Prompt Chunking

A common strategy in token-level detection methods for handling long textual descriptions. The long description is divided into multiple short segments that are scored separately and then fused. Although this covers all tokens, it introduces additional inference overhead and may weaken compositional consistency.