import torch
import gradio
import gradio as gr
from transformers import T5Tokenizer, T5ForConditionalGeneration, T5Config

#retrain


#initialize
device = torch.device("cpu") #("cuda")
model = T5ForConditionalGeneration.from_pretrained('t5-small')#,device_map="auto")
tokenizer = T5Tokenizer.from_pretrained('t5-small')#,device_map="auto")


def summ(text_content):
    
    preprocess_text = text_content.strip().replace("\n","")
    t5_inputText = "summarize: "+preprocess_text
    tokenized_text = tokenizer.encode(t5_inputText, return_tensors="pt").to(device)
    summary_ids = model.generate(tokenized_text,num_beams=4, no_repeat_ngram_size=2,
                                  min_length=30, max_length=300,early_stopping=True).to(device)
    summarized_output = tokenizer.decode(summary_ids[0], skip_special_tokens=True)
    return summarized_output 


def greet(text_content):
    bm25 = summ(text_content)
    return bm25

demo = gr.Interface(fn=greet, inputs="text", outputs="text")
print("Throwing up")
#demo.lauch()
if __name__ == "__main__":    demo.launch()