and Runtime is able to take models from any major framework and specifically tune them to perform better on specific target hardware in the NVIDIA family be it an A100, TITAN V, Jetson Xavier or NVIDIA’s Deep Learning Accelerator. TensorRT performs a couple sets of optimizations to achieve this. TensorRT fuses layers and tensors in the model graph, it then uses a large kernel library to select implementations that perform best on the target GPU. TensorRT also has strong support for reduced operating precision execution which allows users to leverage the Tensor Cores on Volta and newer GPUs as well as reducing memory and computation footprints on device.

Torch-TensorRT is a compiler that uses TensorRT to optimize TorchScript code, compiling standard TorchScript modules into ones that internally run with TensorRT optimizations. This enables you to continue to remain in the PyTorch ecosystem, using all the great features PyTorch has such as module composability, its flexible tensor implementation, data loaders and more. Torch-TensorRT is available to use with both PyTorch and LibTorch.

Learning objectives

This notebook demonstrates the steps for compiling a TorchScript module with Torch-TensorRT on a pretrained ResNet-50 network, and running it to test the speedup obtained.

Content

  1. Requirements

  2. ResNet-50 Overview

  3. Running the model without optimizations

  4. Accelerating with Torch-TensorRT

  5. Conclusion

[2]:

!nvidia-smi
!pip install ipywidgets --trusted-host pypi.org --trusted-host pypi.python.org --trusted-host=files.pythonhosted.org

