The idea would be to take some existing foundational model (like Llama) and perform some type of fine-tuning. Fine-tuning can be hardware intensive (but not as intensive as training a foundational model). Once the model is tuned, using it, called inferencing, is not nearly as hardware intensive.
