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):
  if prompt=="":
     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 or selecting an example prompt below. 
Generation takes <b>15-20 seconds</b> on average.
Enjoy!

![visitor badge](https://visitor-badge.glitch.me/badge?page_id=Auto_Review_Generation_in_Turkish)
"""

#<p>NOTE: Examples can sometimes  generate ERROR. When you see ERROR on the screen <b>just click SUBMIT</b>. Model will generate text in 15-20 secs.</p> 
article = """<p style='text-align: center'>On YouTube:</p>
            <p style='text-align: center'><a href='https://youtube.com/playlist?list=PLQflnv_s49v9d9w-L0S8XUXXdNks7vPBL' target='_blank'>How to Train a Hugging Face Causal Language Model from Scratch with a Custom Dataset and a Custom Tokenizer?</a></p> 
            <p style='text-align: center'><a href='https://youtube.com/playlist?list=PLQflnv_s49v8aajw6m9MRNbAAbL63flKD' target='_blank'>Hugging Face kütüphanesini kullanarak bir GPT2 Transformer Dil Modelini Kendi Veri Setimizle nasıl eğitip kullanabiliriz? (in Turkish)</a></p>
            <p style='text-align: center'>On Medium:</p>
            <p style='text-align: center'><a href='https://medium.com/deep-learning-with-keras/how-to-train-a-hugging-face-causal-language-model-from-scratch-8d08d038168f' target='_blank'>How to Train a Hugging Face Causal Language Model from Scratch with a Custom Dataset and a Custom Tokenizer?</a></p>"""                    
examples=["Bir hafta önce aldığım cep telefonu çalışmıyor.",
          "Tatil için yaptığım rezervasyonu iptal edemiyorum.",
          "Geçen ay sipariş verdiğim ayakkabı gelmedi.",
          "Abone olduğum spor salonu kapandı.",
          "Buzdolabından garip sesler geliyor.",
          "Otel tam bir fiyasko."] 


demo = gr.Interface(fn=generate_review, 
                    inputs= gr.Textbox(lines=5, label="Prompt", placeholder="enter or select a prompt below..."),
                    outputs= gr.Textbox(lines=5, label="Generated Review", placeholder="genereated review will be here..."),
                    examples=examples,
                    title=title,
                    description= description,
                    article = article,
                    #cache_examples = False
                    allow_flagging="manual",
                    flagging_options=["good","moderate", "non-sense", ]
                    #flagging_dir='./flags'
                    )



demo.launch()