Tue Feb  8 19:16:53 2022
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 510.39.01    Driver Version: 510.39.01    CUDA Version: 11.6     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  NVIDIA GeForce ...  On   | 00000000:65:00.0 Off |                  N/A |
| 30%   29C    P8    15W / 350W |      0MiB / 24576MiB |      0%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+
Looking in indexes: https://pypi.org/simple, https://pypi.ngc.nvidia.com
Collecting ipywidgets
  Downloading ipywidgets-7.6.5-py2.py3-none-any.whl (121 kB)
     |████████████████████████████████| 121 kB 5.3 MB/s eta 0:00:01
Requirement already satisfied: ipython-genutils~=0.2.0 in /opt/conda/lib/python3.8/site-packages (from ipywidgets) (0.2.0)
Requirement already satisfied: ipykernel>=4.5.1 in /opt/conda/lib/python3.8/site-packages (from ipywidgets) (6.7.0)
Requirement already satisfied: nbformat>=4.2.0 in /opt/conda/lib/python3.8/site-packages (from ipywidgets) (5.1.3)
Requirement already satisfied: ipython>=4.0.0 in /opt/conda/lib/python3.8/site-packages (from ipywidgets) (7.31.0)
Collecting jupyterlab-widgets>=1.0.0
  Downloading jupyterlab_widgets-1.0.2-py3-none-any.whl (243 kB)
     |████████████████████████████████| 243 kB 2.6 MB/s eta 0:00:01
Collecting widgetsnbextension~=3.5.0
  Downloading widgetsnbextension-3.5.2-py2.py3-none-any.whl (1.6 MB)
     |████████████████████████████████| 1.6 MB 3.5 MB/s eta 0:00:01
Requirement already satisfied: traitlets>=4.3.1 in /opt/conda/lib/python3.8/site-packages (from ipywidgets) (5.1.1)
Requirement already satisfied: jupyter-client<8.0 in /opt/conda/lib/python3.8/site-packages (from ipykernel>=4.5.1->ipywidgets) (7.1.2)
Requirement already satisfied: tornado<7.0,>=4.2 in /opt/conda/lib/python3.8/site-packages (from ipykernel>=4.5.1->ipywidgets) (6.1)
Requirement already satisfied: nest-asyncio in /opt/conda/lib/python3.8/site-packages (from ipykernel>=4.5.1->ipywidgets) (1.5.4)
Requirement already satisfied: debugpy<2.0,>=1.0.0 in /opt/conda/lib/python3.8/site-packages (from ipykernel>=4.5.1->ipywidgets) (1.5.1)
Requirement already satisfied: matplotlib-inline<0.2.0,>=0.1.0 in /opt/conda/lib/python3.8/site-packages (from ipykernel>=4.5.1->ipywidgets) (0.1.3)
Requirement already satisfied: pexpect>4.3 in /opt/conda/lib/python3.8/site-packages (from ipython>=4.0.0->ipywidgets) (4.8.0)
Requirement already satisfied: setuptools>=18.5 in /opt/conda/lib/python3.8/site-packages (from ipython>=4.0.0->ipywidgets) (59.5.0)
Requirement already satisfied: jedi>=0.16 in /opt/conda/lib/python3.8/site-packages (from ipython>=4.0.0->ipywidgets) (0.18.1)
Requirement already satisfied: decorator in /opt/conda/lib/python3.8/site-packages (from ipython>=4.0.0->ipywidgets) (5.1.0)
Requirement already satisfied: prompt-toolkit!=3.0.0,!=3.0.1,<3.1.0,>=2.0.0 in /opt/conda/lib/python3.8/site-packages (from ipython>=4.0.0->ipywidgets) (3.0.24)
Requirement already satisfied: backcall in /opt/conda/lib/python3.8/site-packages (from ipython>=4.0.0->ipywidgets) (0.2.0)
Requirement already satisfied: pickleshare in /opt/conda/lib/python3.8/site-packages (from ipython>=4.0.0->ipywidgets) (0.7.5)
Requirement already satisfied: pygments in /opt/conda/lib/python3.8/site-packages (from ipython>=4.0.0->ipywidgets) (2.11.1)
Requirement already satisfied: parso<0.9.0,>=0.8.0 in /opt/conda/lib/python3.8/site-packages (from jedi>=0.16->ipython>=4.0.0->ipywidgets) (0.8.3)
Requirement already satisfied: entrypoints in /opt/conda/lib/python3.8/site-packages (from jupyter-client<8.0->ipykernel>=4.5.1->ipywidgets) (0.3)
Requirement already satisfied: pyzmq>=13 in /opt/conda/lib/python3.8/site-packages (from jupyter-client<8.0->ipykernel>=4.5.1->ipywidgets) (22.3.0)
Requirement already satisfied: python-dateutil>=2.1 in /opt/conda/lib/python3.8/site-packages (from jupyter-client<8.0->ipykernel>=4.5.1->ipywidgets) (2.8.2)
Requirement already satisfied: jupyter-core>=4.6.0 in /opt/conda/lib/python3.8/site-packages (from jupyter-client<8.0->ipykernel>=4.5.1->ipywidgets) (4.9.1)
Requirement already satisfied: jsonschema!=2.5.0,>=2.4 in /opt/conda/lib/python3.8/site-packages (from nbformat>=4.2.0->ipywidgets) (4.4.0)
Requirement already satisfied: attrs>=17.4.0 in /opt/conda/lib/python3.8/site-packages (from jsonschema!=2.5.0,>=2.4->nbformat>=4.2.0->ipywidgets) (21.4.0)
Requirement already satisfied: importlib-resources>=1.4.0 in /opt/conda/lib/python3.8/site-packages (from jsonschema!=2.5.0,>=2.4->nbformat>=4.2.0->ipywidgets) (5.4.0)
