I have deployed a sentence transformer model for semantic check in an aws server which is under load balancer and in autoscaling group. I am running this server using gunicorn command with uvicorn as worker class with timeout as 3000 seconds. Along with that I have given max requests to 1000 and jitters as 50. But the server is giving response as 504 when trying to access this server to while accessing from application server. I have defined the sentence transformer model at the root of api(fastapi) call. What could be the reason to receive 504 error response from the upstream server?

I have tried to load model at the root of my api file. Then the system which used to hang earlier is not hanging now. But the 504 error still persists