Markit_v2 / src /core /config.py
AnseMin's picture
Update embedding model to Google Generative AI and enhance vector store functionality
4dfec96
"""
Centralized configuration management for Markit application.
"""
import os
from typing import Optional, Dict, Any
from dataclasses import dataclass
@dataclass
class APIConfig:
"""Configuration for external API services."""
google_api_key: Optional[str] = None
openai_api_key: Optional[str] = None
mistral_api_key: Optional[str] = None
def __post_init__(self):
"""Load API keys from environment variables."""
self.google_api_key = os.getenv("GOOGLE_API_KEY")
self.openai_api_key = os.getenv("OPENAI_API_KEY")
self.mistral_api_key = os.getenv("MISTRAL_API_KEY")
@dataclass
class OCRConfig:
"""Configuration for OCR-related settings."""
tesseract_path: Optional[str] = None
tessdata_path: Optional[str] = None
default_language: str = "eng"
def __post_init__(self):
"""Load OCR configuration from environment variables."""
self.tesseract_path = os.getenv("TESSERACT_PATH")
self.tessdata_path = os.getenv("TESSDATA_PATH", "./tessdata")
@dataclass
class ModelConfig:
"""Configuration for AI model settings."""
gemini_model: str = "gemini-2.5-flash"
mistral_model: str = "mistral-ocr-latest"
got_ocr_model: str = "stepfun-ai/GOT-OCR2_0"
temperature: float = 0.1
max_tokens: int = 32768
def __post_init__(self):
"""Load model configuration from environment variables."""
self.gemini_model = os.getenv("GEMINI_MODEL", self.gemini_model)
self.mistral_model = os.getenv("MISTRAL_MODEL", self.mistral_model)
self.got_ocr_model = os.getenv("GOT_OCR_MODEL", self.got_ocr_model)
self.temperature = float(os.getenv("MODEL_TEMPERATURE", self.temperature))
self.max_tokens = int(os.getenv("MODEL_MAX_TOKENS", self.max_tokens))
@dataclass
class DoclingConfig:
"""Configuration for Docling parser."""
artifacts_path: Optional[str] = None
enable_remote_services: bool = False
enable_tables: bool = True
enable_code_enrichment: bool = False
enable_formula_enrichment: bool = False
enable_picture_classification: bool = False
generate_picture_images: bool = False
ocr_cpu_threads: int = 4
def __post_init__(self):
"""Load Docling configuration from environment variables."""
self.artifacts_path = os.getenv("DOCLING_ARTIFACTS_PATH")
self.enable_remote_services = os.getenv("DOCLING_ENABLE_REMOTE_SERVICES", "false").lower() == "true"
self.enable_tables = os.getenv("DOCLING_ENABLE_TABLES", "true").lower() == "true"
self.enable_code_enrichment = os.getenv("DOCLING_ENABLE_CODE_ENRICHMENT", "false").lower() == "true"
self.enable_formula_enrichment = os.getenv("DOCLING_ENABLE_FORMULA_ENRICHMENT", "false").lower() == "true"
self.enable_picture_classification = os.getenv("DOCLING_ENABLE_PICTURE_CLASSIFICATION", "false").lower() == "true"
self.generate_picture_images = os.getenv("DOCLING_GENERATE_PICTURE_IMAGES", "false").lower() == "true"
self.ocr_cpu_threads = int(os.getenv("OMP_NUM_THREADS", self.ocr_cpu_threads))
@dataclass
class RAGConfig:
"""Configuration for RAG (Retrieval-Augmented Generation) functionality."""
# Vector store settings
vector_store_path: str = "./data/vector_store"
collection_name: str = "markit_documents"
# Chat history settings
chat_history_path: str = "./data/chat_history"
# Embedding settings
embedding_model: str = "models/text-embedding-004"
embedding_chunk_size: int = 1000
# Chunking settings
chunk_size: int = 1000
chunk_overlap: int = 200
# Chat limits
max_messages_per_session: int = 50
max_messages_per_hour: int = 100
# Retrieval settings
retrieval_k: int = 4
retrieval_score_threshold: float = 0.5
# LLM settings for RAG
rag_model: str = "gemini-2.5-flash"
rag_temperature: float = 0.1
rag_max_tokens: int = 32768
def __post_init__(self):
"""Load RAG configuration from environment variables."""
