You can use this model directly with a pipeline for text generation:
pip install -U transformers
pip install -U hazmfrom transformers import pipeline
import hazm
normalizer = hazm.Normalizer(persian_numbers=False)
def normalize_input(text):
text = normalizer.normalize(text)
return text
def sents_as_output(text, num_sents=1):
sents = hazm.sent_tokenize(text)
if num_sents > 0:
return " ".join(sents[:num_sents])
return " ".join(sents[0])
generator = pipeline('text-generation', "HooshvareLab/gpt2-fa")
text = "در یک اتفاق شگفت انگیز، پژوهشگران"
text = normalize_input(text)
outputs = generator(text)
for output in outputs:
generated = output["generated_text"]
print(sents_as_output(generated))در یک اتفاق شگفت انگیز، پژوهشگران قصد دارند با استفاده از دادههای حاصل از چندین تلسکوپ، عکسهایی با وضوح مختلف از سیارهی مشتری و زحل تهیه کنند.
This visualization is powered by Ecco (an interactive language modeling visualization).
| Notebook | |
|---|---|
| Ecco Visualization |  |
| Notebook | |
|---|---|
| Simple Version | |
If you fine-tuned gpt2-fa on your dataset, share it with us. All things you need to do, create a pull request and share yours with us. We are looking forward to it.
| Model | Description | How to Use |
|---|---|---|
| HooshvareLab/gpt2-fa-poetry | The model was fine-tuned on ChronologicalPersianPoetryDataset. | |
| HooshvareLab/gpt2-fa-comment | The model can generate comments based on your aspects, and the model was fine-tuned on persiannlp/parsinlu. Currently, the model only supports aspects in the food and movie scope. |
Please cite in publications as the following:
@misc{ParsGPT2,
author = {Hooshvare Team},
title = {ParsGPT2, a Persian version of GPT2},
year = {2021},
publisher = {GitHub},
journal = {GitHub repository},
howpublished = {\url{https://github.com/hooshvare/parsgpt}},
}Post a Github issue on the ParsGPT2 Issues repo.