Requirement already satisfied: pyrsistent!=0.17.0,!=0.17.1,!=0.17.2,>=0.14.0 in /opt/conda/lib/python3.8/site-packages (from jsonschema!=2.5.0,>=2.4->nbformat>=4.2.0->ipywidgets) (0.18.1)
Requirement already satisfied: zipp>=3.1.0 in /opt/conda/lib/python3.8/site-packages (from importlib-resources>=1.4.0->jsonschema!=2.5.0,>=2.4->nbformat>=4.2.0->ipywidgets) (3.7.0)
Requirement already satisfied: ptyprocess>=0.5 in /opt/conda/lib/python3.8/site-packages (from pexpect>4.3->ipython>=4.0.0->ipywidgets) (0.7.0)
Requirement already satisfied: wcwidth in /opt/conda/lib/python3.8/site-packages (from prompt-toolkit!=3.0.0,!=3.0.1,<3.1.0,>=2.0.0->ipython>=4.0.0->ipywidgets) (0.2.5)
Requirement already satisfied: six>=1.5 in /opt/conda/lib/python3.8/site-packages (from python-dateutil>=2.1->jupyter-client<8.0->ipykernel>=4.5.1->ipywidgets) (1.16.0)
Requirement already satisfied: notebook>=4.4.1 in /opt/conda/lib/python3.8/site-packages (from widgetsnbextension~=3.5.0->ipywidgets) (6.4.1)
Requirement already satisfied: Send2Trash>=1.5.0 in /opt/conda/lib/python3.8/site-packages (from notebook>=4.4.1->widgetsnbextension~=3.5.0->ipywidgets) (1.8.0)
Requirement already satisfied: jinja2 in /opt/conda/lib/python3.8/site-packages (from notebook>=4.4.1->widgetsnbextension~=3.5.0->ipywidgets) (3.0.3)
Requirement already satisfied: argon2-cffi in /opt/conda/lib/python3.8/site-packages (from notebook>=4.4.1->widgetsnbextension~=3.5.0->ipywidgets) (21.3.0)
Requirement already satisfied: prometheus-client in /opt/conda/lib/python3.8/site-packages (from notebook>=4.4.1->widgetsnbextension~=3.5.0->ipywidgets) (0.13.0)
Requirement already satisfied: terminado>=0.8.3 in /opt/conda/lib/python3.8/site-packages (from notebook>=4.4.1->widgetsnbextension~=3.5.0->ipywidgets) (0.12.1)
Requirement already satisfied: nbconvert in /opt/conda/lib/python3.8/site-packages (from notebook>=4.4.1->widgetsnbextension~=3.5.0->ipywidgets) (6.4.0)
Requirement already satisfied: argon2-cffi-bindings in /opt/conda/lib/python3.8/site-packages (from argon2-cffi->notebook>=4.4.1->widgetsnbextension~=3.5.0->ipywidgets) (21.2.0)
Requirement already satisfied: cffi>=1.0.1 in /opt/conda/lib/python3.8/site-packages (from argon2-cffi-bindings->argon2-cffi->notebook>=4.4.1->widgetsnbextension~=3.5.0->ipywidgets) (1.15.0)
Requirement already satisfied: pycparser in /opt/conda/lib/python3.8/site-packages (from cffi>=1.0.1->argon2-cffi-bindings->argon2-cffi->notebook>=4.4.1->widgetsnbextension~=3.5.0->ipywidgets) (2.21)
Requirement already satisfied: MarkupSafe>=2.0 in /opt/conda/lib/python3.8/site-packages (from jinja2->notebook>=4.4.1->widgetsnbextension~=3.5.0->ipywidgets) (2.0.1)
Requirement already satisfied: jupyterlab-pygments in /opt/conda/lib/python3.8/site-packages (from nbconvert->notebook>=4.4.1->widgetsnbextension~=3.5.0->ipywidgets) (0.1.2)
Requirement already satisfied: defusedxml in /opt/conda/lib/python3.8/site-packages (from nbconvert->notebook>=4.4.1->widgetsnbextension~=3.5.0->ipywidgets) (0.7.1)
Requirement already satisfied: bleach in /opt/conda/lib/python3.8/site-packages (from nbconvert->notebook>=4.4.1->widgetsnbextension~=3.5.0->ipywidgets) (4.1.0)
Requirement already satisfied: nbclient<0.6.0,>=0.5.0 in /opt/conda/lib/python3.8/site-packages (from nbconvert->notebook>=4.4.1->widgetsnbextension~=3.5.0->ipywidgets) (0.5.10)
Requirement already satisfied: testpath in /opt/conda/lib/python3.8/site-packages (from nbconvert->notebook>=4.4.1->widgetsnbextension~=3.5.0->ipywidgets) (0.5.0)
Requirement already satisfied: pandocfilters>=1.4.1 in /opt/conda/lib/python3.8/site-packages (from nbconvert->notebook>=4.4.1->widgetsnbextension~=3.5.0->ipywidgets) (1.5.0)
Requirement already satisfied: mistune<2,>=0.8.1 in /opt/conda/lib/python3.8/site-packages (from nbconvert->notebook>=4.4.1->widgetsnbextension~=3.5.0->ipywidgets) (0.8.4)
Requirement already satisfied: packaging in /opt/conda/lib/python3.8/site-packages (from bleach->nbconvert->notebook>=4.4.1->widgetsnbextension~=3.5.0->ipywidgets) (21.3)
Requirement already satisfied: webencodings in /opt/conda/lib/python3.8/site-packages (from bleach->nbconvert->notebook>=4.4.1->widgetsnbextension~=3.5.0->ipywidgets) (0.5.1)
Requirement already satisfied: pyparsing!=3.0.5,>=2.0.2 in /opt/conda/lib/python3.8/site-packages (from packaging->bleach->nbconvert->notebook>=4.4.1->widgetsnbextension~=3.5.0->ipywidgets) (3.0.6)
Installing collected packages: widgetsnbextension, jupyterlab-widgets, ipywidgets
Successfully installed ipywidgets-7.6.5 jupyterlab-widgets-1.0.2 widgetsnbextension-3.5.2
WARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv

