Running trained LLMs has, more or less, always been profitable. Inference can only run on one GPU efficiently, you know the size of the model, you know how long it takes… etc.
Inference is approximately as expensive and environment destroying as Call of Duty, which don’t get me wrong, is a waste of humanity for sure.
Running trained LLMs has, more or less, always been profitable. Inference can only run on one GPU efficiently, you know the size of the model, you know how long it takes… etc.
Inference is approximately as expensive and environment destroying as Call of Duty, which don’t get me wrong, is a waste of humanity for sure.
Model training however…