config/settings.py

import os
from dotenv import load_dotenv

load_dotenv()



class Settings:
    GROQ_API_KEY = os.getenv("GROQ_API_KEY")
    
    # Multilingual Model Settings
    VIETNAMESE_EMBEDDING_MODEL = 'keepitreal/vietnamese-sbert'
    VIETNAMESE_LLM_MODEL = "llama-3.1-8b-instant"
    
    MULTILINGUAL_EMBEDDING_MODEL = 'Qwen/Qwen3-Embedding-0.6B'
    MULTILINGUAL_LLM_MODEL = "llama-3.1-8b-instant"
    
    # Fallback models in case primary models fail
    FALLBACK_MULTILINGUAL_EMBEDDING_MODEL = 'sentence-transformers/all-MiniLM-L6-v2'
    
    # Default models (fallback)
    DEFAULT_EMBEDDING_MODEL = 'dangvantuan/vietnamese-embedding'
    DEFAULT_LLM_MODEL = "Vietnamese_LLaMA2_13B_8K_SFT_General_Domain_Knowledge"
    
    WHISPER_MODEL = "whisper-large-v3"
    
    # TTS Settings
    MAX_CHUNK_LENGTH = 200
    SUPPORTED_LANGUAGES = {
        'vi': 'vi', 'en': 'en', 'fr': 'fr', 'es': 'es', 
        'de': 'de', 'ja': 'ja', 'ko': 'ko', 'zh': 'zh'
    }
    
    # RAG Settings
    EMBEDDING_DIMENSION = 768  # For Vietnamese model
    MULTILINGUAL_EMBEDDING_DIMENSION = 4096  # For Nemotron model
    
    TOP_K_RESULTS = 3
    
    # SpeechBrain VAD Settings
    VAD_MODEL = "speechbrain/vad-crdnn-libriparty"
    VAD_THRESHOLD = 0.5
    VAD_MIN_SILENCE_DURATION = 0.5
    VAD_SPEECH_PAD_DURATION = 0.1
    SAMPLE_RATE = 16000

settings = Settings()