rerank abstract method
Scores complete query/document pairs in input order, preserving all classes.
Preflight the entire batch before inference, never truncate or return partial results, and use LlamaBackend.cancelGeneration for cancellation. Unsupported combinations throw LlamaUnsupportedException; overflow, cancellation and execution failures throw LlamaInferenceException.
Implementation
Future<BackendRerankingOutput> rerank(
int contextHandle,
String query,
List<String> documents,
);