cross-encoder-ettin-17m-DistillRankNET

This model is a cross-encoder based on jhu-clsp/ettin-encoder-17m. 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-17m")
model = AutoModelForSequenceClassification.from_pretrained("xpmir/cross-encoder-ettin-17m-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	23.94	28.98
trec2019	80.25	53.28
trec2020	86.96	53.33
fever	62.54	64.01
arguana	8.17	12.20
climate_fever	14.74	10.74
dbpedia	56.75	30.47
fiqa	25.77	20.02
hotpotqa	54.49	38.89
nfcorpus	44.88	24.70
nq	30.91	35.10
quora	72.67	73.54
scidocs	14.23	7.60
scifact	44.12	46.84
touche	57.82	30.06
trec_covid	78.54	57.16
robust04	51.90	30.83
lotte_writing	51.33	41.45
lotte_recreation	43.58	38.96
lotte_science	33.19	27.43
lotte_technology	32.42	25.31
lotte_lifestyle	53.95	45.13
Mean In Domain	63.72	45.20
BEIR 13	43.51	34.72
LoTTE (OOD)	44.39	34.85