Large language models (LLMs) excel at function calling, but inference scaling has been explored mainly for unstructured generation. We propose an inference-scaling framework for structured outputs that combines fine-grained beam search with \textbf{ToolPRM}, a process reward model scoring each intra-call decision (function name and argument filling). We build the first fine-grained intra-call supervision dataset via function masking, rollout collection, and step-level annotation. ToolPRM outperforms outcome and coarse-grained reward models in predictive accuracy and yields consistent test-time gains on multiple function-calling benchmarks. We further show that structured generation follows ``\textbf{explore more but retain less}'', since early JSON errors are unrecoverable.
@article{arxiv.2510.14703,
title = {ToolPRM: Fine-Grained Inference Scaling of Structured Outputs for Function Calling},
author = {Jianghao Lin and Yuanyuan Shi and Xin Peng and Renjie Ding and Hairui Wang and Yuxuan Peng and Bizhe Bai and Weixi Song and Fengshuo Bai and Huacan Chai and Weinan Zhang and Fei Huang and Ying Wen},
journal= {arXiv preprint arXiv:2510.14703},
year = {2026}
}