Muonly

Recipes

Install dependencies

If micromamba is not already installed on your system, you can install it easily using the following command:

"${SHELL}" <(curl -L micro.mamba.pm/install.sh)

See more details at: https://mamba.readthedocs.io/en/latest/installation/micromamba-installation.html

Then, create the environment using the provided environment.yaml file:

micromamba create -y -f ./environment.yaml

Setup environment

Source a shell script depending on your shell:

• bash: source setup.sh
• fish: source setup.fish

Training on local machine

To run a training job with a sanity-check config on your local machine, use the following command:

./train.py debug=sanity-check

After the sanity check passes without any errors, you can start training the model by specifying your desired configurations in the prompt. For example:

./train.py model=latent_attention model.model_dim=128 optimizer.lr=0.0001 datamodule.batch_size=256

Submit training job into a cluster

./submit.py -h

./submit.py --debug sanity-check

./submit.py --model latent_attention -a 'model.model_dim=128 optimizer.lr=0.0001 datamodule.batch_size=256'

Monitor training logs with Aim UI

Open Aim UI

aim up --port <PORT>

If you are working on a remote server, you need to set up port forwarding