A model architecture that takes a query and document together as input and directly outputs a relevance score, unlike dual-encoders that score them separately.
Adhering to complex, structured, or constrained instructions