yaoxuewen/

bert-small-cross-encoder

Student model of nreimers/mmarco-mMiniLMv2-L12-H384-v1

Trained on prajjwal1/bert-small

Dataset: MSMarco (https://microsoft.github.io/msmarco/TREC-Deep-Learning) -> passv2_train_top100.txt -> first 2527 queries with their corresponding 100 passages

Pass each query and passage pair through nreimers/mmarco-mMiniLMv2-L12-H384-v1 and get a score. Train prajjwal1/bert-small to match the score based on the same query/passage pair.

The train code is similar to: https://github.com/UKPLab/sentence-transformers/tree/master/examples/training/cross-encoder

Pearson correlation: 0.83976, Spearman correlation: 0.88378 on the test dataset

Mean rbo score: 0.84 for passage ranking