Export tf.Keras 2.2 model to tflite with Static Tensors for accelerating using delegates

I am exporting tf.Keras 2.2 model using default converter, and everything works good, except it can't use CoreML/GPU accelerator. The reason is "Attempting to use a delegate that only supports static-sized tensors with a graph that has dynamic-sized tensors." It is common problem, even in NNDelegate it described. But there is no info in the docs how to solve this.