Base model: cross-encoder/ms-marco-MiniLM-L-6-v2
Training: 3 epochs on in-domain CVE/CWE pairs (hard negatives from Phase 24 fusion ranker)
Test Hits@10: 0.7165
Test MRR: 0.6026
Fusion: rerank top-200 CWE candidates with cross-encoder weight λ=0.5

Usage

python

from sentence_transformers import CrossEncoder

model = CrossEncoder("alihur-ecu/cve-cwe-crossencoder")
scores = model.predict([("CVE description text", "CWE description text")])