import spaces
import json
import subprocess
import gradio as gr
from huggingface_hub import hf_hub_download

subprocess.run('pip install llama-cpp-python==0.2.75 --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cu124', shell=True)
subprocess.run('pip install llama-cpp-agent==0.2.10', shell=True)

hf_hub_download(repo_id="bartowski/Meta-Llama-3-70B-Instruct-GGUF", filename="Meta-Llama-3-70B-Instruct-Q3_K_M.gguf",  local_dir = "./models")

@spaces.GPU(duration=120)
def respond(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
):
    from llama_cpp import Llama
    from llama_cpp_agent import LlamaCppAgent
    from llama_cpp_agent import MessagesFormatterType
    from llama_cpp_agent.providers import LlamaCppPythonProvider
    from llama_cpp_agent.chat_history import BasicChatHistory
    from llama_cpp_agent.chat_history.messages import Roles
    
    llm = Llama(
        model_path="models/Meta-Llama-3-70B-Instruct-Q3_K_M.gguf",
        n_gpu_layers=81,
    )
    provider = LlamaCppPythonProvider(llm)

    agent = LlamaCppAgent(
        provider,
        system_prompt="You are a helpful assistant.",
        predefined_messages_formatter_type=MessagesFormatterType.LLAMA_3,
        debug_output=True
    )
    
    settings = provider.get_provider_default_settings()
    settings.max_tokens = max_tokens
    settings.stream = True

    messages = BasicChatHistory()
    print("history")
    print(history)

    for msn in history:
        user = {
            'role': Roles.user,
            'content': msn[0]
        }
        assistant = {
            'role': Roles.assistant,
            'content': msn[1]
        }
        print(user)
        print(assistant)
        messages.add_message(user)
        messages.add_message(assistant)
    
    stream = agent.get_chat_response(message, llm_sampling_settings=settings, chat_history=messages, returns_streaming_generator=True, print_output=False)
    
    outputs = ""
    for output in stream:
        outputs += output
        yield outputs

demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(value="You are a helpful assistant.", label="System message"),
        gr.Slider(minimum=1, maximum=4096, value=4096, 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)",
        ),
    ],
    theme=gr.themes.Soft(primary_hue="green", secondary_hue="indigo", neutral_hue="zinc",font=[gr.themes.GoogleFont("Exo 2"), "ui-sans-serif", "system-ui", "sans-serif"]).set(
        block_background_fill_dark="*neutral_800"
    )
)

if __name__ == "__main__":
    demo.launch()