Is it possible to supplement the model with a diff for updates on modular memory, or would severely impact perf?
I imagine you could do something like a LORA
this design at 7 transistors per weight is 99.9% burnt in the silicon forever.
I imagine you could do something like a LORA
this design at 7 transistors per weight is 99.9% burnt in the silicon forever.