Add usage example
Browse files
README.md
CHANGED
@@ -1,5 +1,28 @@
|
|
1 |
## Overview
|
2 |
-
This model is a finetuned version of [mt5-small](https://huggingface.co/google/mt5-small) for question paraphrasing task in Turkish. As a generator model, its
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
|
4 |
## Dataset
|
5 |
I used [TQP dataset V0.1](https://github.com/monatis/tqp) that I've published just recently. This model should be taken as as a baseline model for TQP dataset. A cleaning and further improvements in the dataset and an elaborate hyperparameter tuning may boost the performance.
|
|
|
1 |
## Overview
|
2 |
+
This model is a finetuned version of [mt5-small](https://huggingface.co/google/mt5-small) for question paraphrasing task in Turkish. As a generator model, its capabilities are currently investigated and there is an ongoing effort to further improve it.
|
3 |
+
|
4 |
+
## Usage
|
5 |
+
You can generate 5 paraphrases for the input question with The simple code below.
|
6 |
+
|
7 |
+
```python
|
8 |
+
from transformers import AutoTokenizer, T5ForConditionalGeneration
|
9 |
+
model_name = "mys/mt5-small-turkish-question-paraphrasing"
|
10 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
11 |
+
model = T5ForConditionalGeneration.from_pretrained(model_name)
|
12 |
+
|
13 |
+
tokens = tokenizer.encode_plus("Yarın toplantı kaçta başlıyor?", return_tensors='pt')
|
14 |
+
paraphrases = model.generate(tokens['input_ids'], max_length=128, num_return_sequences=5, num_beams=5)
|
15 |
+
tokenizer.batch_decode(paraphrases, skip_special_tokens=True)
|
16 |
+
```
|
17 |
+
|
18 |
+
And the output will be something like:
|
19 |
+
```shell
|
20 |
+
['Yarın toplantı ne zaman başlıyor?',
|
21 |
+
'Yarın toplantı saat kaçta başlıyor?',
|
22 |
+
'Yarın toplantı saat kaçta başlar?',
|
23 |
+
'Yarın toplantı ne zaman başlayacak?',
|
24 |
+
'Yarın toplantı ne zaman başlar?']
|
25 |
+
```
|
26 |
|
27 |
## Dataset
|
28 |
I used [TQP dataset V0.1](https://github.com/monatis/tqp) that I've published just recently. This model should be taken as as a baseline model for TQP dataset. A cleaning and further improvements in the dataset and an elaborate hyperparameter tuning may boost the performance.
|