“Maria identifies key challenges in deploying fine-tuned LLMs, including the costs of inference, electricity consumption, and the resulting CO2 footprint.”