Hugging Face Carbon Emissions Tracking

Hugging Face supports new features and tools to manage energy consumption and CO2 emissions issues that occur during model training and deployment processes.

Key Features:

• Low-Emission Model Search: An emissions_threshold parameter has been added to the huggingface_hub library. This allows users to filter and search for models based on a specific carbon emissions (g) range, making it easy to find environmentally friendly models.
• Automatic Carbon Emissions Reporting: Integration between the transformers library and codecarbon is supported. When using the Trainer object, if codecarbon is installed, carbon emissions generated during training are automatically tracked and saved as an emissions.csv file.

These features help increase transparency in AI development and assist developers in selecting and reporting models while considering their environmental impact.