import gradio as gr
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer

# Load the model and tokenizer
model = AutoModelForSeq2SeqLM.from_pretrained("vennify/t5-base-grammar-correction")
tokenizer = AutoTokenizer.from_pretrained("vennify/t5-base-grammar-correction")

def correct_text(text, max_length, min_length, max_new_tokens, num_beams, temperature, top_p):
    inputs = tokenizer.encode("grammar: " + text, return_tensors="pt")
    
    if max_new_tokens > 0:
        outputs = model.generate(
            inputs,
            max_length=max_length,
            max_new_tokens=max_new_tokens,
            min_length=min_length,
            num_beams=num_beams,
            temperature=temperature,
            top_p=top_p,
            early_stopping=True
        )
    else:
        outputs = model.generate(
            inputs,
            max_length=max_length,
            min_length=min_length,
            num_beams=num_beams,
            temperature=temperature,
            top_p=top_p,
            early_stopping=True
        )
    
    corrected_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
    yield corrected_text

def respond(prompt, max_length, min_length, max_new_tokens, num_beams, temperature, top_p):
    # Your response generation logic here
    #response = correct_text(message, max_length, max_new_tokens, min_length, num_beams, temperature, top_p)
    #yield response
    #return f"System message: {system_message}, Max Length: {max_length}, Min Length: {min_length}, Max new tokens: {max_new_tokens}, Num Beams: {num_beams}, Temperature: {temperature}, Top-p: {top_p}"
    #response = correct_text(prompt, max_length, max_new_tokens, min_length, num_beams, temperature, top_p)
    return prompt

def update_prompt(text):
    return text

# Create the Gradio interface
with gr.Blocks() as demo:
    gr.Markdown(
    """
    # Grammar Correction App
    """)
    prompt_box = gr.Textbox(lines=2, placeholder="Enter your prompt here...")
    output_box = gr.Textbox()
    submitBtn = gr.Button("Submit")

    # Sample prompts
    with gr.Row():
        samp1 = gr.Button(value="we shood buy an car")
        samp2 = gr.Button(value="she is more taller")
        samp3 = gr.Button(value="John and i saw a sheep over their.")
        
        samp1.click(update_prompt, samp1.value, prompt_box)
        samp2.click(update_prompt, samp2.value, prompt_box)
        samp3.click(update_prompt, samp3.value, prompt_box)
        
    
    with gr.Accordion("Generation Parameters:", open=False):
        max_length = gr.Slider(minimum=1, maximum=256, value=80, step=1, label="Max Length")
        min_length = gr.Slider(minimum=1, maximum=256, value=0,  step=1, label="Min Length")
        max_tokens = gr.Slider(minimum=0, maximum=256, value=0, step=1, label="Max new tokens")
        num_beams = gr.Slider(minimum=1, maximum=10, value=5, step=1, label="Num Beams")
        temperature = gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature")
        top_p = gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)")

    

    submitBtn.click(correct_text, [prompt_box, max_length, min_length, max_tokens, num_beams, temperature, top_p], output_box)

    

demo.launch()