app.py

import os
os.system('pip install torch')  # or 'pip install tensorflow'
os.system('pip install transformers')
os.system('pip install datasets')
os.system('pip install gradio')
os.system('pip install minijinja')
os.system('pip install PyMuPDF')

import gradio as gr
from huggingface_hub import InferenceClient
from transformers import pipeline
from datasets import load_dataset
import fitz  # PyMuPDF

client = InferenceClient()

dataset = load_dataset("ibunescu/qa_legal_dataset_train")

def score_argument_from_outcome(outcome, argument):
    prosecutor_score = 0
    if "Prosecutor" in outcome:
        prosecutor_score = outcome.count("Prosecutor") * 2
        if "won" in outcome and "Prosecutor" in outcome:
            prosecutor_score += 10
    return prosecutor_score

def chat_between_bots(system_message1, system_message2, max_tokens, temperature, top_p, history1, history2, shared_history, message):
    response1, history1 = list(respond(message, history1, system_message1, max_tokens, temperature, top_p))[-1]
    response2, history2 = list(respond(message, history2, system_message2, max_tokens, temperature, top_p))[-1]
    
    return response1, response2, history1, history2, shared_history

def extract_text_from_pdf(pdf_file):
    text = ""
    doc = fitz.open(pdf_file)
    for page in doc:
        text += page.get_text()
    return text

def ask_about_pdf(pdf_text, question):
    prompt = f"PDF Content: {pdf_text}\n\nQuestion: {question}\n\nAnswer:"
    response = ""
    for message in client.chat_completion(
        [{"role": "system", "content": "You are a legal expert answering questions based on the PDF content provided."},
         {"role": "user", "content": prompt}],
        max_tokens=512,
        stream=True,
        temperature=0.6,
        top_p=0.95,
    ):
        token = message.choices[0].delta.content
        if token is not None:
            response += token
    return response

def update_pdf_gallery_and_extract_text(pdf_files):
    if len(pdf_files) > 0:
        pdf_text = extract_text_from_pdf(pdf_files[0].name)
    else:
        pdf_text = ""
    return pdf_files, pdf_text

def add_message(history, message):
    history.append(message)
    return history, gr.Textbox(value=None, interactive=False)

def bot(history):
    system_message = "You are a helpful assistant."
    messages = [{"role": "system", "content": system_message}]
    for val in history:
        if val[0]:
            messages.append({"role": "user", "content": val[0]})
        if val[1]:
            messages.append({"role": "assistant", "content": val[1]})
    response = ""
    for message in client.chat_completion(
        messages,
        max_tokens=150,
        stream=True,
        temperature=0.6,
        top_p=0.95,
    ):
        token = message.choices[0].delta.content
        if token is not None:
            response += token
        history[-1][1] = response
        yield history

def print_like_dislike(x: gr.LikeData):
    print(x.index, x.value, x.liked)

def reset_conversation():
    return [], [], "", "", ""

def save_conversation(history1, history2, shared_history):
    return history1, history2, shared_history

custom_css = """
.scroll-box {
    max-height: 400px;
    overflow-y: auto;
}
"""

with gr.Blocks(css=custom_css) as demo:
    history1 = gr.State([])
    history2 = gr.State([])
    shared_history = gr.State([])
    pdf_files = gr.State([])
    pdf_text = gr.State("")
    
    with gr.Tab("Argument Evaluation"):
        message = gr.Textbox(label="Case to Argue")
        system_message1 = "System message for bot 1"
        system_message2 = "System message for bot 2"
        max_tokens = 150
        temperature = 0.6
        top_p = 0.95
        
        prosecutor_response = gr.Textbox(label="Prosecutor Response", interactive=False)
        defense_response = gr.Textbox(label="Defense Response", interactive=False)
        prosecutor_score_color = gr.Textbox(label="Prosecutor Score Color", interactive=False)
        defense_score_color = gr.Textbox(label="Defense Score Color", interactive=False)
        shared_argument = gr.Textbox(label="Case Outcome", interactive=True)
        
        submit_btn = gr.Button("Argue")
        clear_btn = gr.Button("Clear and Reset")
        save_btn = gr.Button("Save Conversation")
        
        submit_btn.click(chat_between_bots, inputs=[system_message1, system_message2, max_tokens, temperature, top_p, history1, history2, shared_history, message], outputs=[prosecutor_response, defense_response, history1, history2, shared_argument, prosecutor_score_color, defense_score_color])
        clear_btn.click(reset_conversation, outputs=[history1, history2, shared_history, prosecutor_response, defense_response, shared_argument])
        save_btn.click(save_conversation, inputs=[history1, history2, shared_history], outputs=[history1, history2, shared_history])
    
    with gr.Tab("PDF Management"):
        pdf_upload = gr.File(label="Upload Case Files (PDF)", file_types=[".pdf"])
        pdf_gallery = gr.Gallery(label="PDF Gallery")
        pdf_view = gr.Textbox(label="PDF Content", interactive=False, elem_classes=["scroll-box"])
        pdf_question = gr.Textbox(label="Ask a Question about the PDF")
        pdf_answer = gr.Textbox(label="Answer", interactive=False, elem_classes=["scroll-box"])
        pdf_upload_btn = gr.Button("Update PDF Gallery")
        pdf_ask_btn = gr.Button("Ask")
        
        pdf_upload_btn.click(update_pdf_gallery_and_extract_text, inputs=[pdf_upload], outputs=[pdf_gallery, pdf_text])
        pdf_text.change(fn=lambda x: x, inputs=pdf_text, outputs=pdf_view)
        pdf_ask_btn.click(ask_about_pdf, inputs=[pdf_text, pdf_question], outputs=pdf_answer)
    
    with gr.Tab("Chatbot"):
        chatbot = gr.Chatbot()
        
demo.launch()