Chopping Trees: Semantic Similarity Based Dynamic Pruning for Tree-of-Thought Reasoning

Joongho Kim, Xirui Huang, Zarreen Reza, Gabriel Grand, Kevin Zhu, Ryan Lagasse · 2025

Read Chopping Trees: Semantic Similarity Based Dynamic Pruning for Tree-of-Thought Reasoning by Joongho Kim, Xirui Huang, Zarreen Reza, Gabriel Grand, Kevin Zhu, Ryan Lagasse (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-08595.txt.

Description

Joongho Kim and three co-authors speed up Tree-of-Thought reasoning, a method where a language model explores many branching solution paths. The search wastes effort because separate branches often repeat the same idea in different words, a redundancy the authors call semantic overlap.

Their method, Semantic Similarity-Based Dynamic Pruning, watches the search as it runs, groups steps that mean nearly the same thing, and prunes the duplicates in real time. On the GSM8K and MATH500 benchmarks it runs up to two point three times faster than strong tree-search baselines. Accuracy stays within about five percent of the best baseline while the number of explored nodes falls by eighty-five to ninety percent. Code is public.

Catalog record

CREATORJoongho Kim, Xirui Huang, Zarreen Reza, Gabriel Grand, Kevin Zhu, Ryan Lagasse
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE26 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-08595

Browse the full library · About the Tower of Babel