|
--- |
|
language: fa |
|
tags: |
|
- Style transfer |
|
- Formality style transfer |
|
widget: |
|
- text: "من با دوستام میرم بازی." |
|
- text: "من به خونه دوستم رفتم." |
|
|
|
--- |
|
|
|
# Persian-t5-formality-transfer |
|
|
|
This is a formality style transfer model for the Persian language to convert colloquial text into a formal one. It is based on [the monolingual T5 model for Persian.](https://huggingface.co/Ahmad/parsT5-base) and [Persian T5 paraphraser](https://huggingface.co/erfan226/persian-t5-paraphraser) |
|
|
|
Note: This model is still in development and therefore its outputs might not be very good. However, you can experiment with different values for the decoder to get better results. For more info check this [link.](https://huggingface.co/blog/how-to-generate) |
|
|
|
## Usage |
|
|
|
```python |
|
>>> pip install transformers |
|
>>> from transformers import (T5ForConditionalGeneration, AutoTokenizer, pipeline) |
|
>>> import torch |
|
|
|
model_path = 'erfan226/persian-t5-formality-transfer' |
|
model = T5ForConditionalGeneration.from_pretrained(model_path) |
|
tokenizer = AutoTokenizer.from_pretrained(model_path) |
|
pipe = pipeline(task='text2text-generation', model=model, tokenizer=tokenizer) |
|
|
|
def paraphrase(text): |
|
for j in range(3): |
|
out = pipe(text, encoder_no_repeat_ngram_size=4, do_sample=True, num_beams=5, max_length=128)[0]['generated_text'] |
|
print("Paraphrase:", out) |
|
|
|
text = "من با دوستام میرم بازی" |
|
print("Original:", text) |
|
paraphrase(text) |
|
|
|
# Original: من با دوستام میرم بازی |
|
# Paraphrase: دوست دارم با دوستانم بازی کنم. |
|
# Paraphrase: من با دوستانم میرم... |
|
# Paraphrase: من با دوستام بازی می کنم. |
|
|
|
``` |
|
|
|
## Training data |
|
TBD |
