Hugging Face Hub enables CO2 emissions filtering and CodeCarbon tracking for model training

AI Impact Summary

The Hugging Face Hub now supports quantifying and curating model emissions: a new emissions_threshold parameter allows filtering models by CO2 output, and model cards can surface co2_eq_emissions data. Transformers users can auto-track emissions during training via the CodeCarbon integration, with emissions.csv generated and included in model cards for transparency. This enables eco-conscious model selection and ESG reporting, but teams should account for environment-dependent variability in CO2 figures and treat training emissions as one dimension of total deployment energy.