GAISSALabel is a tool designed to help evaluate the energy efficiency of both the training and inference phases of any machine learning model, which could be especially helpful for HPC services. It assesses the footprint of ML models, aiding in improving their efficiency and reducing the costs of running the models and their computation times.
To thoroughly test their tool, the development team aims to study more ML models using GAISSALabel. Consequently, the team would like to validate their work and establish collaborations with AI companies utilizing these ML modules, providing them with reports and recommendations.
The tool operates as a web-based system, allowing users to upload a series of metrics to assess the energy consumption of the ML model.
After the assessment, the tool provides the following metrics on a scale from A to E, along with optimization recommendations:
- CO2 emissions
- Performance score
- Downloads
- Size efficiency
- Dataset size efficiency
Additionally, an energy label is provided as a reference to indicate the model’s efficiency level.
You can find a video in the following link with a more detailed explanation of GAISSALabel.
Institution: Universitat Politècnica de Catalunya
Contact: Pablo Lago / tech@viromii.com
