cross-encoder-ettin-150m-DistillRankNET

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

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-DistillRankNET")

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	36.30	42.93
trec2019	96.98	75.59
trec2020	93.83	72.27
fever	80.10	79.82
arguana	14.52	22.21
climate_fever	27.00	19.76
dbpedia	75.55	45.75
fiqa	47.54	39.63
hotpotqa	85.28	66.73
nfcorpus	57.92	35.41
nq	53.64	58.68
quora	75.29	77.46
scidocs	28.04	15.73
scifact	68.06	70.51
touche	66.31	36.81
trec_covid	96.50	77.81
robust04	73.96	49.37
lotte_writing	73.65	64.31
lotte_recreation	62.02	56.42
lotte_science	51.11	42.43
lotte_technology	56.74	47.53
lotte_lifestyle	72.68	63.49
Mean In Domain	75.70	63.60
BEIR 13	59.67	49.72
LoTTE (OOD)	65.03	53.92