Publication

Language Model Decoding as Likelihood–Utility Alignment