Found 1 document(s)
Page 1 of 1
-
On-policy self-distillation (OPSD) enables a language model to learn from a Teacher instantiated from the same model but given access to privileged information, such as a reference solution. We find that access to suffic...