Targeted instruction tuning for LLMs involves selecting a subset of instructions from a candidate pool using a small query set from target tasks. Despite growing interest, we still lack guidance on what to select. Our new preprint brings clarity to this space (thread 👇).

This work wouldn’t have been possible without my awesome collaborators: Paula Rodriguez-Diaz (@paularodrid), Neha Hulkund (@NHulkund), Sara Beery (@sarameghanbeery), and David Alvarez-Melis (@elmelis).