import gradio as gr
import torch
from transformers import GPT2Tokenizer, GPT2LMHeadModel

checkpoint = "gpt2"
tokenizer = GPT2Tokenizer.from_pretrained(checkpoint)


# Load the fine-tuned model and tokenizer
my_model = GPT2LMHeadModel.from_pretrained("jeevana/EmailSubjectLineGeneration")
my_tokenizer = GPT2Tokenizer.from_pretrained("jeevana/EmailSubjectLineGeneration")

def generate_response(model, tokenizer, prompt):
    input_ids = tokenizer.encode(prompt, return_tensors="pt",truncation=True, max_length=1000)
    # Create the attention mask and pad token id
    attention_mask = torch.ones_like(input_ids)
    pad_token_id = tokenizer.eos_token_id

    output = model.generate(
        input_ids,
        max_new_tokens=15,
        min_new_tokens = 1,
        num_return_sequences=1,
        attention_mask=attention_mask,
        pad_token_id=pad_token_id
    )
    response = tokenizer.decode(output[0], skip_special_tokens=True)
    print("Generated response::", response )
    print("len(prompt)::", len(prompt) )
    response = response[len(prompt) + 9:]
    return response

def predict(input):
    prediction = generate_response(my_model, my_tokenizer, input)
    print("type of response:", type(prediction))
    return prediction 


app = gr.Interface(fn=predict, inputs=[gr.Textbox(label="Email", lines=12)],
                    outputs=[gr.Textbox(label="Subject", lines=3)],
                    title="EmailSubjectLineGeneration",
                    description="EmailSubjectLineGeneration"
                   )
app.launch(share=True, debug=True)