Beam Search
Beam search is a decoding strategy that explores multiple candidate output sequences simultaneously during text generation, keeping the top-k most probable sequences (the "beam width") at each step. Unlike greedy decoding which always picks the single highest-probability token, beam search considers that a lower-probability token at one step might lead to a better overall sequence.
Example
With beam width 3, the model translating "The cat sat" explores three parallel paths: "Le chat était assis" (prob: 0.85), "Le chat s'est assis" (prob: 0.82), and "Le chat se trouvait assis" (prob: 0.71). It then continues expanding all three before selecting the highest-scoring complete sentence.
Frequently asked questions
What is Beam Search?
- Beam search is a decoding strategy that explores multiple candidate output sequences simultaneously during text generation, keeping the top-k most probable sequences (the "beam width") at each step.
Can you give an example of Beam Search?
- With beam width 3, the model translating "The cat sat" explores three parallel paths: "Le chat était assis" (prob: 0.85), "Le chat s'est assis" (prob: 0.82), and "Le chat se trouvait assis" (prob: 0.71). It then continues expanding all three before selecting the highest-scoring complete sentence.
Related Terms
Put this into practice
Build polished, copy-ready prompts in under 60 seconds with SurePrompts.
Try SurePrompts