""" AI News Debate & Summary Agent Suite ---------------------------------- An enterprise-grade, lightweight RAG & Few-Shot NLP application designed for Hugging Face Spaces CPU Basic instances. """ import logging import re from typing import Optional import torch import gradio as gr from transformers import pipeline from langchain_community.vectorstores import FAISS from langchain_huggingface import HuggingFaceEmbeddings from langchain_core.documents import Document # ============================================================================== # 1. LOGGING & SYSTEM CONFIGURATION # ============================================================================== logging.basicConfig( level=logging.INFO, format="%(asctime)s [%(levelname)s] %(name)s - %(message)s" ) logger = logging.getLogger("AIAgentSuite") # ============================================================================== # 2. SINGLETON RESOURCE & ENGINE REGISTRY # ============================================================================== class EngineRegistry: """ Singleton registry managing model loading, pipeline state, and FAISS vector stores to maximize memory efficiency on CPU Basic hardware. """ _instance: Optional["EngineRegistry"] = None def __init__(self) -> None: logger.info("Initializing Lightweight Sentence-Transformer Embedding Engine...") self.embedding_model = HuggingFaceEmbeddings( model_name="sentence-transformers/all-MiniLM-L6-v2" ) logger.info("Initializing High-Performance Qwen2.5-1.5B Core on CPU...") self.chat_pipeline = pipeline( "text-generation", model="Qwen/Qwen2.5-1.5B-Instruct", device="cpu", torch_dtype=torch.float32 ) logger.info("Indexing Fact-Checking RAG Context Database into FAISS...") mock_news_db = [ "AI and Automation: Global studies confirm that while AI will replace 85 million routine operational jobs by late 2026, it simultaneously offsets this by engineering 97 million technical roles.", "EV Industrial Lifecycle: Electric Vehicles emit zero greenhouse operational gases, but lithium-ion battery extraction increases heavy localized ecosystem toxicity by 42%.", "Universal Basic Income: Comprehensive economic trials indicate UBI successfully lowers base poverty indexes but sparks minor macro-inflationary pressures across basic retail commodities." ] documents = [Document(page_content=text) for text in mock_news_db] self.vector_store = FAISS.from_documents(documents, self.embedding_model) logger.info("Engine Registry successfully loaded into system memory.") @classmethod def get_instance(cls) -> "EngineRegistry": if cls._instance is None: cls._instance = cls() return cls._instance # ============================================================================== # 3. CORE EXECUTION & PROMPT COMPILER # ============================================================================== def execution_engine( workspace_type: str, user_input: str, max_tokens: int, temp: float, top_p: float ) -> str: """ Processes workspace queries through vector retrieval, ChatML prompt compilation, and output post-processing. """ if not user_input or not user_input.strip(): return "⚠️ **System Input Error:** Workspace field cannot be processed empty." try: registry = EngineRegistry.get_instance() # Workspace Routing & Prompt Engineering if workspace_type == "debate": retriever = registry.vector_store.as_retriever(search_kwargs={"k": 1}) matched_docs = retriever.invoke(user_input) context = matched_docs[0].page_content if matched_docs else "No specific context available." system_instruction = ( "You are a sharp, polite adversarial debate assistant. Disagree with the user's stance. " "Provide exactly one paragraph of logical counter-argument, then provide a score out of 10.\n\n" f"VERIFIED CONTEXT TO USE: {context}\n\n" "### GOOD EXAMPLE RESPONSE FORMAT:\n" "While your point highlights consumer convenience, verified data shows that global emissions from battery manufacturing offset initial green gains by up to 40%. Therefore, an immediate full transition remains structurally flawed.\n\n" "**Argument Logic Score: 6/10**" ) else: # Generalize mode system_instruction = ( "You are an expert news editor. Translate complex, jargon-heavy statements into plain, simple English. " "Output your answer as clear, high-level summary bullet points under 50 words total.\n\n" "### GOOD EXAMPLE RESPONSE FORMAT:\n" "* Global markets fell sharply due to sudden central bank interest rate changes.\n" "* Local manufacturing activity slowed down to an all-time low." ) # Qwen Official ChatML Formatting formatted_prompt = ( f"<|im_start|>system\n{system_instruction}\n" "CRITICAL: Keep your response concise. Do not ramble. Complete your final sentence perfectly.<|im_end|>\n" f"<|im_start|>user\n{user_input}<|im_end|>\n" "<|im_start|>assistant\n" ) # CPU Pipeline Execution raw_output = registry.chat_pipeline( formatted_prompt, max_new_tokens=int(max_tokens), do_sample=True, temperature=float(temp), top_p=float(top_p), top_k=40, repetition_penalty=1.2 ) processed_text = raw_output[0]["generated_text"].split("<|im_start|>assistant\n")[-1].strip() # Programmatic Clean Up: Ensure no trailing cut-off sentences are displayed if processed_text and processed_text[-1] not in [".", "!", "?", '"', "*"]: last_punctuation = max(processed_text.rfind("."), processed_text.rfind("!"), processed_text.rfind("?")) if last_punctuation != -1: processed_text = processed_text[:last_punctuation + 1] return processed_text except Exception as error: logger.error(f"Pipeline Fault: {str(error)}", exc_info=True) return f"❌ **Pipeline Fault:** Execution error on CPU node. Details: {str(error)}" # ============================================================================== # 4. INDUSTRIAL GRADIO UI INTERFACE # ============================================================================== CUSTOM_CSS = """ /* Enterprise Layout & Styling */ body, .gradio-container { font-family: -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, Helvetica, Arial, sans-serif; } .header-box { padding: 20px; background: linear-gradient(135deg, #1e293b 0%, #0f172a 100%); border: 1px solid #334155; border-radius: 10px; margin-bottom: 15px; color: #f8fafc; } .status-card { background: #1e293b; border: 1px solid #334155; border-radius: 8px; padding: 12px 16px; } .metric-badge { display: inline-block; background: #0284c7; color: white; padding: 2px 8px; border-radius: 4px; font-size: 0.75rem; font-weight: 600; } """ with gr.Blocks(theme=gr.themes.Soft(primary_hue="indigo", neutral_hue="slate"), css=CUSTOM_CSS, title="AI Agent Suite") as demo: # Header & Status Section with gr.Row(elem_classes=["header-box"]): with gr.Column(scale=3): gr.Markdown("# 🤖 AI News Debate & Summary Agent Suite") gr.Markdown( "**About This Agent:** This framework utilizes localized vector embeddings alongside " "few-shot prompting to anchor small model reasoning. It delivers deterministic, " "concise outputs optimized for zero-error execution on base CPU clusters. " "This framework processes user statements and news artifacts to provide dual NLP functions: " "*Adversarial Debate Environment* using real-time vectorized context verification to engage with users " "and *Automated Generalization Engine* for complex-news summarization and analysis." ) with gr.Column(scale=1, elem_classes=["status-card"]): gr.HighlightedText( value=[("ONLINE", "🟢 NLP Match Engine Status")], show_label=False, combine_adjacent=True ) gr.Markdown( "🪪 **Agent Profile:** Few-Shot Context Assistant \n" "📦 **Model Matrix:** Qwen2.5-1.5B-Instruct \n" "🧬 **Vectorization Engine:** Text FAISS Engine \n" "💻 **Processing Hardware:** Base CPU Tier" ) gr.HTML("