## 1. Requirements

NVIDIA’s NGC provides PyTorch Docker Container which contains PyTorch and Torch-TensorRT. We can make use of latest pytorch container to run this notebook.

Otherwise, you can follow the steps in notebooks/README to prepare a Docker container yourself, within which you can run this demo notebook.

## 2. ResNet-50 Overview

PyTorch has a model repository called the PyTorch Hub, which is a source for high quality implementations of common models. We can get our ResNet-50 model from there pretrained on ImageNet.

Model Description

This ResNet-50 model is based on the Deep Residual Learning for Image Recognition paper, which describes ResNet as “a method for detecting objects in images using a single deep neural network”. The input size is fixed to 32x32.

alt

## 3. Running the model without optimizations

PyTorch has a model repository called timm, which is a source for high quality implementations of computer vision models. We can get our EfficientNet model from there pretrained on ImageNet.

[3]:

import torch
import torchvision

torch.hub._validate_not_a_forked_repo=lambda a,b,c: True

resnet50_model = torch.hub.load('pytorch/vision:v0.10.0', 'resnet50', pretrained=True)
resnet50_model.eval()

Downloading: "https://github.com/pytorch/vision/archive/v0.10.0.zip" to /root/.cache/torch/hub/v0.10.0.zip
Downloading: "https://download.pytorch.org/models/resnet50-0676ba61.pth" to /root/.cache/torch/hub/checkpoints/resnet50-0676ba61.pth

[3]:

ResNet(
  (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)
  (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
  (relu): ReLU(inplace=True)
  (maxpool): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)
  (layer1): Sequential(
    (0): Bottleneck(
      (conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
      (downsample): Sequential(
        (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
    )
    (1): Bottleneck(
      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
    (2): Bottleneck(
      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
  )
  (layer2): Sequential(
    (0): Bottleneck(
      (conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
      (downsample): Sequential(
        (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)
        (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
    )
    (1): Bottleneck(
      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
    (2): Bottleneck(
      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
    (3): Bottleneck(
      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
  )
  (layer3): Sequential(
    (0): Bottleneck(
      (conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
      (downsample): Sequential(
        (0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2), bias=False)
        (1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
    )
    (1): Bottleneck(
      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
    (2): Bottleneck(
      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
    (3): Bottleneck(
      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
    (4): Bottleneck(
      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
    (5): Bottleneck(
      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
  )
  (layer4): Sequential(
    (0): Bottleneck(
      (conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
      (downsample): Sequential(
        (0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(2, 2), bias=False)
        (1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
    )
    (1): Bottleneck(
      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
    (2): Bottleneck(
      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU(inplace=True)
    )
  )
  (avgpool): AdaptiveAvgPool2d(output_size=(1, 1))
  (fc): Linear(in_features=2048, out_features=1000, bias=True)
)

