Tensorflow/Yolov5 Memory Leak

We have a simple socket services written in python that loads keras model and YoloV5 model into memory, and prediction is run on images when socket receives messages. The problem is every time a prediction is run on either Keras or Yolo5, the python process's memory consumption grows, and after thousands of predictions the python process will run out of memory. We need you to fix the memory leak, please note the following: 1. [login to view URL]() did't help. 2. We can't reload the model before each prediction, it takes too long. The model must be preloaded into memory for instant inference. 3. We are doing inferences in CPU only 4. You should test the code with large number of different images. 5. If the memory leak cannot be resolve, we will accept solutions that converts keras/yolov5 models to another framework that doesn't leak. See attached for example of memory leaking code.