
Slowly but surely
Great piece.
It is great to see a counter to the ChatGPT bloat. I read a post from someone here that ran the Llama model on a box with 128gb of RAM and it ended in it sprouting nonsense in a short time.
This area of optimisation and compression is a driving force ATM, and that's great 'cause no way can bloat-beasts like ChatGPT continue to consume the World's resources.
These 'lean' LLM face huge knock backs from the power companies. OpenAI is just a front for Sam's power business.