File size: 805 Bytes
1fd87b5 327241e 1fd87b5 327241e c24508a |
1 2 3 4 5 6 7 8 9 10 |
This model is a T5-large reranker fine-tuned on the MS MARCO passage dataset for 10k steps (or 1 epoch).
This model usually has a better zero-shot performance than `monot5-large-msmarco`, i.e., it performs better on datasets different from MS MARCO.
For more details on how to use it, check the following links:
- [A simple reranking example](https://github.com/castorini/pygaggle#a-simple-reranking-example)
- [Rerank MS MARCO passages](https://github.com/castorini/pygaggle/blob/master/docs/experiments-msmarco-passage-subset.md)
- [Rerank Robust04 documents](https://github.com/castorini/pygaggle/blob/master/docs/experiments-robust04-monot5-gpu.md)
Paper describing the model: [Document Ranking with a Pretrained Sequence-to-Sequence Model](https://www.aclweb.org/anthology/2020.findings-emnlp.63/) |