With our model loaded, let’s proceed to downloading some images!

[4]:

!mkdir -p ./data
!wget  -O ./data/img0.JPG "https://d17fnq9dkz9hgj.cloudfront.net/breed-uploads/2018/08/siberian-husky-detail.jpg?bust=1535566590&width=630"
!wget  -O ./data/img1.JPG "https://www.hakaimagazine.com/wp-content/uploads/header-gulf-birds.jpg"
!wget  -O ./data/img2.JPG "https://www.artis.nl/media/filer_public_thumbnails/filer_public/00/f1/00f1b6db-fbed-4fef-9ab0-84e944ff11f8/chimpansee_amber_r_1920x1080.jpg__1920x1080_q85_subject_location-923%2C365_subsampling-2.jpg"
!wget  -O ./data/img3.JPG "https://www.familyhandyman.com/wp-content/uploads/2018/09/How-to-Avoid-Snakes-Slithering-Up-Your-Toilet-shutterstock_780480850.jpg"

!wget  -O ./data/imagenet_class_index.json "https://s3.amazonaws.com/deep-learning-models/image-models/imagenet_class_index.json"

--2022-02-08 19:17:19--  https://d17fnq9dkz9hgj.cloudfront.net/breed-uploads/2018/08/siberian-husky-detail.jpg?bust=1535566590&width=630
Resolving d17fnq9dkz9hgj.cloudfront.net (d17fnq9dkz9hgj.cloudfront.net)... 18.65.227.99, 18.65.227.37, 18.65.227.223, ...
Connecting to d17fnq9dkz9hgj.cloudfront.net (d17fnq9dkz9hgj.cloudfront.net)|18.65.227.99|:443... connected.
HTTP request sent, awaiting response... 200 OK
Length: 24112 (24K) [image/jpeg]
Saving to: ‘./data/img0.JPG’

./data/img0.JPG     100%[===================>]  23.55K  --.-KB/s    in 0.002s

2022-02-08 19:17:19 (14.6 MB/s) - ‘./data/img0.JPG’ saved [24112/24112]

--2022-02-08 19:17:19--  https://www.hakaimagazine.com/wp-content/uploads/header-gulf-birds.jpg
Resolving www.hakaimagazine.com (www.hakaimagazine.com)... 164.92.73.117
Connecting to www.hakaimagazine.com (www.hakaimagazine.com)|164.92.73.117|:443... connected.
HTTP request sent, awaiting response... 200 OK
Length: 452718 (442K) [image/jpeg]
Saving to: ‘./data/img1.JPG’

./data/img1.JPG     100%[===================>] 442.11K  2.40MB/s    in 0.2s

2022-02-08 19:17:19 (2.40 MB/s) - ‘./data/img1.JPG’ saved [452718/452718]

--2022-02-08 19:17:20--  https://www.artis.nl/media/filer_public_thumbnails/filer_public/00/f1/00f1b6db-fbed-4fef-9ab0-84e944ff11f8/chimpansee_amber_r_1920x1080.jpg__1920x1080_q85_subject_location-923%2C365_subsampling-2.jpg
Resolving www.artis.nl (www.artis.nl)... 94.75.225.20
Connecting to www.artis.nl (www.artis.nl)|94.75.225.20|:443... connected.
HTTP request sent, awaiting response... 200 OK
Length: 361413 (353K) [image/jpeg]
Saving to: ‘./data/img2.JPG’

