import gradio as gr
from langchain_community.llms import LlamaCpp

import_llm = LlamaCpp(
        model_path = "./model/model.gguf",
        n_ctx=4096,
        max_tokens=4096,
        temperature=0.0,
        top_p = 0.95,
        top_k = 50,
        verbose=False,
        streaming=True
    )

PROMPT = """Anda adalah seorang asisten virtual, tugas Anda adalah menjawab pertanyaan sesuai pengetahuan umum Anda.

Pertanyaan Pengguna: {}

Jawaban Anda:"""

PROMPT_HISTORY = """Anda adalah seorang asisten virtual, tugas Anda adalah menjawab pertanyaan sesuai pengetahuan umum Anda.

Berikut adalah riwayat percakapan anda:
[RIWAYAT]
{}
[/RIWAYAT]

Pertanyaan Lanjutan: {}

Jawaban Anda:"""

global llm
llm = None

chat_template = {
    "user": "<start_of_turn>user\n{}<end_of_turn>\n",
    "assistant": "<start_of_turn>model\n{}<end_of_turn>\n",
    "generation_prompt": "<start_of_turn>model\n"
}

api_version_file = open("./versions.txt", "r")
api_version = api_version_file.read().strip()

model_version_file = open("./versions.txt", "r")
model_version = model_version_file.read().strip()

with gr.Blocks(title="Gemma2-2b-it-GGUF") as nlp:
    gr.HTML("<h2 style= 'font-size: 3rem'>Gemma2-2b-it-GGUF</h2> \
            <div style = 'display: flex; flex-direction: row; align-items: center; gap: 0 5px;'>\
            <div><span style ='font-size: 16px; background-color: gray; padding: 3px 5px; border-radius: 5px;'>API Version: {}</span></div>\
            <div><span style ='font-size: 16px; background-color: gray; padding: 3px 5px; margin: 2px 0; border-radius: 5px;'>Model Version: {}</span></div>\
            </div>".format(api_version, model_version))
    bot = gr.Chatbot(label="Gemma2-2b-it")
    msg = gr.Textbox(
        placeholder="Tanya sesuatu! (Tekan \"enter\" untuk mengirim)", label="", elem_id="inputTextBox")
    clear = gr.Button('Clear', variant='stop')

    if llm == None:
        llm = import_llm

    def user(user_message, history):
        return "", history + [[user_message, None]]

    def predict(history):
        prompt = ""
    
        if len(history) == 1:
            chat = history[0]
            user_input = PROMPT.format(chat[0])
            prompt += chat_template['user'].format(user_input)
            prompt += chat_template['generation_prompt']
        else:
            chat_history = ""
            for chat in history[:-1]:
                if chat[0] is not None:
                    chat_history += f"Pertanyaan: {chat[0]}\n"
                if chat[1] is not None:
                    chat_history += f"Jawaban: {chat[1]}\n"
            user_input = PROMPT_HISTORY.format(
                chat_history, history[-1][0] if history[-1][0] is not None else ""
            )
            prompt += chat_template['user'].format(user_input)
            prompt += chat_template['generation_prompt']
    
        history[-1][1] = ""
        for token in llm.stream(prompt):
            history[-1][1] += token
            yield history

    msg.submit(user, [msg, bot], [msg, bot], queue=False).then(
        predict, bot, bot
    )
    clear.click(lambda: None, None, bot, queue=False)

nlp.launch(max_threads=1)