You Only Forward Once: An Efficient Compositional Judging Paradigm
This addresses the problem of slow judgment generation in high-throughput settings for researchers and practitioners using MLLMs, offering an incremental improvement in efficiency.
The paper tackles the trade-off between fine-grained understanding and speed in multimodal large language models (MLLMs) used as judges by proposing YOFO, a method that judges all requirements in a single forward pass, achieving orders-of-magnitude speedups and state-of-the-art results on standard datasets.
Multimodal large language models (MLLMs) show strong potential as judges. However, existing approaches face a fundamental trade-off: adapting MLLMs to output a single score misaligns with the generative nature of MLLMs and limits fine-grained requirement understanding, whereas autoregressively generating judging analyses is prohibitively slow in high-throughput settings. Observing that judgment reduces to verifying whether inputs satisfy a set of structured requirements, we propose YOFO, a template-conditioned method that judges all requirements in a single forward pass. Built on an autoregressive model, YOFO accepts a structured requirement template and, in one inference step, produces a binary yes/no decision for each requirement by reading the logits of the final token associated with that requirement. This design yields orders-of-magnitude speedups while preserving interpretability. Extensive experiments show that YOFO not only achieves state-of-the-art results on standard recommendation datasets, but also supports dependency-aware analysis-where subsequent judgments are conditioned on previous ones-and further benefits from post-hoc CoT.