./data/img2.JPG     100%[===================>] 352.94K   366KB/s    in 1.0s

2022-02-08 19:17:24 (366 KB/s) - ‘./data/img2.JPG’ saved [361413/361413]

--2022-02-08 19:17:24--  https://www.familyhandyman.com/wp-content/uploads/2018/09/How-to-Avoid-Snakes-Slithering-Up-Your-Toilet-shutterstock_780480850.jpg
Resolving www.familyhandyman.com (www.familyhandyman.com)... 104.18.201.107, 104.18.202.107, 2606:4700::6812:ca6b, ...
Connecting to www.familyhandyman.com (www.familyhandyman.com)|104.18.201.107|:443... connected.
HTTP request sent, awaiting response... 200 OK
Length: 94328 (92K) [image/jpeg]
Saving to: ‘./data/img3.JPG’

./data/img3.JPG     100%[===================>]  92.12K  --.-KB/s    in 0.005s

2022-02-08 19:17:25 (16.8 MB/s) - ‘./data/img3.JPG’ saved [94328/94328]

--2022-02-08 19:17:25--  https://s3.amazonaws.com/deep-learning-models/image-models/imagenet_class_index.json
Resolving s3.amazonaws.com (s3.amazonaws.com)... 52.217.41.222
Connecting to s3.amazonaws.com (s3.amazonaws.com)|52.217.41.222|:443... connected.
HTTP request sent, awaiting response... 200 OK
Length: 35363 (35K) [application/octet-stream]
Saving to: ‘./data/imagenet_class_index.json’

./data/imagenet_cla 100%[===================>]  34.53K  --.-KB/s    in 0.07s

2022-02-08 19:17:26 (481 KB/s) - ‘./data/imagenet_class_index.json’ saved [35363/35363]

All pre-trained models expect input images normalized in the same way, i.e. mini-batches of 3-channel RGB images of shape (3 x H x W), where H and W are expected to be at least 224. The images have to be loaded in to a range of [0, 1] and then normalized using mean = [0.485, 0.456, 0.406] and std = [0.229, 0.224, 0.225].

Here’s a sample execution.

[5]:

from PIL import Image
from torchvision import transforms
import matplotlib.pyplot as plt
import json

fig, axes = plt.subplots(nrows=2, ncols=2)

for i in range(4):
    img_path = './data/img%d.JPG'%i
    img = Image.open(img_path)
    preprocess = transforms.Compose([
        transforms.Resize(256),
        transforms.CenterCrop(224),
        transforms.ToTensor(),
        transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
    ])
    input_tensor = preprocess(img)
    plt.subplot(2,2,i+1)
    plt.imshow(img)
    plt.axis('off')

# loading labels
with open("./data/imagenet_class_index.json") as json_file:
    d = json.load(json_file)
../_images/_notebooks_Resnet50-example_12_0.png

Throughout this tutorial, we will be making use of some utility functions; rn50_preprocess for preprocessing input images, predict to use the model for prediction and benchmark to benchmark the inference. You do not need to understand/go through these utilities to make use of Torch TensorRT, but are welecomed to do so if you choose.

[6]:

import numpy as np
import time
import torch.backends.cudnn as cudnn
cudnn.benchmark = True

def rn50_preprocess():
    preprocess = transforms.Compose([
        transforms.Resize(256),
        transforms.CenterCrop(224),
        transforms.ToTensor(),
        transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
    ])
    return preprocess

# decode the results into ([predicted class, description], probability)
def predict(img_path, model):
    img = Image.open(img_path)
    preprocess = rn50_preprocess()
    input_tensor = preprocess(img)
    input_batch = input_tensor.unsqueeze(0) # create a mini-batch as expected by the model

