Feature Request: Multiple Models / Backends for Different Tasks

[quack8035]

A model that excels at writing compelling prose may not be necessary for reasoning tasks (e.g. has the location changed, has a piece of state changed, has a specific event occurred). Would it make sense to support multiple backends for prose vs. “maintenace” tasks?

[p-e-w]

Yes, this is indeed planned. Note that different sampling parameters for generation vs. narration are already supported.