Use energy efficient AI/ML models
Large AI/ML and deep learning network models have a significant carbon footprint. Evaluate and use alternative, more energy efficient, models that provide similar functionality.
Evaluate energy efficient AI/ML models for development and inference. For example, DistilBERT provides similar functionality to the BERT model, running 60% faster while preserving 97% of BERT's performance. GPT-Neo 125M provides balanced results for energy consumption and accuracy compared to GPT-J 6B or GPT Neo 2.7B models.
SCI = (E * I) + M per R
Software Carbon Intensity Spec
Using energy efficient AI/ML models impacts SCI as follows:
E: Having an energy-efficient AI/ML model would provide efficient resource utilization and reduces the energy consumption of the compute resources and consequently, the E number should decrease.