Keras model running on Sagemaker endpoint randomly failing

Question

I have an endpoint running in Sagemaker with a Pyfunc model deployed with MLflow that loads a Keras model and runs a prediction on it. The endpoint runs normally with no iss