LLM Learning Hub

workspace/llm-course/home

Search

Search at reasoning time explores multiple solution paths (tree-of-thoughts, beam search, MCTS) and selects the most promising branch, improving accuracy on combinatorial problems.

What is Search?

Search at reasoning time explores multiple solution paths (tree-of-thoughts, beam search, MCTS) and selects the most promising branch, improving accuracy on combinatorial problems.

Search at reasoning time explores multiple solution paths (tree-of-thoughts, beam search, MCTS) and selects the most promising branch, improving accuracy on combinatorial problems.

Where is it used?

Tree-of-Thoughts (ToT), AlphaCode's beam search, and rStar-Math's MCTS explore reasoning branches; LangGraph `ToolNode` with backtracking implements search in agent frameworks.

How to build it

Implement a ToT loop: generate K candidate next-steps, score each with an LLM evaluator, expand the top branch, and backtrack on dead ends using `langgraph` state graphs.