Failed to construct interpreter TensorflowLite

Question

I am using `

Cuda 10.1, cudnn 7.6, Tensorflow 2.3.0, keras 2.4.3

When training model I use save_weights_only=True, and a fo