app.py

import gradio as gr
from gpt4all import GPT4All
from urllib.request import urlopen
import json
import time


# populate all models available from GPT4All
url = "https://raw.githubusercontent.com/nomic-ai/gpt4all/main/gpt4all-chat/metadata/models3.json"
response = urlopen(url)
data_json = json.loads(response.read())


def model_choices():
    model_list = [data_json[i]['filename'] for i in range(len(data_json))]
    return model_list

    
# get each models' description
model_description = {model['filename']: model['description'] for model in data_json}


def llm_intro(selected_model):
    html_string = model_description.get(selected_model, "No description available for this model selection.")
    formatted_description = html_string.replace("<ul>", "").replace("</ul>", "").replace("</li>", "").replace("<br>", "\n").replace("</br>", "").replace("<li>", "\n➤ ")
    return formatted_description


def remove_endtags(html_string, tags):
    """Remove rear HTML tags from the input string."""
    for tag in tags:
        html_string = re.sub(fr"</{tag}>", "", html_string)
    return html_string


def replace_starttags(html_string, replacements):
    """Replace starting HTML tags with the corresponding values."""
    for tag, replacement in replacements.items():
        html_string = html_string.replace(tag, replacement)
    return html_string


def format_html_string(html_string):
    """Format the HTML string to a readable text format."""
    tags_to_remove = ["ul", "li", "br"]
    html_string = remove_endtags(html_string, tags_to_remove)

    tag_replacements = {
        "<li>": "\n➤ ",
        "<br>": "\n",
        "<strong>": "**",
        "</strong>": "**"
    }
    formatted_string = replace_starttags(html_string, tag_replacements)

    return formatted_string

    
# cache models for faster reloads
model_cache = {}  


def load_model(model_name):
    """
    This function checks the cache before loading a model.
    If the model is cached, it returns the cached version.
    Otherwise, it loads the model, caches it, and then returns it.
    """
    if model_name not in model_cache:
        model = GPT4All(model_name)  
        model_cache[model_name] = model
    return model_cache[model_name]

#     clear = gr.ClearButton([input_text, chatbot])

# Construct chatbot
def generate_response(model_name, message, chat_history):
    model = load_model(model_name)
    chat_history = []
    if len(chat_history) > 0:
        past_chat = ", ".join(chat_history)
        input_text = past_chat + " " + message
    else:
        input_text = message
    response = model.generate(input_text, max_tokens=100)
    chat_history.append((input_text, response))
    return chat_history, response

# Create Gradio UI
with gr.Blocks() as demo:
    gr.Markdown("# GPT4All Chatbot")
    with gr.Row():
        with gr.Column(scale=1):
            model_dropdown = gr.Dropdown(
                choices=model_choices(),  
                multiselect=False,
                type="value",
                value="orca-mini-3b-gguf2-q4_0.gguf",
                label="LLMs to choose from"
            )
            explanation = gr.Textbox(label="Model Description", lines=3, interactive=False, value=llm_intro("orca-mini-3b-gguf2-q4_0.gguf"))
    
            # Link the dropdown with the textbox to update the description based on the selected model
            model_dropdown.change(fn=llm_intro, inputs=model_dropdown, outputs=explanation)
            
        with gr.Column(scale=4):
            chatbot = gr.Chatbot(label="Chatroom", value=[(None, "How may I help you today?")])
            
    message = gr.Textbox(label="Message")
    state = gr.State()

    message.submit(generate_response, inputs=[model_dropdown, message, state], outputs=[chatbot, state])

# Launch the Gradio app
demo.launch()