Commit
·
d0551f1
1
Parent(s):
1543302
Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,41 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Question-Answer to Statement Converter
|
2 |
+
|
3 |
+
A question answer pair to statement converter from https://github.com/jifan-chen/QA-Verification-Via-NLI
|
4 |
+
|
5 |
+
See:
|
6 |
+
```
|
7 |
+
@article{chen2021can,
|
8 |
+
title={Can NLI Models Verify QA Systems' Predictions?},
|
9 |
+
author={Chen, Jifan and Choi, Eunsol and Durrett, Greg},
|
10 |
+
journal={EMNLP Findings},
|
11 |
+
year={2021}
|
12 |
+
}
|
13 |
+
```
|
14 |
+
|
15 |
+
**Note:** I am not the maintainer or orginal author just keeping it here to use huggingface APIs to produce statements from question answer pair for downstream applications.
|
16 |
+
|
17 |
+
## TL;DR:
|
18 |
+
We fine-tune a seq2seq model,
|
19 |
+
T5-3B (Raffel et al., 2020), using the $(a, q, d)$ pairs
|
20 |
+
annotated by Demszky et al. (2018).
|
21 |
+
|
22 |
+
Where a is answer, q is question, and d is declerative sentence (i.e. a statement).
|
23 |
+
|
24 |
+
See Appendex B.2 of Chen et al. for more.
|
25 |
+
|
26 |
+
## Usage
|
27 |
+
|
28 |
+
```python
|
29 |
+
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
|
30 |
+
|
31 |
+
tokenizer = AutoTokenizer.from_pretrained('domenicrosati/question_converter-3b')
|
32 |
+
model = AutoModelForSeq2SeqLM.from_pretrained('domenicrosati/question_converter-3b')
|
33 |
+
|
34 |
+
question = "Where in the world is Carmen Sandiego?"
|
35 |
+
answer = "She is in D'Abruzzo"
|
36 |
+
|
37 |
+
prompt = f'{question} </s> {answer}'
|
38 |
+
input_ids = tokenizer(prompt, return_tensors='pt').input_ids
|
39 |
+
output_ids = model.generate(input_ids)
|
40 |
+
responses = tokenizer.batch_decode(output_ids, skip_special_tokens=True)
|
41 |
+
```
|