"""Configuration module for the Document Question Answering (DQA) system.""" import os from dataclasses import dataclass, field from typing import Optional, List @dataclass class DQAConfig: # Model settings # Default to SOTA MTEB model: BAAI/bge-small-en-v1.5 (or sentence-transformers/all-MiniLM-L6-v2) embedding_model_name: str = "BAAI/bge-small-en-v1.5" cross_encoder_model_name: str = "cross-encoder/ms-marco-MiniLM-L-6-v2" # LLM Settings llm_provider: str = "auto" # 'hf', 'openai', 'offline', or 'auto' # Hugging Face Settings hf_token: Optional[str] = field(default_factory=lambda: os.getenv("HF_TOKEN") or os.getenv("HUGGINGFACEHUB_API_TOKEN")) hf_model: str = "Qwen/Qwen2.5-7B-Instruct" # or 'meta-llama/Llama-3.2-3B-Instruct', 'mistralai/Mistral-7B-Instruct-v0.3' # OpenAI / Compatible Settings openai_model: str = "gpt-4o-mini" openai_api_key: Optional[str] = field(default_factory=lambda: os.getenv("OPENAI_API_KEY")) openai_base_url: Optional[str] = field(default_factory=lambda: os.getenv("OPENAI_BASE_URL")) temperature: float = 0.0 max_tokens: int = 1024 # Ingestion & Chunking chunk_size_tokens: int = 400 chunk_overlap_tokens: int = 50 min_chunk_tokens: int = 40 ocr_fallback_enabled: bool = True ocr_min_chars_per_page: int = 40 # Retrieval initial_top_k: int = 20 rerank_top_k: int = 5 rrf_k: int = 60 dense_weight: float = 0.5 sparse_weight: float = 0.5 # Verification verification_threshold: float = 0.55 flag_unsupported_claims: bool = True # Storage storage_dir: str = os.path.join(os.path.dirname(os.path.abspath(__file__)), "storage") index_file_name: str = "faiss_index.bin" metadata_file_name: str = "metadata.json" default_config = DQAConfig()