zhou-2023-roberta-nli-sts-b-retrieval
IN premise — summaries/2026/08/24/zhou-2023-context-faithful-prompting-s3-method.md
Created 2026-08-25T02:59:10+00:00
Counterfactual demonstration retrieval encodes the test instance and candidate counterfactual instances using the RoBERTa_nli+sts-b model and selects the top-k by cosine similarity via the KATE framework.
Summary
The system finds the most relevant counterfactual examples for a given test case by passing both the test instance and the candidates through a combined natural-language-inference and sentence-similarity encoder, then ranking them by how closely their embeddings align. This matters because the specific model choice and cosine-similarity ranking inside the KATE pipeline define exactly which counterfactuals downstream reasoning will draw on, making the retrieval step a bottleneck for overall system behavior.