​​​​​​​

System requirements

Operating systems Konduit Serving is supported on Linux, macOS and Windows.

Dependencies Ensure that you have JDK 8.0 installed. To use the Python SDK, install Python 3.7 and above.

Hardware requirements Binaries are provided for Intel/x86 architectures. For ARM support, see the Building from source page.

GPU: Hardware acceleration with CUDA version 10.1 (included in GPU build) is supported.

Installation

Install konduit from PyPI with

pip install konduit

If using the Anaconda distribution, you may install konduit from the konduitai Anaconda channel. First add the konduitai channel:

conda config --add channels konduitai

then install konduit with:

conda install -c konduitai konduit

You may need to install Cython before installing konduit using

pip install cython

We recommend using Python 3.7+.

Set environment variables manually

In the absence of the KONDUIT_JAR_PATH environment variable, the Python SDK looks for the Konduit Serving JAR file in ~/.konduit/konduit-serving. To overwrite this default, you can set a default location for the Konduit Serving JAR using environment variables.

setx KONDUIT_JAR_PATH "C:\Users\User\konduit-serving\konduit.jar"

export KONDUIT_JAR_PATH="~/konduit-serving/konduit.jar"

Common installation issues

1. Installing pyjnius returns

   WARNING: Not able to assign machine() = AMD64 to a cpu value! Using cpu = 'i386' instead!

   Fix: Ensure your JAVA environment variables point to a 64-bit version of Java if you're using a 64-bit version of Python, or a 32-bit version of Java if you're using a 32-bit version of Python (see kivy/pyjnius#390).

2. When running konduit commands on Windows, the following error message is returned:

   ImportError: DLL load failed: The specified module could not be found.

   Fix: On Windows, pyjnius requires an additional PATH variable to locate jvm.dll. Refer to the pyjnius documentation for details.