Latest News

In this demo, we provide a complete walkthrough of the SustainML User Interface, a framework designed to prioritize energy efficiency and reduce the carbon footprint of machine learning applications.

Whether you are a data scientist or an ML engineer, this tutorial shows you how to define problems, upload datasets, and compare models based on their environmental impact.

⬇️ Resources & Links

• SustainML Repository: https://github.com/eProsima/SustainML-Framework

📺 What’s Inside This Video:

1. Getting Started & Prerequisites. We cover the necessary installation steps and show you how to launch the interface directly from your terminal using the sustainml-framework command.

2. Defining Your ML Problem. Learn how to use the Problem Definition Screen to set up your project. We explain how to configure:

• Modality & Metrics: Select input types (like NLP) and evaluation metrics.
• Type Limiter & Hardware: Restrict model families (e.g., Transformers, CNNs) and target hardware.
• Model Goals: Define optimization objectives or let SustainML infer them automatically.

3. Analyzing Results & Green Metrics. Once you submit a problem, we dive into the Results Screen. You will see how to:

• View the Model Overview Table to check Power Consumption (W), Carbon Footprint (gCO2e), and Carbon Intensity.
• Select specific models to view detailed Iteration data, including Latency (ms) and Energy Consumption (kWh).
• Use the Comparison Graph to compare the environmental impact of different models side-by-side visually.

4. Advanced Features: Datasets & Hugging Face. We also explore the independent tools available in the interface:

• Upload Dataset: How to upload a file and let the system automatically extract metadata, keywords, and semantic profiles.
• Hugging Face Integration: How to discover and analyze models directly from the Hugging Face repository based on your problem description.
• U-Net Models: A look at the dedicated view for inspecting available U-Net architectures.

5. System Status Finally, we briefly look at the Settings panel to monitor the status of internal nodes like Hardware Constraints and Carbon Footprint estimators.