[hazrmard]

Do I understand this right?

A light-weight speculative model adapts to usage, keeping the acceptance rate for the static heavy-weight model within acceptable bounds.

Do they adapt with LoRAs?