Changing the weights and controls does not alleviate the context limits causing LLMs, local or cloud, to forget over time. I was asking how local LLMs would amelioratile this but looks like there is no solution.