Update config/settings.py

config/settings.py

@@ -6,35 +6,81 @@ load_dotenv()
 class Settings:
     GROQ_API_KEY = os.getenv("GROQ_API_KEY")
     
-    # Multilingual Model Settings - SỬA: Dùng models nhẹ và phổ biến
-    VIETNAMESE_EMBEDDING_MODEL = 'sentence-transformers/all-MiniLM-L6-v2'  # Model nhẹ, hỗ trợ nhiều ngôn ngữ
-    VIETNAMESE_LLM_MODEL = "llama-3.1-8b-instant"  # Model Groq
+    # Multilingual Model Settings
+    VIETNAMESE_EMBEDDING_MODEL = 'dangvantuan/vietnamese-embedding'
+    VIETNAMESE_LLM_MODEL = "Vietnamese_LLaMA2_13B_8K_SFT_General_Domain_Knowledge"
     
-    MULTILINGUAL_EMBEDDING_MODEL = 'sentence-transformers/all-MiniLM-L6-v2'  # Dùng model chung
-    MULTILINGUAL_LLM_MODEL = "llama-3.1-8b-instant"  # Model Groq
+    MULTILINGUAL_EMBEDDING_MODEL = 'Qwen/Qwen3-Embedding-4B'
+    MULTILINGUAL_LLM_MODEL = "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 = 'dangvantuan/vietnamese-embedding'
+    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-turbo"
+    
+    # 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
     
-    # Fallback models
-    FALLBACK_EMBEDDING_MODEL = 'sentence-transformers/all-MiniLM-L6-v2'
+    # 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()"
+    
+    # Fallback models in case primary models fail
+    FALLBACK_MULTILINGUAL_EMBEDDING_MODEL = 'sentence-transformers/all-MiniLM-L6-v2'
     
-    # Default models
-    DEFAULT_EMBEDDING_MODEL = 'sentence-transformers/all-MiniLM-L6-v2'
-    DEFAULT_LLM_MODEL = "llama-3.1-8b-instant"
+    # 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-turbo"
     
     # TTS Settings
     MAX_CHUNK_LENGTH = 200
     SUPPORTED_LANGUAGES = {
-        'vi': 'Vietnamese', 'en': 'English', 'fr': 'French', 'es': 'Spanish', 
-        'de': 'German', 'ja': 'Japanese', 'ko': 'Korean', 'zh': 'Chinese'
+        'vi': 'vi', 'en': 'en', 'fr': 'fr', 'es': 'es', 
+        'de': 'de', 'ja': 'ja', 'ko': 'ko', 'zh': 'zh'
     }
     
     # RAG Settings
-    EMBEDDING_DIMENSION = 384  # Dimension của all-MiniLM-L6-v2
+    EMBEDDING_DIMENSION = 768  # For Vietnamese model
+    MULTILINGUAL_EMBEDDING_DIMENSION = 4096  # For Nemotron model
+    
     TOP_K_RESULTS = 3
     
-    # SpeechBrain VAD Settings - TẠM TẮT để tránh lỗi
-    VAD_MODEL = None  # Tạm tắt VAD
+    # 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