Verdict-MQ Enables Multi-Question Inference via Single-Pass Encoding
October 16, 2026
Verdict-MQ uses a structured attention mask to allow an encoder to answer multiple typed questions about a single state in one pass. The 395M parameter model outperforms the 421M Laya specialist on typed-decisions benchmarks by achieving lower KL divergence and Brier scores.
HOW THIS AFFECTS YOU
●
builderYou can reduce compute costs and latency when processing high-volume, small-scale classification tasks on single documents.
●
researcherThis provides a method for making state encoding independent of question order and influence.