import gradio as gr
import torch
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
import time
import sys
import traceback

# Global variables to store error information
error_message = ""

# Global variables for model and tokenizer
model = None
tokenizer = None
device = None

# Load the model and tokenizer from Hugging Face
model_name = "ambrosfitz/history-qa-flan-t5-large"
try:
    global model, tokenizer, device
    model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
    tokenizer = AutoTokenizer.from_pretrained(model_name, use_fast=False)
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    model.to(device)
except Exception as e:
    error_message = f"Error loading model or tokenizer: {str(e)}\n{traceback.format_exc()}"
print(error_message)

def generate_qa(text, max_length=512):
    global model, tokenizer, device
    try:
        input_text = f"Generate a history question and answer based on this text: {text}"
        input_ids = tokenizer(input_text, return_tensors="pt", max_length=max_length, truncation=True).input_ids.to(device)
        
        with torch.no_grad():
            outputs = model.generate(input_ids, max_length=max_length, num_return_sequences=1, do_sample=True, temperature=0.7)
            generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
        
        # Parse the generated text
        parts = generated_text.split("Question: ")
        if len(parts) > 1:
            qa_parts = parts[1].split("Answer: ")
            question = qa_parts[0].strip()
            answer = qa_parts[1].strip() if len(qa_parts) > 1 else "No answer provided."
            return f"Question: {question}\n\nAnswer: {answer}"
        else:
            return "Unable to generate a proper question and answer. Please try again with a different input."
    except Exception as e:
        return f"An error occurred: {str(e)}\n{traceback.format_exc()}"

def slow_qa(message, history):
    try:
        full_response = generate_qa(message)
        for i in range(len(full_response)):
            time.sleep(0.01)
            yield full_response[:i+1]
    except Exception as e:
        yield f"An error occurred: {str(e)}\n{traceback.format_exc()}"

# Create and launch the Gradio interface
try:
    iface = gr.ChatInterface(
        slow_qa,
        chatbot=gr.Chatbot(height=500),
        textbox=gr.Textbox(placeholder="Enter historical text here...", container=False, scale=7),
        title="History Q&A Generator (FLAN-T5)",
        description="Enter a piece of historical text, and the model will generate a related question and answer.",
        theme="soft",
        examples=[
            "The American Revolution was a colonial revolt that took place between 1765 and 1783.",
            "World War II was a global conflict that lasted from 1939 to 1945, involving many of the world's nations.",
            "The Renaissance was a period of cultural, artistic, political, and economic revival following the Middle Ages."
        ],
        cache_examples=False,
        retry_btn="Regenerate",
        undo_btn="Remove last",
        clear_btn="Clear",
    )
    
    if error_message:
        print("Launching interface with error message.")
    else:
        print("Launching interface normally.")
    iface.launch(debug=True)
except Exception as e:
    print(f"An error occurred while creating or launching the interface: {str(e)}\n{traceback.format_exc()}")