ProcessBench: Identifying Process Errors in Mathematical Reasoning
This work addresses scalable oversight of language models in mathematical reasoning, but it is incremental as it builds on existing datasets and methods.
The paper tackles the problem of language models making errors in mathematical reasoning by introducing ProcessBench, a dataset of 3,400 test cases with annotated error locations, and finds that existing process reward models fail to generalize to challenging problems, while critic models and fine-tuned PRMs perform better, with QwQ-32B-Preview showing critique capability competitive with GPT-4o.
As language models regularly make mistakes when solving math problems, automated identification of errors in the reasoning process becomes increasingly significant for their scalable oversight. In this paper, we introduce ProcessBench for measuring the ability to identify erroneous steps in mathematical reasoning. It consists of 3,400 test cases, primarily focused on competition- and Olympiad-level math problems. Each test case contains a step-by-step solution with error location annotated by human experts. Models are required to identify the earliest step that contains an error, or conclude that all steps are correct. We conduct extensive evaluation on ProcessBench, involving two types of models: process reward models (PRMs) and critic models, where for the latter we prompt general language models to critique each solution step by step. We draw two main observations: (1) Existing PRMs typically fail to generalize to more challenging math problems beyond GSM8K and MATH. They underperform both critic models (i.e., prompted general language models) and our own trained PRM that is straightforwardly fine-tuned on the PRM800K dataset. (2) The best open-source model, QwQ-32B-Preview, has demonstrated the critique capability competitive with the proprietary model GPT-4o, despite that it still lags behind the reasoning-specialized o1-mini. We hope ProcessBench can foster future research in reasoning process assessment, paving the way toward scalable oversight of language models.