    # move the input and model to GPU for speed if available
    if torch.cuda.is_available():
        input_batch = input_batch.to('cuda')
        model.to('cuda')

    with torch.no_grad():
        output = model(input_batch)
        # Tensor of shape 1000, with confidence scores over Imagenet's 1000 classes
        sm_output = torch.nn.functional.softmax(output[0], dim=0)

    ind = torch.argmax(sm_output)
    return d[str(ind.item())], sm_output[ind] #([predicted class, description], probability)

def benchmark(model, input_shape=(1024, 1, 224, 224), dtype='fp32', nwarmup=50, nruns=10000):
    input_data = torch.randn(input_shape)
    input_data = input_data.to("cuda")
    if dtype=='fp16':
        input_data = input_data.half()

    print("Warm up ...")
    with torch.no_grad():
        for _ in range(nwarmup):
            features = model(input_data)
    torch.cuda.synchronize()
    print("Start timing ...")
    timings = []
    with torch.no_grad():
        for i in range(1, nruns+1):
            start_time = time.time()
            features = model(input_data)
            torch.cuda.synchronize()
            end_time = time.time()
            timings.append(end_time - start_time)
            if i%10==0:
                print('Iteration %d/%d, ave batch time %.2f ms'%(i, nruns, np.mean(timings)*1000))

    print("Input shape:", input_data.size())
    print("Output features size:", features.size())
    print('Average batch time: %.2f ms'%(np.mean(timings)*1000))

With the model downloaded and the util functions written, let’s just quickly see some predictions, and benchmark the model in its current un-optimized state.

[7]:

for i in range(4):
    img_path = './data/img%d.JPG'%i
    img = Image.open(img_path)

    pred, prob = predict(img_path, resnet50_model)
    print('{} - Predicted: {}, Probablility: {}'.format(img_path, pred, prob))

    plt.subplot(2,2,i+1)
    plt.imshow(img);
    plt.axis('off');
    plt.title(pred[1])

./data/img0.JPG - Predicted: ['n02110185', 'Siberian_husky'], Probablility: 0.49785590171813965
./data/img1.JPG - Predicted: ['n01820546', 'lorikeet'], Probablility: 0.6445754766464233
./data/img2.JPG - Predicted: ['n02481823', 'chimpanzee'], Probablility: 0.9899807572364807
./data/img3.JPG - Predicted: ['n01749939', 'green_mamba'], Probablility: 0.39485082030296326
../_images/_notebooks_Resnet50-example_16_1.png 
[8]:

# Model benchmark without Torch-TensorRT
model = resnet50_model.eval().to("cuda")
benchmark(model, input_shape=(128, 3, 224, 224), nruns=100)

Warm up ...
Start timing ...
Iteration 10/100, ave batch time 75.34 ms
Iteration 20/100, ave batch time 75.33 ms
Iteration 30/100, ave batch time 75.35 ms
Iteration 40/100, ave batch time 75.37 ms
Iteration 50/100, ave batch time 75.38 ms
Iteration 60/100, ave batch time 75.38 ms
Iteration 70/100, ave batch time 75.39 ms
Iteration 80/100, ave batch time 75.39 ms
Iteration 90/100, ave batch time 75.40 ms
Iteration 100/100, ave batch time 75.41 ms
Input shape: torch.Size([128, 3, 224, 224])
Output features size: torch.Size([128, 1000])
Average batch time: 75.41 ms

## 4. Accelerating with Torch-TensorRT

Onwards to the next step, accelerating with Torch TensorRT. In these examples we showcase the results for FP32 (single precision) and FP16 (half precision). We do not demonstrat specific tuning, just showcase the simplicity of usage. If you want to learn more about the possible customizations, visit our documentation.

