Skip to content

Yaml usage

Using LaikaLLM via .yaml is really simple:

  • You define the steps of your experiment following the data-model-evaluation logic
  • You invoke LaikaLLM with the following command:
python LaikaLLM.py -c params.yml # (1)
  1. Instead of params.yml, specify the path to your .yaml file

Info

The necessary requirement for using LaikaLLM with the user-defined .yaml configuration is to set the root of the repository as the Working Directory

Yaml interface

The .yaml file parameters can be grouped into four different macro-section:

  1. The first section contains general parameters needed by all the other sections, like the experiment name, the random state, the device to use, etc.
  2. The second section contains all the parameters needed by the data phase of the experiment, such as the dataset to use and its parameters
  3. The third section contains all the parameters needed by the model phase of the experiment, such as the model to use, the number of epochs to train the model, the train batch size, etc.
  4. The fourth section contains all the parameters needed for the eval phase of the experiment, such as the eval batch size, the metrics to use, etc.

You can check each subsection to see all its customizable parameters!