Beam Search as Implicit Step-Level Search
Beam search, which maintains a set of most promising partial sequences at each generation step, functions as a form of step-level search. In this framework, the language model's own probability distribution implicitly serves as the verifier, and the pruning mechanism used to maintain the beam size acts as the search strategy. This differs from explicit step-level search, which utilizes more sophisticated and specialized verifier components.
0
1
Tags
Foundations of Large Language Models
Ch.5 Inference - Foundations of Large Language Models
Foundations of Large Language Models Course
Computing Sciences
Related
An AI system is designed to solve multi-step logic puzzles. Two generation strategies are being considered. Strategy 1 generates the next step in the reasoning process by selecting the sequence of words that the model calculates as most probable. Strategy 2 generates a potential next step and then uses a separate, specialized model to judge whether that step is logically valid before committing to it. Which statement best analyzes the primary advantage of Strategy 2's evaluation method over Strategy 1's?
Analyzing AI Reasoning Strategies
Evaluating Reasoning Steps in Generative Models
Beam Search as Implicit Step-Level Search