# Corrected RAG System Configuration
# This config works with the corrected code implementation

# API Configuration
apis:
  siliconflow:
    api_key: "${SILICONFLOW_API_KEY}"
    base_url: "https://api.siliconflow.com/v1"
  groq:
    api_key: "${GROQ_API_KEY}" 
    base_url: "https://api.groq.com/openai/v1"
    
# Model Configuration
models:
  # SiliconFlow Models
  embedding_model: "Qwen/Qwen3-Embedding-8B"
  reranker_model: "Qwen/Qwen3-Reranker-8B"
  
  # Groq Models (OpenAI compatible)
  llm_model: "openai/gpt-oss-120b"  # Primary model
  alternative_models:
    - "openai/gpt-oss-20b"           # Faster, less capable
    - "qwen/qwen3-32b"             # Good balance
    
# RAG System Configuration
rag_system:
  # Chunking settings
  chunk_size: 512
  chunk_overlap: 50
  max_context_chunks: 5
  


  # Retrieval settings
  similarity_threshold: 0.7
  rerank_top_k: 20
  final_top_k: 5
  
  # Generation settings
  temperature: 0.1
  max_tokens: 4096
  
  # Performance settings
  batch_size: 32
  max_retries: 3
  enable_caching: true

# Text Processing Configuration
text_processing:
  # Text splitting
  separators: ["\n\n", "\n", " ", ""]
  
  # Language settings
  language: "korean"
  encoding: "utf-8"

# Vector Store Configuration (if using)
vector_store:
  provider: "qdrant"
  collection_name: "manufacturing_docs"
  vector_size: 1024
  distance_metric: "cosine"
  
  # Qdrant specific
  qdrant_url: "${QDRANT_URL}"
  qdrant_api_key: "${QDRANT_API_KEY}"

# Document Processing Configuration
document_processing:
  # Supported formats
  supported_formats: [".pdf", ".xlsx", ".xls", ".txt", ".md", ".png", ".jpg", ".jpeg"]
  
  # File size limits
  max_file_size_mb: 100
  
  # Processing options
  extract_images: true
  extract_tables: true
  ocr_enabled: true
  
  # OCR settings
  ocr_language: "kor+eng"
  ocr_confidence_threshold: 0.6

# Logging Configuration
logging:
  level: "INFO"
  format: "%(asctime)s - %(name)s - %(levelname)s - %(message)s"
  file_logging: true
  log_file: "logs/rag_system.log"
  max_file_size_mb: 10
  backup_count: 5

# Performance Configuration
performance:
  # Rate limiting
  siliconflow_requests_per_minute: 60
  groq_requests_per_minute: 30
  api_timeout: 60  # Increased timeout for large documents
  max_retries: 3
  retry_delay: 5
  batch_size: 32  # Consider reducing this for large documents
  api_timeout_seconds: 30
  max_workers: 4
  


# Cache Configuration
cache:
  enabled: true
  type: "memory"  # Options: memory, redis, file
  ttl_seconds: 3600
  max_size: 1000

# Evaluation Configuration
evaluation:
  metrics: ["accuracy", "relevance", "coherence"]
  threshold: 0.8
  save_results: true
  results_dir: "evaluation_results"

# Development/Debug Configuration
debug:
  verbose_logging: false
  save_intermediate_results: false
  profiling_enabled: false