Installing AQUA-Lib

The cornerstone of Aqua is a pip installable python library which enables sharing idle GPU memory over NvLinks and dynamically reclaiming it when the demands are high. Follow these instructions to install Aqua in your python environment

git clone https://github.com/aquaml/aqua

Next, activate your conda environment (e.g., mygenai) and navigate to the Aqua directory:

conda activate mygenai
cd aqua

Finally, install Aqua using pip:

python3 -m pip install -e .

Installing AQUA-coordinator

Aqua requires workloads on GPUs to communicate to offer, reclaim, allocate and deallocate memory on each other. Aqua-coordinator enables this communication, you should start the coordinator on the server where you are running Aqua.

git clone https://github.com/aquaml/coordinator

Run the coordinator in your preferred python environment

cd coordinator
python3 -m pip install "fastapi[standard]"
python3 aqua_serve.py