Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,62 @@
|
|
1 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
license: apache-2.0
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
+
tags:
|
3 |
+
- MRC
|
4 |
+
- SQuAD 1.1
|
5 |
+
- roberta-large
|
6 |
+
language:
|
7 |
+
- English
|
8 |
license: apache-2.0
|
9 |
---
|
10 |
+
|
11 |
+
# Model description
|
12 |
+
|
13 |
+
An RoBERTa reading comprehension model for [SQuAD 1.1](https://aclanthology.org/D16-1264/).
|
14 |
+
|
15 |
+
The model is initialized with [roberta-large](https://huggingface.co/roberta-large/) and fine-tuned on the [SQuAD 1.1 train data](https://huggingface.co/datasets/squad).
|
16 |
+
|
17 |
+
## Intended uses & limitations
|
18 |
+
|
19 |
+
You can use the raw model for the reading comprehension task. Biases associated with the pre-existing language model, roberta-large, that we used may be present in our fine-tuned model, squad-v1-roberta-large.
|
20 |
+
|
21 |
+
|
22 |
+
## Usage
|
23 |
+
|
24 |
+
You can use this model directly with the [PrimeQA](https://github.com/primeqa/primeqa) pipeline for reading comprehension [squad.ipynb](https://github.com/primeqa/primeqa/blob/main/notebooks/mrc/squad.ipynb).
|
25 |
+
|
26 |
+
```bibtex
|
27 |
+
@article{2016arXiv160605250R,
|
28 |
+
author = {{Rajpurkar}, Pranav and {Zhang}, Jian and {Lopyrev},
|
29 |
+
Konstantin and {Liang}, Percy},
|
30 |
+
title = "{SQuAD: 100,000+ Questions for Machine Comprehension of Text}",
|
31 |
+
journal = {arXiv e-prints},
|
32 |
+
year = 2016,
|
33 |
+
eid = {arXiv:1606.05250},
|
34 |
+
pages = {arXiv:1606.05250},
|
35 |
+
archivePrefix = {arXiv},
|
36 |
+
eprint = {1606.05250},
|
37 |
+
}
|
38 |
+
```
|
39 |
+
```bibtex
|
40 |
+
@article{DBLP:journals/corr/abs-1907-11692,
|
41 |
+
author = {Yinhan Liu and
|
42 |
+
Myle Ott and
|
43 |
+
Naman Goyal and
|
44 |
+
Jingfei Du and
|
45 |
+
Mandar Joshi and
|
46 |
+
Danqi Chen and
|
47 |
+
Omer Levy and
|
48 |
+
Mike Lewis and
|
49 |
+
Luke Zettlemoyer and
|
50 |
+
Veselin Stoyanov},
|
51 |
+
title = {RoBERTa: {A} Robustly Optimized {BERT} Pretraining Approach},
|
52 |
+
journal = {CoRR},
|
53 |
+
volume = {abs/1907.11692},
|
54 |
+
year = {2019},
|
55 |
+
url = {http://arxiv.org/abs/1907.11692},
|
56 |
+
archivePrefix = {arXiv},
|
57 |
+
eprint = {1907.11692},
|
58 |
+
timestamp = {Thu, 01 Aug 2019 08:59:33 +0200},
|
59 |
+
biburl = {https://dblp.org/rec/journals/corr/abs-1907-11692.bib},
|
60 |
+
bibsource = {dblp computer science bibliography, https://dblp.org}
|
61 |
+
}
|
62 |
+
```
|