cross-encoder-MiniLM-L12-infoNCE

This model is a cross-encoder based on microsoft/MiniLM-L12-H384-uncased. 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("microsoft/MiniLM-L12-H384-uncased")
model = AutoModelForSequenceClassification.from_pretrained("xpmir/cross-encoder-MiniLM-L12-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	40.11	46.78
trec2019	97.67	75.70
trec2020	95.06	73.47
fever	81.53	81.33
arguana	22.23	32.86
climate_fever	30.43	22.10
dbpedia	76.22	44.88
fiqa	45.92	38.18
hotpotqa	88.89	72.07
nfcorpus	55.54	33.83
nq	53.58	58.73
quora	73.37	76.46
scidocs	28.96	16.35
scifact	68.27	71.04
touche	62.50	32.81
trec_covid	93.45	71.70
robust04	73.21	49.57
lotte_writing	68.25	59.26
lotte_recreation	61.74	56.11
lotte_science	44.50	37.01
lotte_technology	54.43	45.65
lotte_lifestyle	72.64	63.40
Mean In Domain	77.61	65.32
BEIR 13	60.07	50.18
LoTTE (OOD)	62.46	51.83