cross-encoder-ettin-150m-Hinge

This model is a cross-encoder based on jhu-clsp/ettin-encoder-150m. It was trained on Ms-Marco using loss hingeLoss 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 hingeLoss

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

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	38.26	44.88
trec2019	96.98	75.09
trec2020	91.98	69.89
fever	81.90	81.41
arguana	16.83	25.15
climate_fever	27.62	20.27
dbpedia	72.22	42.88
fiqa	48.39	40.23
hotpotqa	84.55	67.46
nfcorpus	55.66	34.96
nq	52.09	56.95
quora	65.37	68.50
scidocs	28.41	16.25
scifact	68.82	71.82
touche	64.70	35.97
trec_covid	92.33	78.15
robust04	68.72	45.20
lotte_writing	74.43	65.16
lotte_recreation	63.23	58.08
lotte_science	50.55	41.48
lotte_technology	58.66	49.55
lotte_lifestyle	73.19	64.08
Mean In Domain	75.74	63.29
BEIR 13	58.38	49.23
LoTTE (OOD)	64.80	53.93