fix: for backend production

main.py

@@ -13,6 +13,7 @@ from infra.logger_structlog import StructLogger
 import asyncio
 import mlflow
 from backend.utils.experiment import log_config
+import torch
 
 @asynccontextmanager
 async def lifespan(app: FastAPI):
@@ -48,6 +49,7 @@ async def lifespan(app: FastAPI):
     yield
 
     logger.warn("Shutting down the server....")
+    torch.cuda.empty_cache()
     # You can remove connections and release gpu here .  
 
 mlflow.set_tracking_uri("sqlite:///config/logs/mlflow.db")

requirements.txt

@@ -8,4 +8,4 @@ pydantic
 pydantic_settings
 structlog
 
-tracking_system_ai@git+https://github.com/E1250/p-tracking_system.git
+tracking_system@git+https://github.com/E1250/p-tracking_system.git