import gradio as gr
from transformers import AutoTokenizer, TFGPT2LMHeadModel
review_model = TFGPT2LMHeadModel.from_pretrained("kmkarakaya/turkishReviews-ds")
review_tokenizer = AutoTokenizer.from_pretrained("kmkarakaya/turkishReviews-ds")
def generate_review(prompt):
input_ids = review_tokenizer.encode(prompt, return_tensors='tf')
context_length = 40
output = review_model.generate(
input_ids,
do_sample=True,
max_length=context_length,
top_k=10,
no_repeat_ngram_size=2,
early_stopping=True
)
return(review_tokenizer.decode(output[0], skip_special_tokens=True))
title="Turkish Review Generator: A GPT2 based Text Generator Trained with a Custom Dataset"
description= """Generate a review in Turkish by providing a prompt.
Generation takes 15-20 seconds on average.""""
article = """On YouTube:
How to Train a Hugging Face Causal Language Model from Scratch with a Custom Dataset and a Custom Tokenizer? (in English):
Hugging Face kütüphanesini kullanarak bir GPT2 Transformer Dil Modelini Kendi Veri Setimizle nasıl eğitip kullanabiliriz? (in Turkish):
On Medium:
Hugging Face kütüphanesini kullanarak bir GPT2 Transformer Dil Modelini Kendi Veri Setimizle nasıl eğitip kullanabiliriz? (in Turkish):
"""
examples=["Bir hafta önce aldığım cep telefonu",
"Tatil için rezervasyon yaptırdım.",
"Henüz alalı bir",
"Spor salonuna 1 yıllık abone oldum ama"]
demo = gr.Interface(fn=generate_review,
inputs="text",
outputs="text",
examples=examples,
title=title,
description= description,
article = article
)
demo.launch()