Dependencies and runtime environment

By default, all multinode compute primitives run on the python:3-bullseye Docker image.

Although this may suffice, you may require additional dependencies, or even a completely different runtime environment.

Additional python dependencies

multinode will look for a requirements.txt file in your current working directory, and will automatically install all the listed dependencies on the cloud workers. If a requirements.txt file is not found, a warning will be raised, and the standard python:3-bullseye image will be used.

Sometimes, you may want to install a particular dependency, but only for a specific function. For example, this may reduce the time it takes to pull the image when the worker is provisioning. To do this, use the additional_dependencies keyword argument in the decorator.

# file: requirements.txt
pandas==1.4.3
numpy==1.21.2
pillow==8.3.1

# file: code.py

import multinode as mn

@mn.function()
def some_standard_function():
    # Code will run on a worker with
    # pandas, numpy, and pillow installed

@mn.function(additional_dependencies=[
    "torch==1.9.0",
    "botorch==0.6.5",
])
def some_torch_function():
    # Code will run on a worker with
    # pandas, numpy, pillow, torch, and botorch installed

Custom docker images