$ϕ$-Decoding: Adaptive Foresight Sampling for Balanced Inference-Time Exploration and Exploitation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Xu, Fangzhi, Yan, Hang, Ma, Chang, Zhao, Haiteng, Liu, Jun, Lin, Qika, Wu, Zhiyong
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910879011831808
author Xu, Fangzhi
Yan, Hang
Ma, Chang
Zhao, Haiteng
Liu, Jun
Lin, Qika
Wu, Zhiyong
author_facet Xu, Fangzhi
Yan, Hang
Ma, Chang
Zhao, Haiteng
Liu, Jun
Lin, Qika
Wu, Zhiyong
contents Inference-time optimization scales computation to derive deliberate reasoning steps for effective performance. While previous search-based strategies address the short-sightedness of auto-regressive generation, the vast search space leads to excessive exploration and insufficient exploitation. To strike an efficient balance to derive the optimal step, we frame the decoding strategy as foresight sampling, leveraging simulated future steps to obtain globally optimal step estimation. Built on it, we propose a novel decoding strategy, named $ϕ$-Decoding. To provide a precise and expressive estimation of step value, $ϕ$-Decoding approximates two distributions via foresight and clustering. Sampling from the joint distribution, the optimal steps can be selected for exploitation. To support adaptive computation allocation, we propose in-width and in-depth pruning strategies, featuring a light-weight solution to achieve inference efficiency. Extensive experiments across seven benchmarks show $ϕ$-Decoding outperforms strong baselines in both performance and efficiency. Additional analysis demonstrates its generalization across various LLMs and scalability across a wide range of computing budgets. The code will be released at https://github.com/xufangzhi/phi-Decoding, and the open-source PyPI package is coming soon.
format Preprint
id arxiv_https___arxiv_org_abs_2503_13288
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle $ϕ$-Decoding: Adaptive Foresight Sampling for Balanced Inference-Time Exploration and Exploitation
Xu, Fangzhi
Yan, Hang
Ma, Chang
Zhao, Haiteng
Liu, Jun
Lin, Qika
Wu, Zhiyong
Machine Learning
Artificial Intelligence
Computation and Language
Inference-time optimization scales computation to derive deliberate reasoning steps for effective performance. While previous search-based strategies address the short-sightedness of auto-regressive generation, the vast search space leads to excessive exploration and insufficient exploitation. To strike an efficient balance to derive the optimal step, we frame the decoding strategy as foresight sampling, leveraging simulated future steps to obtain globally optimal step estimation. Built on it, we propose a novel decoding strategy, named $ϕ$-Decoding. To provide a precise and expressive estimation of step value, $ϕ$-Decoding approximates two distributions via foresight and clustering. Sampling from the joint distribution, the optimal steps can be selected for exploitation. To support adaptive computation allocation, we propose in-width and in-depth pruning strategies, featuring a light-weight solution to achieve inference efficiency. Extensive experiments across seven benchmarks show $ϕ$-Decoding outperforms strong baselines in both performance and efficiency. Additional analysis demonstrates its generalization across various LLMs and scalability across a wide range of computing budgets. The code will be released at https://github.com/xufangzhi/phi-Decoding, and the open-source PyPI package is coming soon.
title $ϕ$-Decoding: Adaptive Foresight Sampling for Balanced Inference-Time Exploration and Exploitation
topic Machine Learning
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2503.13288