import os import logging import gradio as gr import json from typing import List from datetime import datetime, timezone from pydantic import BaseModel, Field from trafilatura import fetch_url, extract from langchain import HuggingFaceHub from llama_cpp_agent import MessagesFormatterType from llama_cpp_agent.chat_history import BasicChatHistory from llama_cpp_agent.chat_history.messages import Roles from llama_cpp_agent.llm_output_settings import ( LlmStructuredOutputSettings, LlmStructuredOutputType, ) from llama_cpp_agent.tools import WebSearchTool from llama_cpp_agent.prompt_templates import web_search_system_prompt, research_system_prompt # UI related imports and definitions css = """ .message-row { justify-content: space-evenly !important; } .message-bubble-border { border-radius: 6px !important; } .message-buttons-bot, .message-buttons-user { right: 10px !important; left: auto !important; bottom: 2px !important; } .dark.message-bubble-border { border-color: #1b0f0f !important; } .dark.user { background: #140b0b !important; } .dark.assistant.dark, .dark.pending.dark { background: #0c0505 !important; } """ PLACEHOLDER = """
Logo

llama-cpp-agent

DDG Agent allows users to interact with it using natural language, making it easier for them to find the information they need. Offers a convenient and secure way for users to access web-based information.

Mistral 7B Instruct v0.3 Mixtral 8x7B Instruct v0.1 Meta Llama 3 8B Instruct
Discord GitHub
""" # Global variables huggingface_token = os.environ.get("HUGGINGFACE_TOKEN") # Example queries examples = [ ["latest news about Yann LeCun"], ["Latest news site:github.blog"], ["Where I can find best hotel in Galapagos, Ecuador intitle:hotel"], ["filetype:pdf intitle:python"] ] # Utility functions def get_server_time(): utc_time = datetime.now(timezone.utc) return utc_time.strftime("%Y-%m-%d %H:%M:%S") def get_website_content_from_url(url: str) -> str: try: downloaded = fetch_url(url) result = extract(downloaded, include_formatting=True, include_links=True, output_format='json', url=url) if result: result = json.loads(result) return f'=========== Website Title: {result["title"]} ===========\n\n=========== Website URL: {url} ===========\n\n=========== Website Content ===========\n\n{result["raw_text"]}\n\n=========== Website Content End ===========\n\n' else: return "" except Exception as e: return f"An error occurred: {str(e)}" class CitingSources(BaseModel): sources: List[str] = Field( ..., description="List of sources to cite. Should be an URL of the source. E.g. GitHub URL, Blogpost URL or Newsletter URL." ) # Model function def get_model(temperature, top_p, repetition_penalty): return HuggingFaceHub( repo_id="mistralai/Mistral-7B-Instruct-v0.3", model_kwargs={ "temperature": temperature, "top_p": top_p, "repetition_penalty": repetition_penalty, "max_length": 1000 }, huggingfacehub_api_token=huggingface_token ) def get_messages_formatter_type(model_name): model_name = model_name.lower() if any(keyword in model_name for keyword in ["meta", "aya"]): return MessagesFormatterType.LLAMA_3 elif any(keyword in model_name for keyword in ["mistral", "mixtral"]): return MessagesFormatterType.MISTRAL elif any(keyword in model_name for keyword in ["einstein", "dolphin"]): return MessagesFormatterType.CHATML elif "phi" in model_name: return MessagesFormatterType.PHI_3 else: return MessagesFormatterType.CHATML # Main response function def respond( message, history: list[tuple[str, str]], system_message, max_tokens, temperature, top_p, repeat_penalty, ): model = get_model(temperature, top_p, repeat_penalty) chat_template = MessagesFormatterType.MISTRAL search_tool = WebSearchTool( llm_provider=model, message_formatter_type=chat_template, max_tokens_search_results=12000, max_tokens_per_summary=2048, ) messages = BasicChatHistory() for msn in history: user = {"role": Roles.user, "content": msn[0]} assistant = {"role": Roles.assistant, "content": msn[1]} messages.add_message(user) messages.add_message(assistant) # Perform web search search_result = search_tool.run(message) outputs = "" # Generate response response_prompt = f"""Write a detailed and complete research document that fulfills the following user request: '{message}', based on the information from the web below. {search_result} Respond in a clear and concise manner, citing sources where appropriate.""" response = model(response_prompt) outputs += response # Generate citations citation_prompt = "Cite the sources you used in your response." citing_sources = model(citation_prompt) outputs += "\n\nSources:\n" outputs += citing_sources return outputs # Gradio interface demo = gr.ChatInterface( respond, additional_inputs=[ gr.Dropdown([ 'Mistral-7B-Instruct-v0.3-Q6_K.gguf', 'mixtral-8x7b-instruct-v0.1.Q5_K_M.gguf', 'Meta-Llama-3-8B-Instruct-Q6_K.gguf' ], value="Mistral-7B-Instruct-v0.3-Q6_K.gguf", label="Model" ), gr.Textbox(value=web_search_system_prompt, label="System message"), gr.Slider(minimum=1, maximum=4096, value=2048, step=1, label="Max tokens"), gr.Slider(minimum=0.1, maximum=1.0, value=0.45, step=0.1, label="Temperature"), gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p"), gr.Slider(minimum=0, maximum=100, value=40, step=1, label="Top-k"), gr.Slider(minimum=0.0, maximum=2.0, value=1.1, step=0.1, label="Repetition penalty"), ], theme=gr.themes.Soft( primary_hue="orange", secondary_hue="amber", neutral_hue="gray", font=[gr.themes.GoogleFont("Exo"), "ui-sans-serif", "system-ui", "sans-serif"]).set( body_background_fill_dark="#0c0505", block_background_fill_dark="#0c0505", block_border_width="1px", block_title_background_fill_dark="#1b0f0f", input_background_fill_dark="#140b0b", button_secondary_background_fill_dark="#140b0b", border_color_accent_dark="#1b0f0f", border_color_primary_dark="#1b0f0f", background_fill_secondary_dark="#0c0505", color_accent_soft_dark="transparent", code_background_fill_dark="#140b0b" ), css=css, retry_btn="Retry", undo_btn="Undo", clear_btn="Clear", submit_btn="Send", cache_examples=False, examples=examples, description="Llama-cpp-agent: Chat with DuckDuckGo Agent", analytics_enabled=False, chatbot=gr.Chatbot( scale=1, placeholder=PLACEHOLDER, show_copy_button=True ) ) if __name__ == "__main__": demo.launch()