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

# Load the model and tokenizer from Hugging Face repository
model_repo_id = "Ajay12345678980/QA_GPT_J"  # Replace with your model repository ID

# Initialize the model and tokenizer
device = "cuda" if torch.cuda.is_available() else "cpu"
model = GPT2LMHeadModel.from_pretrained(model_repo_id).to(device)
tokenizer = GPT2Tokenizer.from_pretrained(model_repo_id)

# Define the prediction function
def generate_answer(question):
    input_ids = tokenizer.encode(question, return_tensors="pt").to(device)
    attention_mask = torch.ones_like(input_ids).to(device)
    pad_token_id = tokenizer.eos_token_id

    output = model.generate(
        input_ids,
        max_new_tokens=100,
        num_return_sequences=1,
        attention_mask=attention_mask,
        pad_token_id=pad_token_id
    )
    decoded_output = tokenizer.decode(output[0], skip_special_tokens=True)
    start_index = decoded_output.find("Answer")
    end_index = decoded_output.find("<ANSWER_ENDED>")

    if start_index != -1:
        if end_index != -1:
            answer_text = decoded_output[start_index + len("Answer"):end_index].strip()
        else:
            answer_text = decoded_output[start_index + len("Answer"):].strip()
        return answer_text
    else:
        return "Sorry, I couldn't generate an answer."

# Gradio interface setup
interface = gr.Interface(
    fn=generate_answer, 
    inputs="text", 
    outputs="text", 
    title="GPT-2 Text Generation",
    description="Enter a question and see what the model generates!"
)

# Launch the Gradio app
if __name__ == "__main__":
    interface.launch()