This model is ANCE-Tele trained on MS MARCO. The training details and evaluation results are as follows:
|
Model
|
Pretrain Model
|
Train w/ Marco Title
|
Marco Dev MRR@10
|
BEIR Avg NDCG@10
|
|
ANCE-Tele
|
cocodr-base
|
w/o
|
37.3
|
44.2
|
|
BERI Dataset
|
NDCG@10
|
|
TREC-COVID
|
77.4
|
|
NFCorpus
|
34.4
|
|
FiQA
|
29.0
|
|
ArguAna
|
45.6
|
|
Touché-2020
|
22.3
|
|
Quora
|
85.8
|
|
SCIDOCS
|
14.6
|
|
SciFact
|
71.0
|
|
NQ
|
50.5
|
|
HotpotQA
|
58.8
|
|
Signal-1M
|
27.2
|
|
TREC-NEWS
|
34.7
|
|
DBPedia-entity
|
36.2
|
|
Fever
|
71.4
|
|
Climate-Fever
|
17.9
|
|
BioASQ
|
42.1
|
|
Robust04
|
41.4
|
|
CQADupStack
|
34.9
|
The implementation is the same as our EMNLP 2022 paper
"Reduce Catastrophic Forgetting of Dense Retrieval Training with Teleportation Negatives"
. The associated GitHub repository is available at
https://github.com/OpenMatch/ANCE-Tele
.
@inproceedings{sun2022ancetele,
title={Reduce Catastrophic Forgetting of Dense Retrieval Training with Teleportation Negatives},
author={Si, Sun and Chenyan, Xiong and Yue, Yu and Arnold, Overwijk and Zhiyuan, Liu and Jie, Bao},
booktitle={Proceedings of EMNLP 2022},
year={2022}
}