# For HF Spaces, ensure data directories are created
if os.getenv("SPACE_ID"): # HF Spaces environment
base_data_path = "/tmp/data" if not os.access("./data", os.W_OK) else "./data"
self.vector_store_path = os.getenv("VECTOR_STORE_PATH", f"{base_data_path}/vector_store")
self.chat_history_path = os.getenv("CHAT_HISTORY_PATH", f"{base_data_path}/chat_history")
else:
self.vector_store_path = os.getenv("VECTOR_STORE_PATH", self.vector_store_path)
self.chat_history_path = os.getenv("CHAT_HISTORY_PATH", self.chat_history_path)
self.collection_name = os.getenv("VECTOR_STORE_COLLECTION", self.collection_name)
self.embedding_model = os.getenv("EMBEDDING_MODEL", self.embedding_model)
self.embedding_chunk_size = int(os.getenv("EMBEDDING_CHUNK_SIZE", self.embedding_chunk_size))
self.chunk_size = int(os.getenv("CHUNK_SIZE", self.chunk_size))
self.chunk_overlap = int(os.getenv("CHUNK_OVERLAP", self.chunk_overlap))
self.max_messages_per_session = int(os.getenv("MAX_MESSAGES_PER_SESSION", self.max_messages_per_session))
self.max_messages_per_hour = int(os.getenv("MAX_MESSAGES_PER_HOUR", self.max_messages_per_hour))
self.retrieval_k = int(os.getenv("RETRIEVAL_K", self.retrieval_k))
self.retrieval_score_threshold = float(os.getenv("RETRIEVAL_SCORE_THRESHOLD", self.retrieval_score_threshold))
self.rag_model = os.getenv("RAG_MODEL", self.rag_model)
self.rag_temperature = float(os.getenv("RAG_TEMPERATURE", self.rag_temperature))
self.rag_max_tokens = int(os.getenv("RAG_MAX_TOKENS", self.rag_max_tokens))
@dataclass
class AppConfig:
"""Main application configuration."""
debug: bool = False
max_file_size: int = 10 * 1024 * 1024 # 10MB
allowed_extensions: tuple = (".pdf", ".png", ".jpg", ".jpeg", ".tiff", ".bmp", ".webp", ".tex", ".xlsx", ".docx", ".pptx", ".html", ".xhtml", ".md", ".csv")
temp_dir: str = "./temp"
# Multi-document batch processing settings
max_batch_files: int = 5
max_batch_size: int = 20 * 1024 * 1024 # 20MB combined
batch_processing_types: tuple = ("combined", "individual", "summary", "comparison")
def __post_init__(self):
"""Load application configuration from environment variables."""
self.debug = os.getenv("DEBUG", "false").lower() == "true"
self.max_file_size = int(os.getenv("MAX_FILE_SIZE", self.max_file_size))
self.temp_dir = os.getenv("TEMP_DIR", self.temp_dir)
# Load batch processing configuration
self.max_batch_files = int(os.getenv("MAX_BATCH_FILES", self.max_batch_files))
self.max_batch_size = int(os.getenv("MAX_BATCH_SIZE", self.max_batch_size))
class Config:
"""Main configuration container."""
def __init__(self):
self.api = APIConfig()
self.ocr = OCRConfig()
self.model = ModelConfig()
self.docling = DoclingConfig()
self.app = AppConfig()
self.rag = RAGConfig()
def validate(self) -> Dict[str, Any]:
"""Validate configuration and return validation results."""
validation_results = {
"valid": True,
"warnings": [],
"errors": []
}
# Check API keys
if not self.api.google_api_key:
validation_results["warnings"].append("Google API key not found - Gemini parser will be unavailable")
if not self.api.mistral_api_key:
validation_results["warnings"].append("Mistral API key not found - Mistral parser will be unavailable")
# Check RAG dependencies
if not self.api.google_api_key:
validation_results["warnings"].append("Google API key not found - RAG embeddings will be unavailable")
if not self.api.google_api_key:
validation_results["warnings"].append("Google API key not found - RAG chat will be unavailable")
# Check tesseract setup
if not self.ocr.tesseract_path and not os.path.exists("/usr/bin/tesseract"):
validation_results["warnings"].append("Tesseract not found in system PATH - OCR functionality may be limited")
# Check temp directory
try:
os.makedirs(self.app.temp_dir, exist_ok=True)
except Exception as e:
validation_results["errors"].append(f"Cannot create temp directory {self.app.temp_dir}: {e}")
validation_results["valid"] = False
# Check RAG directories
try:
os.makedirs(self.rag.vector_store_path, exist_ok=True)
os.makedirs(self.rag.chat_history_path, exist_ok=True)
except Exception as e:
validation_results["errors"].append(f"Cannot create RAG directories: {e}")
validation_results["valid"] = False
return validation_results
def get_available_parsers(self) -> list:
"""Get list of available parsers based on current configuration."""
available = ["markitdown"] # Always available
if self.api.google_api_key:
available.append("gemini_flash")
if self.api.mistral_api_key:
available.append("mistral_ocr")
# GOT-OCR is available if we have GPU or can use ZeroGPU
available.append("got_ocr")
# Docling is available if package is installed
try:
import docling
available.append("docling")
except ImportError:
pass
return available
# Global configuration instance
config = Config()