DeepPrune: Parallel Scaling without Inter-trace Redundancy

2025-10-09

AI Summary

Parallel chain-of-thought (CoT) inference in large language models (LLMs) suffers from severe computational inefficiency due to high inter-path redundancy (>80% of reasoning trajectories yield identical answers). To address this, we propose DeepPrune, the first dynamic pruning framework that jointly integrates discriminative equivalence prediction with online greedy clustering. A discriminative model—trained with focal loss and oversampling—identifies redundant reasoning paths in real time; subsequently, equivalent branches are pruned dynamically under an answer-diversity constraint. Evaluated across multiple benchmarks and mainstream LLMs, DeepPrune achieves over 80% token savings while incurring ≤3 percentage points accuracy degradation. This substantially improves parallel CoT inference efficiency and establishes a novel paradigm for resource-efficient reasoning.

Technology Category

• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Parallel scaling has emerged as a powerful paradigm to enhance reasoning capabilities in large language models (LLMs) by generating multiple Chain-of-Thought (CoT) traces simultaneously. However, this approach introduces significant computational inefficiency due to inter-trace redundancy -- our analysis reveals that over 80% of parallel reasoning traces yield identical final answers, representing substantial wasted computation. To address this critical efficiency bottleneck, we propose DeepPrune, a novel framework that enables efficient parallel scaling through dynamic pruning. Our method features a specialized judge model trained with focal loss and oversampling techniques to accurately predict answer equivalence from partial reasoning traces which realizes 0.87 AUROC on equivalence prediction, combined with an online greedy clustering algorithm that dynamically prunes redundant paths while preserving answer diversity. Comprehensive evaluations across three challenging benchmarks (AIME 2024, AIME 2025, and GPQA) and multiple reasoning models demonstrate that DeepPrune achieves remarkable token reduction by over 80% compared to conventional consensus sampling on most cases, while maintaining competitive accuracy within 3 percentage points. Our work establishes a new standard for efficient parallel reasoning, making high-performance reasoning more efficient. Our code and data are here: https://deepprune.github.io/

Problem

• Reduces computational redundancy in parallel reasoning traces
• Prunes redundant paths while preserving answer diversity
• Maintains accuracy with over 80% token reduction

Innovation

• Dynamic pruning framework for parallel scaling
• Judge model predicts answer equivalence early
• Greedy clustering algorithm preserves answer diversity