1 paper · 1 filter
Ramin Pishehvar, Andrea Morandi, Mahesh Viswanathan
Deciding when to escalate a query from a small language model to a larger one requires a cheap signal that predicts, before the large model is called, whether escalating would help…