cross-encoder-MiniLM-L12-BCE

This model is a cross-encoder based on microsoft/MiniLM-L12-H384-uncased. It was trained on Ms-Marco using loss bce 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 bce

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-BCE")

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	37.86	44.20
trec2019	98.06	68.86
trec2020	91.51	69.35
fever	74.84	75.68
arguana	22.98	33.77
climate_fever	27.01	19.69
dbpedia	66.92	39.41
fiqa	43.39	35.12
hotpotqa	84.86	68.52
nfcorpus	51.70	31.16
nq	49.36	54.80
quora	61.96	66.04
scidocs	25.52	14.31
scifact	64.86	68.10
touche	58.12	31.28
trec_covid	82.41	59.39
robust04	67.67	44.71
lotte_writing	63.17	54.71
lotte_recreation	58.43	52.92
lotte_science	41.01	33.83
lotte_technology	49.53	41.23
lotte_lifestyle	70.19	60.76
Mean In Domain	75.81	60.80
BEIR 13	54.92	45.94
LoTTE (OOD)	58.33	48.03