app.py

import gradio as gr
import streamlit as st
from transformers import pipeline
from datasets import load_dataset
from huggingface_hub import hf_hub_download
import subprocess
import os

# Clone the dataset repository if not already cloned
repo_url = "https://huggingface.co/datasets/BEE-spoke-data/survivorslib-law-books"
repo_dir = "./survivorslib-law-books"

if not os.path.exists(repo_dir):
    subprocess.run(["git", "clone", repo_url], check=True)

# Load the dataset from the cloned repository
dataset_path = os.path.join(repo_dir, "train.parquet")
ds = load_dataset("parquet", data_files=dataset_path)

# Initialize text-generation pipeline with the model
model_name = "nvidia/Llama-3.1-Nemotron-70B-Instruct-HF"
pipe = pipeline("text-generation", model=model_name)

# Preprocess dataset (assuming it has a 'text' or 'content' column for feeding to the model)
# If the dataset is different, update the field names accordingly
def preprocess_data(dataset):
    # Here, we assume that the dataset has a 'content' column with legal text
    # Adjust the column name as needed (for example, it might be 'text' or 'paragraph')
    return dataset['content'][:5]  # Displaying only the first 5 entries for brevity

# Gradio Interface setup
def respond(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
):
    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]})
    
    messages.append({"role": "user", "content": message})

    response = ""
    
    for message in pipe(
        prompt=message,
        max_length=max_tokens,
        do_sample=True,
        temperature=temperature,
        top_p=top_p,
    ):
        token = message["generated_text"]
        response += token
        yield response

# Streamlit Interface setup
def streamlit_interface():
    st.title("Canadian Legal Text Generator")
    st.write("Enter a prompt related to Canadian legal data and generate text using Llama-3.1.")
    
    # Show dataset sample (first 5 entries)
    st.subheader("Sample Data from Canadian Legal Dataset:")
    sample_data = preprocess_data(ds['train'])  # Assuming 'train' split
    st.write(sample_data)  # Display the first 5 rows of the dataset

    # Prompt input
    prompt = st.text_area("Enter your prompt:", placeholder="Type something...")

    if st.button("Generate Response"):
        if prompt:
            # Generate text based on the prompt
            with st.spinner("Generating response..."):
                generated_text = pipe(prompt, max_length=100, do_sample=True, temperature=0.7)[0]["generated_text"]
                st.write("**Generated Text:**")
                st.write(generated_text)
        else:
            st.write("Please enter a prompt to generate a response.")


# Running Gradio and Streamlit interfaces
if __name__ == "__main__":
    st.sidebar.title("Choose an Interface")
    interface = st.sidebar.radio("Select", ("Streamlit", "Gradio"))

    if interface == "Streamlit":
        streamlit_interface()
    else:
        demo = gr.ChatInterface(
            respond,
            additional_inputs=[
                gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
                gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
                gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
                gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)"),
            ],
        )
        demo.launch()