monoT5 trained on MS-Marco

Implementation of

Nogueira, R., Jiang, Z., Lin, J., 2020. Document Ranking with a Pretrained Sequence-to-Sequence Model. arXiv:2003.06713 [cs].

This model has been trained on MsMarco v1, and uses the t5-base model

Parameters based on PyGaggle

Using the model

The model can be loaded with experimaestro IR

If you want to use the model in further experiments with XPMIR, use this code:

from xpmir.models import AutoModel
from xpmir.models import AutoModel

model, init_tasks = AutoModel.load_from_hf_hub("xpmir/monot5")

Use this code if you want to use the model in inference only:

from xpmir.models import AutoModel
from xpmir.models import AutoModel

model = AutoModel.load_from_hf_hub("xpmir/monot5", as_instance=True)
model.rsv("walgreens store sales average", "The average Walgreens salary ranges...")

Results

Dataset	AP	P@20	RR	RR@10	Success@5	nDCG	nDCG@10	nDCG@20
msmarco_dev	0.3797	0.0384	0.3851	0.3762	0.5497	0.4835	0.4382	0.4602
trec2019	0.4874	0.7209	0.9671	0.9671	1.0000	0.6918	0.7217	0.6939
trec2020	0.4605	0.6139	0.9396	0.9389	0.9815	0.6796	0.6925	0.6581