Lumpen1/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 12hf_hub_download(13    repo_id="bartowski/Meta-Llama-3-70B-Instruct-GGUF",14    filename="Meta-Llama-3-70B-Instruct-Q3_K_M.gguf",15    local_dir = "./models"16)17hf_hub_download(18    repo_id="bartowski/Mistral-7B-Instruct-v0.3-GGUF",19    filename="Mistral-7B-Instruct-v0.3-f32.gguf",20    local_dir = "./models"21)22 23css = """24.message-row {25    justify-content: space-evenly !important;26}27.message-bubble-border {28    border-radius: 6px !important;29}30.dark.message-bubble-border {31    border-color: #343140 !important;32}33.dark.user {34    background: #1e1c26 !important;35}36.dark.assistant.dark, .dark.pending.dark {37    background: #16141c !important;38}39"""40 41def get_messages_formatter_type(model_name):42    if "Llama" in model_name:43        return MessagesFormatterType.LLAMA_344    elif "Mistral" in model_name:45        return MessagesFormatterType.MISTRAL46    else:47        raise ValueError(f"Unsupported model: {model_name}")48 49@spaces.GPU(duration=120)50def respond(51    message,52    history: list[tuple[str, str]],53    model,54    system_message,55    max_tokens,56    temperature,57    top_p,58    top_k,59    repeat_penalty,60):61    chat_template = get_messages_formatter_type(model)62 63    llm = Llama(64        model_path=f"models/{model}",65        flash_attn=True,66        n_threads=40,67        n_gpu_layers=81,68        n_batch=1024,69        n_ctx=8192,70    )71    provider = LlamaCppPythonProvider(llm)72 73    agent = LlamaCppAgent(74        provider,75        system_prompt=f"{system_message}",76        predefined_messages_formatter_type=chat_template,77        debug_output=True78    )79    80    settings = provider.get_provider_default_settings()81    settings.temperature = temperature82    settings.top_k = top_k83    settings.top_p = top_p84    settings.max_tokens = max_tokens85    settings.repeat_penalty = repeat_penalty86    settings.stream = True87 88    messages = BasicChatHistory()89 90    for msn in history:91        user = {92            'role': Roles.user,93            'content': msn[0]94        }95        assistant = {96            'role': Roles.assistant,97            'content': msn[1]98        }99        messages.add_message(user)100        messages.add_message(assistant)101    102    stream = agent.get_chat_response(103        message,104        llm_sampling_settings=settings,105        chat_history=messages,106        returns_streaming_generator=True,107        print_output=False108    )109    110    outputs = ""111    for output in stream:112        outputs += output113        yield outputs114 115PLACEHOLDER = """116<div class="message-bubble-border" style="display:flex; max-width: 600px; border-radius: 6px; border-width: 1px; box-shadow: 0 4px 6px rgba(0, 0, 0, 0.1); backdrop-filter: blur(10px);">117    <figure style="margin: 0;">118        <img src="https://huggingface.co/spaces/pabloce/llama-cpp-agent/resolve/main/llama.jpg" alt="Logo" style="width: 100%; height: 100%; border-radius: 8px;">119    </figure>120    <div style="padding: .5rem 1.5rem;">121        <h2 style="text-align: left; font-size: 1.5rem; font-weight: 700; margin-bottom: 0.5rem;">llama-cpp-agent</h2>122        <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>123        <div style="display: flex; justify-content: space-between; align-items: center;">124            <div style="display: flex; flex-flow: column; justify-content: space-between;">125                <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;">126                    Mistral 7B Instruct v0.3127                </span>128                <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;">129                    Meta Llama 3 70B Instruct130                </span>131            </div>132            <div style="display: flex; justify-content: flex-end; align-items: center;">133                <a href="https://discord.gg/fgr5RycPFP" target="_blank" rel="noreferrer" style="padding: .5rem;">134                    <svg width="24" height="24" fill="currentColor" xmlns="http://www.w3.org/2000/svg" viewBox="0 5 30.67 23.25">135                        <title>Discord</title>136                        <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>137                    </svg>138                </a>139                <a href="https://github.com/Maximilian-Winter/llama-cpp-agent" target="_blank" rel="noreferrer" style="padding: .5rem;">140                    <svg width="24" height="24" fill="currentColor" viewBox="3 3 18 18">141                        <title>GitHub</title>142                        <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>143                    </svg>144                </a>145            </div>146        </div>147    </div>148</div>149"""150 151demo = gr.ChatInterface(152    respond,153    additional_inputs=[154        gr.Dropdown([155                'Meta-Llama-3-70B-Instruct-Q3_K_M.gguf',156                'Mistral-7B-Instruct-v0.3-f32.gguf'157            ],158            value="Meta-Llama-3-70B-Instruct-Q3_K_M.gguf",159            label="Model"160        ),161        gr.Textbox(value="You are a helpful assistant.", label="System message"),162        gr.Slider(minimum=1, maximum=4096, value=2048, step=1, label="Max tokens"),163        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),164        gr.Slider(165            minimum=0.1,166            maximum=1.0,167            value=0.95,168            step=0.05,169            label="Top-p",170        ),171        gr.Slider(172            minimum=0,173            maximum=100,174            value=40,175            step=1,176            label="Top-k",177        ),178        gr.Slider(179            minimum=0.0,180            maximum=2.0,181            value=1.1,182            step=0.1,183            label="Repetition penalty",184        ),185    ],186    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(187        body_background_fill_dark="#16141c",188        block_background_fill_dark="#16141c",189        block_border_width="1px",190        block_title_background_fill_dark="#1e1c26",191        input_background_fill_dark="#292733",192        button_secondary_background_fill_dark="#24212b",193        border_color_primary_dark="#343140",194        background_fill_secondary_dark="#16141c",195        color_accent_soft_dark="transparent"196    ),197    css=css,198    retry_btn="Retry",199    undo_btn="Undo",200    clear_btn="Clear",201    submit_btn="Send",202    description="Llama-cpp-agent: Chat multi llm selection",203    chatbot=gr.Chatbot(scale=1, placeholder=PLACEHOLDER)204)205 206if __name__ == "__main__":207    demo.launch()