cross-encoder-bert-base-infoNCE

This model is a cross-encoder based on bert-base-uncased. It was trained on Ms-Marco using loss infoNCE as part of a reproducibility paper for training cross encoders: "Reproducing and Comparing Distillation Techniques for Cross-Encoders", see the paper for more details.

Model Description
Usage
Evals

Model Description

This model is intended for re-ranking the top results returned by a retrieval system (like BM25, Bi-Encoders or SPLADE).

Training Data: MS MARCO Passage
Language: English
Loss infoNCE

Training can be easily reproduced using the assiciated repository. The exact training configuration used for this model is also detailed in config.yaml.

Usage

Quick Start:

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
model = AutoModelForSequenceClassification.from_pretrained("xpmir/cross-encoder-bert-base-infoNCE")

features = tokenizer("What is experimaestro ?", "Experimaestro is a powerful framework for ML experiments management...", padding=True, truncation=True, return_tensors="pt")

model.eval()
with torch.no_grad():
    scores = model(**features).logits
    print(scores)

Evaluations

We provide evaluations of this cross-encoder re-ranking the top 1000 documents retrieved by naver/splade-v3-distilbert.

dataset	RR@10	nDCG@10
msmarco_dev	40.13	46.68
trec2019	98.26	75.65
trec2020	93.36	73.30
fever	81.43	81.33
arguana	23.01	34.22
climate_fever	31.31	23.24
dbpedia	78.14	45.69
fiqa	42.83	35.87
hotpotqa	89.63	73.49
nfcorpus	55.04	34.24
nq	54.25	59.13
quora	78.34	80.38
scidocs	26.07	15.06
scifact	69.26	71.47
touche	61.20	33.31
trec_covid	90.57	67.53
robust04	71.40	48.48
lotte_writing	68.55	58.87
lotte_recreation	59.75	54.52
lotte_science	43.67	36.39
lotte_technology	51.72	42.85
lotte_lifestyle	71.37	61.89
Mean In Domain	77.25	65.21
BEIR 13	60.08	50.38
LoTTE (OOD)	61.08	50.50