Manta Light¶

Manta Light is designed to run inside containers deployed on nodes, allowing tasks to execute in isolated environments. This section guides you on building Docker images with Manta Light and the necessary dependencies for your custom algorithms.

Note

Manta Light will always be used within containers, so the installation procedure focuses on building the right Docker images that include Manta Light, required dependencies, and your algorithm-specific packages.

Downloading the Wheel File¶

Download the Manta Light Wheel File: Obtain the pre-built wheel file for Manta Light from the official release repository.
Install the Wheel File: Use pip to install the downloaded wheel file. Run the following command in your terminal:

pip install manta_light-<version>-py3-none-any.whl

This method is ideal for most users as it reduces compatibility issues and streamlines the installation process.

Building the Docker Image¶

To build a Docker image with Manta Light, create a Dockerfile. Below is an example Dockerfile that demonstrates how to set up the image using the Manta Light wheel file and indicates where to add your custom dependencies.

Dockerfile Example¶

# Use official manta light image, version 0.3b1 of manta-light with python 3.12
FROM hugomir/manta_light:0.3b1_py3.12 AS build

# Install your dependencies
# For example : torch and torchvision here
RUN pip3 install --no-cache-dir torch torchvision --index-url https://download.pytorch.org/whl/cpu

Once you have set your dependencies, you can build it by running :

docker build -t <your-tag> .

For example, if you want the tag fl-pytorch:latest, you can run :

docker build -t fl-pytorch:latest .