Contextual loading of a bigger model would incur a lot of latency. In this case it might be a good idea to farm out the request to a non-local model. Could be a

Contextual loading of a bigger model would incur a lot of latency. In this case it might be a good idea to farm out the request to a non-local model. Could be a good avenue for a hybrid local & remote completion system.