FP32 (single precision)

[9]:

import torch_tensorrt

# The compiled module will have precision as specified by "op_precision".
# Here, it will have FP32 precision.
trt_model_fp32 = torch_tensorrt.compile(model, inputs = [torch_tensorrt.Input((128, 3, 224, 224), dtype=torch.float32)],
    enabled_precisions = torch.float32, # Run with FP32
    workspace_size = 1 << 22
)

WARNING: [Torch-TensorRT] - Dilation not used in Max pooling converter

[10]:

# Obtain the average time taken by a batch of input
benchmark(trt_model_fp32, input_shape=(128, 3, 224, 224), nruns=100)

Warm up ...
Start timing ...
Iteration 10/100, ave batch time 41.02 ms
Iteration 20/100, ave batch time 41.12 ms
Iteration 30/100, ave batch time 41.22 ms
Iteration 40/100, ave batch time 41.14 ms
Iteration 50/100, ave batch time 41.20 ms
Iteration 60/100, ave batch time 41.20 ms
Iteration 70/100, ave batch time 41.19 ms
Iteration 80/100, ave batch time 41.23 ms
Iteration 90/100, ave batch time 41.20 ms
Iteration 100/100, ave batch time 41.21 ms
Input shape: torch.Size([128, 3, 224, 224])
Output features size: torch.Size([128, 1000])
Average batch time: 41.21 ms

FP16 (half precision)

[11]:

import torch_tensorrt

# The compiled module will have precision as specified by "op_precision".
# Here, it will have FP16 precision.
trt_model_fp16 = torch_tensorrt.compile(model, inputs = [torch_tensorrt.Input((128, 3, 224, 224), dtype=torch.half)],
    enabled_precisions = {torch.half}, # Run with FP32
    workspace_size = 1 << 22
)

WARNING: [Torch-TensorRT] - For input x.1, found user specified input dtype as Float16, however when inspecting the graph, the input type expected was inferred to be Float
The compiler is going to use the user setting Float16
This conflict may cause an error at runtime due to partial compilation being enabled and therefore
compatibility with PyTorch's data type convention is required.
If you do indeed see errors at runtime either:
- Remove the dtype spec for x.1
- Disable partial compilation by setting require_full_compilation to True
WARNING: [Torch-TensorRT] - Dilation not used in Max pooling converter

[13]:

# Obtain the average time taken by a batch of input
benchmark(trt_model_fp16, input_shape=(128, 3, 224, 224), dtype='fp16', nruns=100)

Warm up ...
Start timing ...
Iteration 10/100, ave batch time 14.48 ms
Iteration 20/100, ave batch time 14.58 ms
Iteration 30/100, ave batch time 14.72 ms
Iteration 40/100, ave batch time 14.73 ms
Iteration 50/100, ave batch time 14.70 ms
Iteration 60/100, ave batch time 14.79 ms
Iteration 70/100, ave batch time 14.73 ms
Iteration 80/100, ave batch time 14.69 ms
Iteration 90/100, ave batch time 14.68 ms
Iteration 100/100, ave batch time 14.69 ms
Input shape: torch.Size([128, 3, 224, 224])
Output features size: torch.Size([128, 1000])
Average batch time: 14.69 ms

## 5. Conclusion

In this notebook, we have walked through the complete process of compiling TorchScript models with Torch-TensorRT for EfficientNet-B0 model and test the performance impact of the optimization. With Torch-TensorRT, we observe a speedup of 1.84x with FP32, and 5.2x with FP16 on an NVIDIA 3090 GPU. These acceleration numbers will vary from GPU to GPU(as well as implementation to implementation based on the ops used) and we encorage you to try out latest generation of Data center compute cards for maximum acceleration.

What’s next

Now it’s time to try Torch-TensorRT on your own model. If you run into any issues, you can fill them at https://github.com/NVIDIA/Torch-TensorRT. Your involvement will help future development of Torch-TensorRT.

Docs

Access comprehensive developer documentation for PyTorch

View Docs

Tutorials

Get in-depth tutorials for beginners and advanced developers

View Tutorials

Resources

Find development resources and get your questions answered

View Resources

To analyze traffic and optimize your experience, we serve cookies on this site. By clicking or navigating, you agree to allow our usage of cookies. As the current maintainers of this site, Facebook’s Cookies Policy applies. Learn more, including about available controls: Cookies Policy.