import gradio as gr
from transformers import BartTokenizer, BartForConditionalGeneration

# Load the trained model and tokenizer from the specified directory
model_path = "bart_QA"
tokenizer = BartTokenizer.from_pretrained(model_path)
model = BartForConditionalGeneration.from_pretrained(model_path)

def answer_question(question, context):
    if question.strip() == "" or context.strip() == "":
        return "Please provide both a question and context."
    
    # Generate input sequence
    input_text = f"question: {question} context: {context}"
    inputs = tokenizer(input_text, return_tensors="pt", max_length=512, truncation=True)
    
    # Generate answer
    outputs = model.generate(**inputs, max_length=150, num_beams=4, early_stopping=True)
    answer = tokenizer.decode(outputs[0], skip_special_tokens=True)
    
    return answer

# Define the Gradio interface
with gr.Blocks() as demo:
    gr.Markdown("# BART QA System")
    
    with gr.Row():
        question_input = gr.Textbox(lines=1, placeholder="Enter your question here", label="Question")
        context_input = gr.Textbox(lines=5, placeholder="Enter the context here", label="Context")
    
    answer_output = gr.Textbox(label="Answer", interactive=False)
    
    with gr.Row():
        submit_btn = gr.Button("Submit")
        clear_btn = gr.Button("Clear")
    
    submit_btn.click(fn=answer_question, inputs=[question_input, context_input], outputs=answer_output)
    clear_btn.click(fn=lambda: ("", ""), inputs=[], outputs=[question_input, context_input])

demo.launch()