changed to uvicorn setup for HF v17

main/schemas.py

@@ -1,5 +1,5 @@
 from pydantic import BaseModel, Field
-from typing import List, Optional, Dict, Union, Literal
+from typing import List, Optional, Dict, Union
 from time import time
 
 class ChatMessage(BaseModel):
@@ -65,13 +65,26 @@ class EmbeddingResponse(BaseModel):
     embedding: List[float]
     dimension: int
 
+class ModelStatus(BaseModel):
+    is_loaded: bool
+    current_model: Optional[str]
+    has_chat_template: Optional[bool] = None
+
+class EmbeddingModelStatus(BaseModel):
+    is_loaded: bool
+    current_model: Optional[str]
+
+class ModelStatusInfo(BaseModel):
+    generation_model: ModelStatus
+    embedding_model: EmbeddingModelStatus
+
 class SystemStatusResponse(BaseModel):
     """Pydantic model for system status response"""
     cpu: Optional[Dict[str, Union[float, str]]] = None
     memory: Optional[Dict[str, Union[float, str]]] = None
     gpu: Optional[Dict[str, Union[bool, str, float]]] = None
     storage: Optional[Dict[str, str]] = None
-    model: Optional[Dict[str, Union[bool, str]]] = None
+    model: ModelStatusInfo
 
 class ValidationResponse(BaseModel):
     config_validation: Dict[str, bool]