Cut Your Losses! Learning to Prune Paths Early for Efficient Parallel Reasoning

Hugging Face Daily Papers Papers
efficient-reasoning token-pruning parallel-decoding kv-cache inference-optimization

Summary

This paper introduces STOP (Super Token for Pruning), a lightweight method that learns to prune unpromising reasoning paths early during parallel decoding by appending learnable tokens and reading KV cache states, achieving 70% token reduction while improving performance on AIME and GPQA benchmarks.

Parallel reasoning enhances Large Reasoning Models (LRMs) but incurs prohibitive costs due to futile paths caused by early errors. To mitigate this, path pruning at the prefix level is essential, yet existing research remains fragmented without a standardized framework. In this work, we propose the first systematic taxonomy of path pruning, categorizing methods by their signal source (internal vs. external) and learnability (learnable vs. non-learnable). This classification reveals the unexplored potential of learnable internal methods, motivating our proposal of STOP (Super TOken for Pruning). Extensive evaluations across LRMs ranging from 1.5B to 20B parameters demonstrate that STOP achieves superior effectiveness and efficiency compared to existing baselines. Furthermore, we rigorously validate the scalability of STOP under varying compute budgets - for instance, boosting GPT-OSS-20B accuracy on AIME25 from 84% to nearly 90% under fixed compute budgets. Finally, we distill our findings into formalized empirical guidelines to facilitate optimal real-world deployment. Code, data and models are available at https://bijiaxihh.github.io/STOP
Original Article
View Cached Full Text

Cached at: 04/20/26, 08:27 AM

Paper page - Cut Your Losses! Learning to Prune Paths Early for Efficient Parallel Reasoning

Source: https://huggingface.co/papers/2604.16029 STOP! You might already be on the wrong reasoning path.

In parallel reasoning, many sampled trajectories are already doomed from their early prefixes, yet still consume full decoding budgets.

We propose STOP (Super Token for Pruning), a lightweight method that appends a short sequence of learnable [STOP] tokens and directly reads KV cache states to decide whether a trajectory should be continued. This enables early pruning of unpromising paths without re-encoding or external models.

STOP significantly improves reasoning performance on AIME and GPQA, while reducing token usage by over 70% in many settings.

Similar Articles

Cut Your Losses! Learning to Prune Paths Early for Efficient Parallel Reasoning

arXiv cs.CL

This paper proposes STOP (SuperTOken for Pruning), a systematic framework for pruning inefficient reasoning paths early in parallel reasoning with Large Reasoning Models. The method achieves superior efficiency and effectiveness across models from 1.5B to 20B parameters, boosting GPT-OSS-20B accuracy on AIME25 from 84% to 90% under fixed compute budgets.

SWE-Pruner Pro: The Coder LLM Already Knows What to Prune

Hugging Face Daily Papers

SWE-Pruner Pro leverages the coding agent's own internal representations to prune long code context, saving up to 39% of tokens while maintaining or improving task performance on multi-turn benchmarks.