pabloce/llama-cpp-agent

1import spaces2import json3import subprocess4from llama_cpp import Llama5from llama_cpp_agent import LlamaCppAgent, MessagesFormatterType6from llama_cpp_agent.providers import LlamaCppPythonProvider7from llama_cpp_agent.chat_history import BasicChatHistory8from llama_cpp_agent.chat_history.messages import Roles9import gradio as gr10from huggingface_hub import hf_hub_download11 12llm = None13llm_model = None14 15hf_hub_download(16    repo_id="bartowski/Meta-Llama-3-70B-Instruct-GGUF",17    filename="Meta-Llama-3-70B-Instruct-Q3_K_M.gguf",18    local_dir = "./models"19)20hf_hub_download(21    repo_id="bartowski/Mistral-Small-24B-Instruct-2501-GGUF",22    filename="Mistral-Small-24B-Instruct-2501-Q8_0.gguf",23    local_dir = "./models"24)25 26css = """27.bubble-wrap {28    padding-top: calc(var(--spacing-xl) * 3) !important;29}30.message-row {31    justify-content: space-evenly !important;32    width: 100% !important;33    max-width: 100% !important;34    margin: calc(var(--spacing-xl)) 0 !important;35    padding: 0 calc(var(--spacing-xl) * 3) !important;36}37.flex-wrap.user {38    border-bottom-right-radius: var(--radius-lg) !important;39}40.flex-wrap.bot {41    border-bottom-left-radius: var(--radius-lg) !important;42}43.message.user{44    padding: 10px;45}46.message.bot{47    text-align: right;48    width: 100%;49    padding: 10px;50    border-radius: 10px;51}52.message-bubble-border {53    border-radius: 6px !important;54}55.message-buttons {56    justify-content: flex-end !important;57}58.message-buttons-left {59    align-self: end !important;60}61.message-buttons-bot, .message-buttons-user {62    right: 10px !important;63    left: auto !important;64    bottom: 2px !important;65}66.dark.message-bubble-border {67    border-color: #343140 !important;68}69.dark.user {70    background: #1e1c26 !important;71}72.dark.assistant.dark, .dark.pending.dark {73    background: #16141c !important;74}75"""76 77def get_messages_formatter_type(model_name):78    if "Llama" in model_name:79        return MessagesFormatterType.LLAMA_380    elif "unsloth" in model_name:81        return MessagesFormatterType.CHATML82    else:83        raise ValueError(f"Unsupported model: {model_name}")84 85@spaces.GPU(duration=120)86def respond(87    message,88    history: list[tuple[str, str]],89    model,90    system_message,91    max_tokens,92    temperature,93    top_p,94    top_k,95    repeat_penalty,96):97    global llm98    global llm_model99    100    chat_template = get_messages_formatter_type(model)101    102    if llm is None or llm_model != model:103        llm = Llama(104            model_path=f"models/{model}",105            flash_attn=True,106            n_gpu_layers=81,107            n_batch=1024,108            n_ctx=8192,109        )110        llm_model = model111    112    provider = LlamaCppPythonProvider(llm)113 114    agent = LlamaCppAgent(115        provider,116        system_prompt=f"{system_message}",117        predefined_messages_formatter_type=chat_template,118        debug_output=True119    )120    121    settings = provider.get_provider_default_settings()122    settings.temperature = temperature123    settings.top_k = top_k124    settings.top_p = top_p125    settings.max_tokens = max_tokens126    settings.repeat_penalty = repeat_penalty127    settings.stream = True128 129    messages = BasicChatHistory()130 131    for msn in history:132        user = {133            'role': Roles.user,134            'content': msn[0]135        }136        assistant = {137            'role': Roles.assistant,138            'content': msn[1]139        }140        messages.add_message(user)141        messages.add_message(assistant)142    143    stream = agent.get_chat_response(144        message,145        llm_sampling_settings=settings,146        chat_history=messages,147        returns_streaming_generator=True,148        print_output=False149    )150    151    outputs = ""152    for output in stream:153        outputs += output154        yield outputs155 156PLACEHOLDER = """157<div class="message-bubble-border" style="display:flex; max-width: 600px; border-radius: 6px; border-width: 1px; border-color: #e5e7eb; box-shadow: 0 4px 6px rgba(0, 0, 0, 0.1); backdrop-filter: blur(10px);">158    <figure style="margin: 0;max-width: 200px;min-height: 300px;">159        <img src="https://huggingface.co/spaces/pabloce/llama-cpp-agent/resolve/main/llama.jpg" alt="Logo" style="width: 100%; height: 100%; border-radius: 8px;">160    </figure>161    <div style="padding: .5rem 1.5rem;display: flex;flex-direction: column;justify-content: space-evenly;">162        <h2 style="text-align: left; font-size: 1.5rem; font-weight: 700; margin-bottom: 0.5rem;">llama-cpp-agent</h2>163        <p style="text-align: left; font-size: 16px; line-height: 1.5; margin-bottom: 15px;">The llama-cpp-agent framework simplifies interactions with Large Language Models (LLMs), providing an interface for chatting, executing function calls, generating structured output, performing retrieval augmented generation, and processing text using agentic chains with tools.</p>164        <div style="display: flex; justify-content: space-between; align-items: center;">165            <div style="display: flex; flex-flow: column; justify-content: space-between;">166                <span style="display: inline-flex; align-items: center; border-radius: 0.375rem; background-color: rgba(229, 70, 77, 0.1); padding: 0.1rem 0.75rem; font-size: 0.75rem; font-weight: 500; color: #f88181; margin-bottom: 2.5px;">167                    Mistral Small 24B Instruct 2501168                </span>169                <span style="display: inline-flex; align-items: center; border-radius: 0.375rem; background-color: rgba(79, 70, 229, 0.1); padding: 0.1rem 0.75rem; font-size: 0.75rem; font-weight: 500; color: #60a5fa; margin-top: 2.5px;">170                    Meta Llama 3 70B Instruct171                </span>172            </div>173            <div style="display: flex; justify-content: flex-end; align-items: center;">174                <a href="https://discord.gg/fgr5RycPFP" target="_blank" rel="noreferrer" style="padding: .5rem;">175                    <svg width="24" height="24" fill="currentColor" xmlns="http://www.w3.org/2000/svg" viewBox="0 5 30.67 23.25">176                        <title>Discord</title>177                        <path d="M26.0015 6.9529C24.0021 6.03845 21.8787 5.37198 19.6623 5C19.3833 5.48048 19.0733 6.13144 18.8563 6.64292C16.4989 6.30193 14.1585 6.30193 11.8336 6.64292C11.6166 6.13144 11.2911 5.48048 11.0276 5C8.79575 5.37198 6.67235 6.03845 4.6869 6.9529C0.672601 12.8736 -0.41235 18.6548 0.130124 24.3585C2.79599 26.2959 5.36889 27.4739 7.89682 28.2489C8.51679 27.4119 9.07477 26.5129 9.55525 25.5675C8.64079 25.2265 7.77283 24.808 6.93587 24.312C7.15286 24.1571 7.36986 23.9866 7.57135 23.8161C12.6241 26.1255 18.0969 26.1255 23.0876 23.8161C23.3046 23.9866 23.5061 24.1571 23.7231 24.312C22.8861 24.808 22.0182 25.2265 21.1037 25.5675C21.5842 26.5129 22.1422 27.4119 22.7621 28.2489C25.2885 27.4739 27.8769 26.2959 30.5288 24.3585C31.1952 17.7559 29.4733 12.0212 26.0015 6.9529ZM10.2527 20.8402C8.73376 20.8402 7.49382 19.4608 7.49382 17.7714C7.49382 16.082 8.70276 14.7025 10.2527 14.7025C11.7871 14.7025 13.0425 16.082 13.0115 17.7714C13.0115 19.4608 11.7871 20.8402 10.2527 20.8402ZM20.4373 20.8402C18.9183 20.8402 17.6768 19.4608 17.6768 17.7714C17.6768 16.082 18.8873 14.7025 20.4373 14.7025C21.9717 14.7025 23.2271 16.082 23.1961 17.7714C23.1961 19.4608 21.9872 20.8402 20.4373 20.8402Z"></path>178                    </svg>179                </a>180                <a href="https://github.com/Maximilian-Winter/llama-cpp-agent" target="_blank" rel="noreferrer" style="padding: .5rem;">181                    <svg width="24" height="24" fill="currentColor" viewBox="3 