|
--- |
|
license: mit |
|
datasets: |
|
- ms_marco |
|
language: |
|
- en |
|
pipeline_tag: feature-extraction |
|
--- |
|
|
|
# MASTER: Multi-task Pre-trained Bottlenecked Masked Autoencoders are Better Dense Retrievers |
|
|
|
Paper: [https://arxiv.org/abs/2212.07841](https://arxiv.org/abs/2212.07841). |
|
|
|
Code: [https://github.com/microsoft/SimXNS/tree/main/MASTER](https://github.com/microsoft/SimXNS/tree/main/MASTER). |
|
|
|
## Overview |
|
|
|
This is the checkpoint after pretraining on the MS-MARCO corpus. **You may use this checkpoint as the initialization for finetuning.** |
|
|
|
## Useage |
|
|
|
|
|
To load this checkpoint for initialization, you may follow: |
|
```python |
|
from transformers import AutoModel |
|
|
|
model = AutoModel.from_pretrained('lx865712528/master-base-pretrained-msmarco') |
|
``` |
|
|
