← All models · MS MARCO reranker
use case · wild
The hard test for relevance: bury the one genuinely-correct answer among distractors that share lots of words with the query but don't actually answer it. Keyword search falls for the traps; a cross-encoder reads for meaning and pulls out the real answer. Edit the candidates and try to fool it.
The margin — the gap between the top logit and the runner-up — is a rough confidence read. A big margin means the reranker is sure; a thin one means the distractors are genuinely close and you may want a human in the loop.
← Practical · MS MARCO reranker · Next: Multi-model →