Running an inference of between 10 and 50 queries on a large language model uses about 500 ml of water ... footprint among ...