Cross-Encoder for Text Ranking
This model is a port of the webis/monoelectra-base model from lightning-ir to Sentence Transformers and Transformers.
The original model was introduced in the paper A Systematic Investigation of Distilling Large Language Models into Cross-Encoders for Passage Re-ranking. See https://github.com/webis-de/rank-distillm for code used to train the original model.
The model can be used as a reranker in a 2-stage "retrieve-rerank" pipeline, where it reorders passages returned by a retriever model (e.g. an embedding model or BM25) given some query. See SBERT.net Retrieve & Re-rank for more details.
Usage with Sentence Transformers
The usage is easy when you have SentenceTransformers installed.
pip install sentence-transformers
Then you can use the pre-trained model like this:
from sentence_transformers import CrossEncoder
model = CrossEncoder("cross-encoder/monoelectra-base", trust_remote_code=True)
scores = model.predict([
("How many people live in Berlin?", "Berlin had a population of 3,520,031 registered inhabitants in an area of 891.82 square kilometers."),
("How many people live in Berlin?", "Berlin is well known for its museums."),
])
print(scores)
# [ 8.122868 -4.292924]
Usage with Transformers
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch
model = AutoModelForSequenceClassification.from_pretrained("cross-encoder/monoelectra-base", trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained("cross-encoder/monoelectra-base")
features = tokenizer(
[
("How many people live in Berlin?", "Berlin had a population of 3,520,031 registered inhabitants in an area of 891.82 square kilometers."),
("How many people live in Berlin?", "Berlin is well known for its museums."),
],
padding=True,
truncation=True,
return_tensors="pt",
)
model.eval()
with torch.no_grad():
scores = model(**features).logits.view(-1)
print(scores)
# tensor([ 8.1229, -4.2929])