app.py

import gradio as gr
from transformers import T5Tokenizer, T5ForConditionalGeneration

# Load model and tokenizer
model = T5ForConditionalGeneration.from_pretrained("Addaci/mT5-small-experiment-13-checkpoint-2790")
tokenizer = T5Tokenizer.from_pretrained("Addaci/mT5-small-experiment-13-checkpoint-2790")

# Define task-specific prompts
def correct_htr_text(input_text, max_new_tokens, temperature):
    prompt = f"Correct the following handwritten transcription for obvious errors while preserving C17th spelling: {input_text}"
    inputs = tokenizer(prompt, return_tensors="pt")
    outputs = model.generate(
        inputs.input_ids, 
        max_new_tokens=max_new_tokens, 
        temperature=temperature
    )
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

def summarize_legal_text(input_text, max_new_tokens, temperature):
    prompt = f"Summarize this legal text: {input_text}"
    inputs = tokenizer(prompt, return_tensors="pt")
    outputs = model.generate(
        inputs.input_ids, 
        max_new_tokens=max_new_tokens, 
        temperature=temperature
    )
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

def answer_legal_question(input_text, question, max_new_tokens, temperature):
    prompt = f"Answer this question based on the legal text: '{question}' Text: {input_text}"
    inputs = tokenizer(prompt, return_tensors="pt")
    outputs = model.generate(
        inputs.input_ids, 
        max_new_tokens=max_new_tokens, 
        temperature=temperature
    )
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

# Define Gradio interface functions
def correct_htr_interface(text, max_new_tokens, temperature):
    return correct_htr_text(text, max_new_tokens, temperature)

def summarize_interface(text, max_new_tokens, temperature):
    return summarize_legal_text(text, max_new_tokens, temperature)

def question_interface(text, question, max_new_tokens, temperature):
    return answer_legal_question(text, question, max_new_tokens, temperature)

def clear_all():
    return "", ""

# External clickable buttons
def clickable_buttons():
    button_html = """
    <div style="display: flex; justify-content: space-between; margin-bottom: 10px;">
        <a href="http://www.marinelives.org/wiki/Tools:_Admiralty_court_legal_glossary" 
        style="border: 1px solid black; padding: 5px; text-align: center; width: 48%; background-color: #f0f0f0;">
        Admiralty Court Legal Glossary</a>
        <a href="https://github.com/Addaci/HCA/blob/main/HCA_13_70_Full_Volume_Processed_Text_EDITED_Ver.1.2_18062024.txt" 
        style="border: 1px solid black; padding: 5px; text-align: center; width: 48%; background-color: #f0f0f0;">
        HCA 13/70 Ground Truth</a>
    </div>
    """
    return button_html

# Interface layout
with gr.Blocks() as demo:
    gr.HTML("<h1>Marinelives mT5-small Legal Assistant</h1>")
    gr.HTML(clickable_buttons())
    
    with gr.Tab("Correct Raw HTR"):
        input_text = gr.Textbox(lines=10, label="Input Text")
        output_text = gr.Textbox(label="Corrected Text")
        max_new_tokens = gr.Slider(10, 512, value=128, label="Max New Tokens")
        temperature = gr.Slider(0.1, 1.0, value=0.7, label="Temperature")
        correct_button = gr.Button("Correct HTR")
        clear_button = gr.Button("Clear")
        
        correct_button.click(fn=correct_htr_interface, 
                             inputs=[input_text, max_new_tokens, temperature], 
                             outputs=output_text)
        clear_button.click(fn=clear_all, outputs=[input_text, output_text])

    with gr.Tab("Summarize Legal Text"):
        input_text_summarize = gr.Textbox(lines=10, label="Input Text")
        output_text_summarize = gr.Textbox(label="Summary")
        max_new_tokens_summarize = gr.Slider(10, 512, value=256, label="Max New Tokens")
        temperature_summarize = gr.Slider(0.1, 1.0, value=0.5, label="Temperature")
        summarize_button = gr.Button("Summarize Text")
        clear_button_summarize = gr.Button("Clear")
        
        summarize_button.click(fn=summarize_interface, 
                               inputs=[input_text_summarize, max_new_tokens_summarize, temperature_summarize], 
                               outputs=output_text_summarize)
        clear_button_summarize.click(fn=clear_all, outputs=[input_text_summarize, output_text_summarize])

    with gr.Tab("Answer Legal Question"):
        input_text_question = gr.Textbox(lines=10, label="Input Text")
        question = gr.Textbox(label="Question")
        output_text_question = gr.Textbox(label="Answer")
        max_new_tokens_question = gr.Slider(10, 512, value=128, label="Max New Tokens")
        temperature_question = gr.Slider(0.1, 1.0, value=0.7, label="Temperature")
        question_button = gr.Button("Get Answer")
        clear_button_question = gr.Button("Clear")
        
        question_button.click(fn=question_interface, 
                              inputs=[input_text_question, question, max_new_tokens_question, temperature_question], 
                              outputs=output_text_question)
        clear_button_question.click(fn=clear_all, outputs=[input_text_question, question, output_text_question])

demo.launch()