Install TorchServe¶

Prerequisites¶

Python: Required. Model Server for PyTorch (TorchServe) works with Python 3. When installing TorchServe, we recommend that you use a Python and Conda environment to avoid conflicts with your other Torch installations.
java 8: Required. TorchServe use java to serve HTTP requests. You must install java 8 (or later) and make sure java is on available in $PATH environment variable before installing TorchServe. If you have multiple java installed, you can use $JAVA_HOME environment variable to control which java to use.

For Ubuntu:

sudo apt-get install openjdk-11-jdk

For CentOS:

openjdk-11-jdk
sudo yum install java-11-openjdk

For macOS:

brew tap AdoptOpenJDK/openjdk
brew cask install adoptopenjdk11

You can also download and install Oracle JDK manually if you have trouble with above commands.

Torch: Recommended. TorchServe won’t install torch by default. Torch is required for most of examples in this project. TorchServe won’t install torch engine by default. And you can also choose specific version of torch if you want.
For virtualenv

#For CPU/GPU
pip install torch torchvision torchtext

For conda

#For CPU
conda install psutil pytorch torchvision torchtext -c pytorch

#For GPU
conda install future psutil pytorch torchvision cudatoolkit=10.1 torchtext -c pytorch

Curl: Optional. Curl is used in all of the examples. Install it with your preferred package manager.
Unzip: Optional. Unzip allows you to easily extract model files and inspect their content. If you choose to use it, associate it with .mar extensions.

Installation with pip¶

From PyTorch official repositories

pip install -f https://download.pytorch.org/torch_stable.html torchserve

Installation with conda¶

conda install -c pytorch torchserve

Running with Docker¶

CPU only¶

For specific versions use:

docker run --rm -it pytorch/torchserve:0.1-cpu

For the latest version you can use the latest tag:

docker run --rm -it pytorch/torchserve:latest

GPU¶

For the latest version you can use the gpu-latest tag:

docker run --rm -it --gpus all pytorch/torchserve:latest-gpu

For specific versions use:

docker run --rm -it --gpus all pytorch/torchserve:0.1-cuda10.1-cudnn7-runtime

Install TorchServe from Source Code¶

If you prefer, you can clone TorchServe from source code. First, run the following command:

git clone https://github.com/pytorch/serve.git
cd serve
pip install .

Notes:

In case pip install . step fails, try using python setup.py install and install the following python packages using pip install : Pillow, psutil, future

Install TorchServe for Development¶

If you plan to develop with TorchServe and change some of the source code, install it from source code and make your changes executable with this command:

pip install -e .

To upgrade TorchServe from source code and make changes executable, run:

pip install -U -e .

Troubleshooting Installation¶

Issue

Solution

java not found, please make sure JAVA_HOME is set properly.

Make sure java is installed. java is on the $PATH or $JAVA_HOME is set properly.

Your PYTHONPATH points to a site-packages dir for Python 3.x but you are running Python 2.x!

You do one of following:

use virtualenv
unset PYTHONPATH
set PYTHONPATH properly