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


# 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 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."""
    endtags_to_remove = ["ul", "li", "br", "strong", "a"]
    html_string = remove_endtags(html_string, endtags_to_remove)

    starttag_replacements = {
        "<ul>": "",
        "<li>": "\n➤ ",
        "<br>": "\n",
        "<strong>": "",
        '<a href="https://opensource.org/license/mit">': "",
        '<a href="https://llama.meta.com/llama3/license/">': "",
        '<a href="https://atlas.nomic.ai/">': "",
    }
    formatted_string = replace_starttags(html_string, starttag_replacements)

    return formatted_string


def llm_intro(selected_model):
    html_string = model_description.get(selected_model, "No description available for this model selection.")
    formatted_description = format_html_string(html_string)
    return formatted_description

    
# 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]