import gradio as gr
from llama_cpp import Llama

llm = Llama(model_path="model.gguf", n_ctx=8000, n_threads=2, chat_format="chatml")
  
def generate(message, history,temperature=0.3,max_tokens=512):
    system_prompt = """You are a super Inteligent AI assistant.
I want you to think smartly, step by step. 
Once you've thought through things step by step, check the responses
before issuing them. I want you to answer clearly, accurately,
and without any unnecessary words. I want you to be concise and provide exact answers,
with known data, without making things up. You're called "Little Llama", 
you're a language model that was compressed but you're still the smartest!"""
    formatted_prompt = [{"role": "system", "content": system_prompt}]
    for user_prompt, bot_response  in history:
        formatted_prompt.append({"role": "user", "content": user_prompt})
        formatted_prompt.append({"role": "assistant", "content": bot_response })
    formatted_prompt.append({"role": "user", "content": message})
    stream_response = llm.create_chat_completion(messages=formatted_prompt, temperature=temperature, max_tokens=max_tokens, stream=True)
    response  = ""
    for chunk in stream_response:
        if len(chunk['choices'][0]["delta"]) != 0 and "content" in chunk['choices'][0]["delta"]:
            response  += chunk['choices'][0]["delta"]["content"]
        yield response 

mychatbot = gr.Chatbot(
avatar_images=["user.png", "botnb.png"], bubble_full_width=False, show_label=False, show_copy_button=True, likeable=True,)
        
iface = gr.ChatInterface(fn=generate, chatbot=mychatbot, retry_btn=None, undo_btn=None)

with gr.Blocks() as demo:
    gr.HTML("<center><h1>Llama 13b - GGUF Q_4_K_M</h1></center>")
    iface.render()

demo.queue().launch(show_api=False, server_name="0.0.0.0")