cross-encoder-ettin-32m-infoNCE

This model is a cross-encoder based on jhu-clsp/ettin-encoder-32m. 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.

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 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("jhu-clsp/ettin-encoder-32m")
model = AutoModelForSequenceClassification.from_pretrained("xpmir/cross-encoder-ettin-32m-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	37.70	44.15
trec2019	96.90	71.56
trec2020	89.88	68.48
fever	76.76	77.15
arguana	15.73	23.75
climate_fever	23.08	17.20
dbpedia	68.54	40.44
fiqa	42.39	34.58
hotpotqa	86.68	70.40
nfcorpus	50.82	30.83
nq	47.61	52.60
quora	76.39	78.32
scidocs	26.05	14.46
scifact	66.63	68.91
touche	54.51	33.03
trec_covid	88.52	72.68
robust04	64.26	40.56
lotte_writing	69.82	60.37
lotte_recreation	57.36	52.06
lotte_science	46.68	38.95
lotte_technology	52.38	43.85
lotte_lifestyle	68.75	59.53
Mean In Domain	74.83	61.40
BEIR 13	55.67	47.26
LoTTE (OOD)	59.88	49.22