app.py

import spaces
import os
import threading
from collections import deque

import plotly.graph_objs as go
import pynvml

import gradio as gr
from huggingface_hub import snapshot_download

from vptq.app_utils import get_chat_loop_generator

models = [
    {
        "name": "VPTQ-community/Meta-Llama-3.1-8B-Instruct-v8-k65536-256-woft",
        "bits": "3 bits"
    },
    {
        "name": "VPTQ-community/Meta-Llama-3.1-8B-Instruct-v8-k65536-65536-woft",
        "bits": "4 bits"
    },
    {
        "name": "VPTQ-community/Meta-Llama-3.1-70B-Instruct-v16-k65536-65536-woft",
        "bits": "2 bits"
    },
    {
        "name": "VPTQ-community/Meta-Llama-3.1-70B-Instruct-v8-k65536-256-woft",
        "bits": "3 bits"
    },
    {
        "name": "VPTQ-community/Meta-Llama-3.1-70B-Instruct-v8-k65536-65536-woft",
        "bits": "4 bits"
    },
    {
        "name": "VPTQ-community/Qwen2.5-72B-Instruct-v8-k65536-65536-woft",
        "bits": "4 bits"
    },
    {
        "name": "VPTQ-community/Qwen2.5-72B-Instruct-v8-k65536-256-woft",
        "bits": "3 bits"
    },
    {
        "name": "VPTQ-community/Qwen2.5-72B-Instruct-v16-k65536-65536-woft",
        "bits": "2 bits"
    },
]

# Queues for storing historical data (saving the last 100 GPU utilization and memory usage values)
gpu_util_history = deque(maxlen=100)
mem_usage_history = deque(maxlen=100)


def initialize_nvml():
    """
    Initialize NVML (NVIDIA Management Library).
    """
    pynvml.nvmlInit()


def get_gpu_info():
    """
    Get GPU utilization and memory usage information.

    Returns:
        dict: A dictionary containing GPU utilization and memory usage information.
    """
    handle = pynvml.nvmlDeviceGetHandleByIndex(0)  # Assuming a single GPU setup
    utilization = pynvml.nvmlDeviceGetUtilizationRates(handle)
    memory = pynvml.nvmlDeviceGetMemoryInfo(handle)

    gpu_info = {
        'gpu_util': utilization.gpu,
        'mem_used': memory.used / 1024**2,  # Convert bytes to MiB
        'mem_total': memory.total / 1024**2,  # Convert bytes to MiB
        'mem_percent': (memory.used / memory.total) * 100
    }
    return gpu_info


def _update_charts(chart_height: int = 200) -> go.Figure:
    """
    Update the GPU utilization and memory usage charts.

    Args:
        chart_height (int, optional): used to set the height of the chart. Defaults to 200.

    Returns:
        plotly.graph_objs.Figure: The updated figure containing the GPU and memory usage charts.
    """
    # obtain GPU information
    gpu_info = get_gpu_info()

    # records the latest GPU utilization and memory usage values
    gpu_util = round(gpu_info.get('gpu_util', 0), 1)
    mem_used = round(gpu_info.get('mem_used', 0) / 1024, 2)  # Convert MiB to GiB
    gpu_util_history.append(gpu_util)
    mem_usage_history.append(mem_used)

    # create GPU utilization line chart
    gpu_trace = go.Scatter(
        y=list(gpu_util_history),
        mode='lines+markers',
        text=list(gpu_util_history),
        line=dict(shape='spline', color='blue'),  # Make the line smooth and set color
        yaxis='y1'  # Link to y-axis 1
    )

    # create memory usage line chart
    mem_trace = go.Scatter(
        y=list(mem_usage_history),
        mode='lines+markers',
        text=list(mem_usage_history),
        line=dict(shape='spline', color='red'),  # Make the line smooth and set color
        yaxis='y2'  # Link to y-axis 2
    )

    # set the layout of the chart
    layout = go.Layout(
        xaxis=dict(title=None, showticklabels=False, ticks=''),
        yaxis=dict(
            title='GPU Utilization (%)',
            range=[-5, 110],
            titlefont=dict(color='blue'),
            tickfont=dict(color='blue'),
        ),
        yaxis2=dict(title='Memory Usage (GiB)',
                    range=[0, max(24,
                                  max(mem_usage_history) + 1)],
                    titlefont=dict(color='red'),
                    tickfont=dict(color='red'),
                    overlaying='y',
                    side='right'),
        height=chart_height,  # set the height of the chart
        margin=dict(l=10, r=10, t=0, b=0),  # set the margin of the chart
        showlegend=False  # disable the legend
    )

    fig = go.Figure(data=[gpu_trace, mem_trace], layout=layout)
    return fig


def initialize_history():
    """
    Initializes the GPU utilization and memory usage history.
    """
    for _ in range(100):
        gpu_info = get_gpu_info()
        gpu_util_history.append(round(gpu_info.get('gpu_util', 0), 1))
        mem_usage_history.append(round(gpu_info.get('mem_percent', 0), 1))


def enable_gpu_info():
    pynvml.nvmlInit()


def disable_gpu_info():
    pynvml.nvmlShutdown()
    
model_choices = [f"{model['name']} ({model['bits']})" for model in models]
display_to_model = {f"{model['name']} ({model['bits']})": model['name'] for model in models}


def download_model(model):
    print(f"Downloading {model['name']}...")
    snapshot_download(repo_id=model['name'])


def download_models_in_background():
    print('Downloading models for the first time...')
    for model in models:
        download_model(model)


download_thread = threading.Thread(target=download_models_in_background)
download_thread.start()

loaded_models = {}

@spaces.GPU(duration=120)
def respond(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
    selected_model_display_label,
):
    model_name = display_to_model[selected_model_display_label]

    # Check if the model is already loaded
    if model_name not in loaded_models:
        # Load and store the model in the cache
        loaded_models[model_name] = get_chat_loop_generator(model_name)

    chat_completion = loaded_models[model_name]

    messages = [{"role": "system", "content": system_message}]

    for val in history:
        if val[0]:
            messages.append({"role": "user", "content": val[0]})
        if val[1]:
            messages.append({"role": "assistant", "content": val[1]})

    messages.append({"role": "user", "content": message})

    response = ""

    for message in chat_completion(
            messages,
            max_tokens=max_tokens,
            stream=True,
            temperature=temperature,
            top_p=top_p,
    ):
        token = message

        response += token
        yield response


"""
For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
"""
# enable_gpu_info()
with gr.Blocks(fill_height=True) as demo:
    # with gr.Row():
    #   def update_chart():
    #       return _update_charts(chart_height=200)
    #       gpu_chart = gr.Plot(update_chart, every=0.1)  # update every 0.1 seconds

    with gr.Column():
        chat_interface = gr.ChatInterface(
            respond,
            additional_inputs=[
                gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
                gr.Slider(minimum=1, maximum=2048, value=512, 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)",
                ),
                gr.Dropdown(
                    choices=model_choices,
                    value=model_choices[0],
                    label="Select Model",
                ),
            ],
        )

if __name__ == "__main__":
    share = os.getenv("SHARE_LINK", None) in ["1", "true", "True"]
    demo.launch(share=share)
    # disable_gpu_info()