3 18 18">182                        <title>GitHub</title>183                        <path d="M12 3C7.0275 3 3 7.12937 3 12.2276C3 16.3109 5.57625 19.7597 9.15374 20.9824C9.60374 21.0631 9.77249 20.7863 9.77249 20.5441C9.77249 20.3249 9.76125 19.5982 9.76125 18.8254C7.5 19.2522 6.915 18.2602 6.735 17.7412C6.63375 17.4759 6.19499 16.6569 5.8125 16.4378C5.4975 16.2647 5.0475 15.838 5.80124 15.8264C6.51 15.8149 7.01625 16.4954 7.18499 16.7723C7.99499 18.1679 9.28875 17.7758 9.80625 17.5335C9.885 16.9337 10.1212 16.53 10.38 16.2993C8.3775 16.0687 6.285 15.2728 6.285 11.7432C6.285 10.7397 6.63375 9.9092 7.20749 9.26326C7.1175 9.03257 6.8025 8.08674 7.2975 6.81794C7.2975 6.81794 8.05125 6.57571 9.77249 7.76377C10.4925 7.55615 11.2575 7.45234 12.0225 7.45234C12.7875 7.45234 13.5525 7.55615 14.2725 7.76377C15.9937 6.56418 16.7475 6.81794 16.7475 6.81794C17.2424 8.08674 16.9275 9.03257 16.8375 9.26326C17.4113 9.9092 17.76 10.7281 17.76 11.7432C17.76 15.2843 15.6563 16.0687 13.6537 16.2993C13.98 16.5877 14.2613 17.1414 14.2613 18.0065C14.2613 19.2407 14.25 20.2326 14.25 20.5441C14.25 20.7863 14.4188 21.0746 14.8688 20.9824C16.6554 20.364 18.2079 19.1866 19.3078 17.6162C20.4077 16.0457 20.9995 14.1611 21 12.2276C21 7.12937 16.9725 3 12 3Z"></path>184                    </svg>185                </a>186            </div>187        </div>188    </div>189</div>190"""191 192demo = gr.ChatInterface(193    respond,194    additional_inputs=[195        gr.Dropdown([196                'Mistral-Small-24B-Instruct-2501-Q8_0.gguf',197                'Meta-Llama-3-70B-Instruct-Q3_K_M.gguf'198            ],199            value="Mistral-Small-24B-Instruct-2501-Q8_0.gguf",200            label="Model"201        ),202        gr.Textbox(value="You are a deep thinking AI, you may use extremely long chains of thought to deeply consider the problem and deliberate with yourself via systematic reasoning processes to help come to a correct solution prior to answering. You should enclose your thoughts and internal monologue inside <think> </think> tags, and then provide your solution or response to the problem.", label="System message"),203        gr.Slider(minimum=1, maximum=4096, value=2048, step=1, label="Max tokens"),204        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),205        gr.Slider(206            minimum=0.1,207            maximum=1.0,208            value=0.95,209            step=0.05,210            label="Top-p",211        ),212        gr.Slider(213            minimum=0,214            maximum=100,215            value=40,216            step=1,217            label="Top-k",218        ),219        gr.Slider(220            minimum=0.0,221            maximum=2.0,222            value=1.1,223            step=0.1,224            label="Repetition penalty",225        ),226    ],227    theme=gr.themes.Soft(primary_hue="violet", secondary_hue="violet", neutral_hue="gray",font=[gr.themes.GoogleFont("Exo"), "ui-sans-serif", "system-ui", "sans-serif"]).set(228        body_background_fill_dark="#16141c",229        block_background_fill_dark="#16141c",230        block_border_width="1px",231        block_title_background_fill_dark="#1e1c26",232        input_background_fill_dark="#292733",233        button_secondary_background_fill_dark="#24212b",234        border_color_accent_dark="#343140",235        border_color_primary_dark="#343140",236        background_fill_secondary_dark="#16141c",237        color_accent_soft_dark="transparent",238        code_background_fill_dark="#292733",239    ),240    css=css,241    retry_btn="Retry",242    undo_btn="Undo",243    clear_btn="Clear",244    submit_btn="Send",245    description="Llama-cpp-agent: Chat multi llm selection",246    chatbot=gr.Chatbot(247        scale=1, 248        placeholder=PLACEHOLDER,249        likeable=False,250        show_copy_button=True251    )252)253 254if __name__ == "__main__":255    demo.launch()