Tree-of-Thought

Quick Definition

Framework exploring multiple reasoning paths for optimal solutions

A framework exploring multiple reasoning paths simultaneously using search algorithms.

complex problem solving, strategic planning, game playing

chain-of-thought reasoning

Training smaller LLMs to replicate larger model behavior

Using draft model candidates verified by large model

Generating LLM responses in predefined machine-readable formats

Generation using cumulative probability threshold for token selection

Maximum tokens a language model can process at once

Alignment method bypassing reward model for preference learning