$φ$-Decoding: Adaptive Foresight Sampling for Balanced Inference-Time Exploration and Exploitation
This work addresses a key bottleneck in inference-time optimization for large language models, offering an incremental improvement over existing search-based strategies.
The paper tackles the problem of balancing exploration and exploitation in inference-time optimization for language models by proposing $\phi$-Decoding, a novel decoding strategy that uses foresight sampling and clustering to select optimal steps, achieving improved performance and efficiency across seven benchmarks.
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.