cross-encoder-ettin-150m-MarginMSE

This model is a cross-encoder based on jhu-clsp/ettin-encoder-150m. It was trained on Ms-Marco using loss marginMSE as part of a reproducibility paper for training cross encoders: "Reproducing and Comparing Distillation Techniques for Cross-Encoders", see the paper for more details.

Contents

• 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 marginMSE

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("xpmir/cross-encoder-ettin-150m-MarginMSE")
model = AutoModelForSequenceClassification.from_pretrained("xpmir/cross-encoder-ettin-150m-MarginMSE")

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	40.64	47.28
trec2019	95.23	73.77
trec2020	94.75	73.89
fever	83.48	82.97
arguana	24.04	35.30
climate_fever	36.86	27.26
dbpedia	78.60	48.26
fiqa	48.22	40.32
hotpotqa	90.50	74.92
nfcorpus	56.81	35.59
nq	55.57	60.48
quora	80.97	82.83
scidocs	29.38	16.70
scifact	70.00	73.05
touche	59.89	34.80
trec_covid	93.83	75.79
robust04	70.15	48.62
lotte_writing	74.20	64.82
lotte_recreation	62.97	57.69
lotte_science	49.32	41.38
lotte_technology	57.70	49.28
lotte_lifestyle	75.63	65.67
Mean In Domain	76.87	64.98
BEIR 13	62.17	52.94
LoTTE (OOD)	65.00	54.58