To run experiments, it is necessary to clone this repo, instad of installing fluid
from PyPI.
fluid
uses poetry
to manage its dependencies, so poetry
has to be
installed first.
curl -sSL https://raw.githubusercontent.com/python-poetry/poetry/master/get-poetry.py | python -
git clone https://github.com/SymbioticLab/fluid
cd fluid
poetry install --extras "exp"
poetry shell
The last command spawns a new shell within the virtual environment just created with all dependencies installed.
cluster.yaml
provides an example Ray cluster config file that you can use as the starting point to bring up a Ray cluster.
Each tune_*.py
file in this directory is an experiment that can run individualy.
To run them, first ensure the Ray cluster can be connected, then run python -m workloads.tune_asha_cifar
from the root directory of the repo.
After it finishes, some statistical results will be saved in trail_dataframe.csv
.