Source: Notebook experiment 17-tree-of-thoughts (Tree of Thoughts)
The owner's runnable LangChain experiment demonstrating Tree of Thoughts.
Contents
- What: runnable experiment (Python, LangChain LCEL, LangSmith tracing) in the owner’s Notebook repo
- Author: Loic
- Date: accessed 2026-11-14
- Original: experiments/17-tree-of-thoughts
Summary
Branch into candidate reasoning steps, score them, expand the best: search over a tree rather than one chain. The experiment is a simplified beam search: at each depth a proposer model suggests K next steps, a scorer model rates them, the best are kept for DEPTH levels, then the model answers.
Key claims
- When to use: Problems needing exploration or backtracking (puzzles, planning, constraint satisfaction) where a single chain gets stuck.