Abstract
Step-level reasoning evaluators are commonly based on autoregressive language models, whose causal attention restricts each step representation to the problem, previous steps, and the current step. Yet, when the complete solution is available, the validity of an earlier step may become clearer only through its downstream consequences. We validate this hypothesis through a controlled 54-run comparison of causal and bidirectional LLaDA evaluators at 1B--3B scale, changing only the self-attention mask, and find bidirectional attention yields consistent improvements. Building on this finding, we introduce \prm{}, an 8B bidirectional evaluator that reaches 88.8 step-level F1 on MR-MATH-invalid and 83.8 on the out-of-distribution MR-GSM8K original-question subset, outperforming ReasonEval-Llemma-34B by 11.3 and 10.3 F1 points, respectively. \prm{} also remains effective when evaluating incomplete reasoning traces in online settings, outperforming the strongest baselines on both benchmarks by a large margin. We further show that \prm{} provides an effective training-data selection signal, improving Mistral-7B performance on MATH-500.
Keywords
Subject
Publication details
- Journal
- Not available
- Open access
- Green open access
Cite this article
APA 7
Feng, Y., Deng, N., Chen, Y., & Mihalcea, R. (2026). LLaDA-PRM: A Bidirectional Step-Level Reasoning Evaluator. https://omanscience.com/en/articles/llada-prm-a-bidirectional-step-level-reasoning-evaluator
MLA 9
Feng, Yiming, et al. "LLaDA-PRM: A Bidirectional Step-Level Reasoning Evaluator." https://omanscience.com/en/articles/llada-prm-a-bidirectional-step-level-reasoning-evaluator.
Chicago (author–date)
Feng, Yiming, Naihao Deng, Yulong Chen, and Rada Mihalcea. 2026. "LLaDA-PRM: A Bidirectional Step-Level Reasoning Evaluator." https://omanscience.com/en/articles/llada-prm-a-bidirectional-step-level-reasoning-evaluator.
Harvard
Feng, Y., Deng, N., Chen, Y. and Mihalcea, R. (2026) 'LLaDA-PRM: A Bidirectional Step-Level Reasoning Evaluator', Available at: https://omanscience.com/en/articles/llada-prm-a-bidirectional-step-level-reasoning-evaluator.
Vancouver
Feng Y, Deng N, Chen Y, Mihalcea R. LLaDA-PRM: A Bidirectional Step-Level Reasoning Evaluator. https://omanscience.com/en/articles/llada-prm-a-bidirectional-step-level-reasoning-evaluator
IEEE
Y. Feng, N. Deng, Y. Chen, and R. Mihalcea, "LLaDA-PRM: A Bidirectional Step-Level Reasoning Evaluator," https://omanscience.com/en/articles/llada-prm-a-bidirectional-step-level-reasoning-evaluator.