[hugozap]

I wonder if these improvements in memory alteration will make it possible to create micro models using an approach based on pruning non relevant connections but preserving the reasoning abilities of large models like GPT4.