Neerajkadari/Context-Aware_Conversational_Intelligence_System

1"""2Global Configuration for the Conversational Intelligence System.3Centralizes all paths, hyperparameters, and model settings.4"""5 6import os7from pathlib import Path8 9# ──────────────────────────────────────────────────────────────10# PATHS11# ──────────────────────────────────────────────────────────────12BASE_DIR = Path(__file__).parent.resolve()13DATA_DIR = BASE_DIR / "data"14MODEL_DIR = BASE_DIR / "models"15REPORT_DIR = BASE_DIR / "reports"16VECTOR_DB_DIR = BASE_DIR / "vector_store"17LOG_DIR = BASE_DIR / "logs"18 19# Create directories20for d in [DATA_DIR, MODEL_DIR, REPORT_DIR, VECTOR_DB_DIR, LOG_DIR]:21    d.mkdir(parents=True, exist_ok=True)22 23# ──────────────────────────────────────────────────────────────24# DATASET25# ──────────────────────────────────────────────────────────────26DATASET_NAME = "ag_news"27NUM_CLASSES = 428CLASS_NAMES = ["World", "Sports", "Business", "Sci/Tech"]29MAX_SAMPLES_TRAIN = 200        # Small sample for fast automatic demo30MAX_SAMPLES_TEST = 5031 32# ──────────────────────────────────────────────────────────────33# MODEL CONFIGURATIONS34# ──────────────────────────────────────────────────────────────35MODELS = {36    "bert": {37        "name": "bert-base-uncased",38        "architecture": "Bidirectional Transformer Encoder",39        "description": "Reads text in both directions simultaneously for strong contextual understanding",40        "max_length": 128,41        "batch_size": 16,42        "learning_rate": 2e-5,43        "epochs": 3,44        "weight_decay": 0.01,45    },46    "gpt2": {47        "name": "gpt2",48        "architecture": "Autoregressive Transformer Decoder",49        "description": "Token-by-token text generation with causal attention mechanism",50        "max_length": 128,51        "batch_size": 8,52        "learning_rate": 5e-5,53        "epochs": 1,54        "weight_decay": 0.01,55    },56    "llama": {57        "name": "TinyLlama/TinyLlama-1.1B-Chat-v1.0",58        "architecture": "Decoder-based Large Language Model",59        "description": "Efficient transformer architecture with strong reasoning ability",60        "max_length": 128,61        "batch_size": 4,62        "learning_rate": 2e-5,63        "epochs": 1,64        "weight_decay": 0.01,65        "use_lora": True,66        "lora_r": 8,67        "lora_alpha": 16,68        "lora_dropout": 0.1,69    },70    "mistral": {71        "name": "mistralai/Mistral-7B-v0.1",72        "architecture": "Transformer with Sliding Window Attention",73        "description": "Efficient long-context processing with fast inference",74        "max_length": 128,75        "batch_size": 2,76        "learning_rate": 2e-5,77        "epochs": 1,78        "weight_decay": 0.01,79        "use_lora": True,80        "lora_r": 8,81        "lora_alpha": 16,82        "lora_dropout": 0.1,83        "local_training": False,84    },85    "xlnet": {86        "name": "xlnet-base-cased",87        "architecture": "Permutation-based Autoregressive Transformer",88        "description": "Combines autoregressive and bidirectional modeling for improved contextual understanding",89        "max_length": 128,90        "batch_size": 16,91        "learning_rate": 2e-5,92        "epochs": 1,93        "weight_decay": 0.01,94    },95}96 97# ──────────────────────────────────────────────────────────────98# EMBEDDING / RAG99# ──────────────────────────────────────────────────────────────100EMBEDDING_MODEL = "all-MiniLM-L6-v2"101CHROMA_COLLECTION = "project_knowledge"102CHUNK_SIZE = 500103CHUNK_OVERLAP = 50104TOP_K_RESULTS = 5105 106# ──────────────────────────────────────────────────────────────107# CONVERSATIONAL AI108# ──────────────────────────────────────────────────────────────109GROQ_MODEL = "llama-3.1-8b-instant"110MAX_RESPONSE_TOKENS = 1000111TEMPERATURE = 0.3112 113# ──────────────────────────────────────────────────────────────114# SERVER115# ──────────────────────────────────────────────────────────────116HOST = os.environ.get("HOST", "0.0.0.0")117PORT = int(os.environ.get("PORT", "8509"))118DEBUG = True119# Model Download from HF Hub120import shutil121import logging122 123logger = logging.getLogger(__name__)124 125def download_models_if_needed():126    """Download models from Neerajkadari/my_project_models if not present locally."""127    from huggingface_hub import snapshot_download128    129    models_needed = ["bert", "gpt2", "llama", "xlnet"]130    any_missing = any(131        not (MODEL_DIR / m).exists() 132        for m in models_needed133    )134    if any_missing:135        logger.info("Downloading models from Neerajkadari/my_project_models...")136        local_path = snapshot_download(137            repo_id="Neerajkadari/my_project_models",138            repo_type="model",139            local_dir=str(BASE_DIR / "downloaded_models"),140            force_download=True141        )142        # Move models into expected location143        src = Path(local_path) / "models"144        for model_name in models_needed:145            src_model = src / model_name146            dst_model = MODEL_DIR / model_name147            if src_model.exists() and not dst_model.exists():148                shutil.copytree(str(src_model), str(dst_model))149                logger.info(f"Model '{model_name}' placed at {dst_model}")150            elif dst_model.exists():151                logger.info(f"Model '{model_name}' already exists at {dst_model}")152        logger.info("Model download complete!")153 154# Auto-download models on import155if not any((MODEL_DIR / m).exists() for m in ["bert", "gpt2", "llama", "xlnet"]):156    download_models_if_needed()