cross-encoder-ettin-150m-ADR-MSE

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

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-ADR-MSE")

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	36.25	42.86
trec2019	96.90	75.17
trec2020	92.65	73.81
fever	81.31	80.92
arguana	13.60	20.95
climate_fever	26.48	19.29
dbpedia	76.04	45.85
fiqa	47.31	39.27
hotpotqa	85.30	66.63
nfcorpus	57.15	34.99
nq	53.73	58.57
quora	78.36	80.10
scidocs	28.71	15.95
scifact	67.17	70.12
touche	67.36	36.33
trec_covid	94.57	77.18
robust04	74.06	50.17
lotte_writing	73.44	64.64
lotte_recreation	60.66	55.80
lotte_science	50.58	42.32
lotte_technology	55.38	46.37
lotte_lifestyle	72.58	63.07
Mean In Domain	75.27	63.95
BEIR 13	59.78	49.70
LoTTE (OOD)	64.45	53.73