Have You Tried AWS Inferentia2 for ML Deployments?