Kubnal Bridge

Techniques & Methods

Beam Search

Beam search explores multiple token sequences simultaneously, keeping the top-k (beam width) most probable partial sequences at each step. Unlike greedy decoding which commits to one token at a time, beam search can recover from locally suboptimal choices by considering alternatives.

Larger beam widths improve sequence quality but increase computation linearly. Beam search is widely used in machine translation and summarization but can produce generic, repetitive outputs for open-ended generation where diversity is preferred.

Authority Links

Related Terms