[UPDATE]: add configuration model

Author: fahmiaziz98

src/config/settings.py

@@ -28,35 +28,35 @@ class Settings(BaseSettings):
     HOST: str = "0.0.0.0"
     PORT: int = 7860
     WORKERS: int = 1
-    RELOAD: bool = False  # Auto-reload on code changes (dev only)
+    RELOAD: bool = False  
 
     # Model Configuration
     MODEL_CONFIG_PATH: str = "src/config/models.yaml"
     MODEL_CACHE_DIR: str = "./model_cache"
-    PRELOAD_MODELS: bool = True  # Load all models at startup
+    PRELOAD_MODELS: bool = True  
 
     # Request Limits
-    MAX_TEXT_LENGTH: int = 32000  # Maximum characters per text
-    MAX_BATCH_SIZE: int = 100  # Maximum texts per batch request
-    REQUEST_TIMEOUT: int = 30  # Request timeout in seconds
+    MAX_TEXT_LENGTH: int = 32000  
+    MAX_BATCH_SIZE: int = 100  
+    REQUEST_TIMEOUT: int = 30  
 
     # Logging
     LOG_LEVEL: str = "INFO"  # DEBUG, INFO, WARNING, ERROR, CRITICAL
-    LOG_FILE: bool = False  # Write logs to file
+    LOG_FILE: bool = True  # Write logs to file
     LOG_DIR: str = "logs"
 
-    # CORS (if needed for web frontends)
     CORS_ENABLED: bool = False
     CORS_ORIGINS: list[str] = ["*"]
 
     # Model Settings
+    DEVICE: str = "cpu"  # "cpu" or "cuda
     TRUST_REMOTE_CODE: bool = True  # For models requiring remote code
 
     model_config = SettingsConfigDict(
         env_file=".env",
         env_file_encoding="utf-8",
         case_sensitive=True,
-        extra="ignore",  # Ignore extra fields in .env
+        extra="ignore", 
     )
 
     @property
@@ -86,10 +86,8 @@ def validate_paths(self) -> None:
                 f"Model configuration file not found: {self.MODEL_CONFIG_PATH}"
             )
 
-        # Create cache directory if it doesn't exist
         Path(self.MODEL_CACHE_DIR).mkdir(parents=True, exist_ok=True)
 
-        # Create log directory if logging to file
         if self.LOG_FILE:
             Path(self.LOG_DIR).mkdir(parents=True, exist_ok=True)
 

src/models/embeddings/dense.py

@@ -9,6 +9,7 @@
 from sentence_transformers import SentenceTransformer
 from loguru import logger
 
+from src.config.settings import get_settings
 from src.core.base import BaseEmbeddingModel
 from src.core.config import ModelConfig
 from src.core.exceptions import ModelLoadError, EmbeddingGenerationError
@@ -36,6 +37,7 @@ def __init__(self, config: ModelConfig):
         """
         super().__init__(config)
         self.model: Optional[SentenceTransformer] = None
+        self.settings = get_settings()
 
     def load(self) -> None:
         """
@@ -52,7 +54,9 @@ def load(self) -> None:
 
         try:
             self.model = SentenceTransformer(
-                self.config.name, device="cpu", trust_remote_code=True
+                self.config.name, 
+                device=self.settings.DEVICE, 
+                trust_remote_code=self.settings.TRUST_REMOTE_CODE
             )
             self._loaded = True
             logger.success(f"✓ Loaded dense model: {self.model_id}")

src/models/embeddings/sparse.py

@@ -9,6 +9,7 @@
 from sentence_transformers import SparseEncoder
 from loguru import logger
 
+from src.config.settings import get_settings
 from src.core.base import BaseEmbeddingModel
 from src.core.config import ModelConfig
 from src.core.exceptions import ModelLoadError, EmbeddingGenerationError
@@ -36,6 +37,7 @@ def __init__(self, config: ModelConfig):
         """
         super().__init__(config)
         self.model: Optional[SparseEncoder] = None
+        self.settings = get_settings()
 
     def load(self) -> None:
         """
@@ -51,7 +53,11 @@ def load(self) -> None:
         logger.info(f"Loading sparse embedding model: {self.config.name}")
 
         try:
-            self.model = SparseEncoder(self.config.name)
+            self.model = SparseEncoder(
+                self.config.name,
+                device=self.settings.DEVICE,
+                trust_remote_code=self.settings.TRUST_REMOTE_CODE,
+            )
             self._loaded = True
             logger.success(f"✓ Loaded sparse model: {self.model_id}")