XLA Compilation at the time of inferencing

Hi ,

I want to check whether XLA compilation can be performed while inferencing with loading saved graph model. what I am performing is applying @tf.function( jit_compile=True) while inferencing with loaded saved model graph. I am seeing performance improvement in throughput by 7x .

Is above doing is correct way or I am doing it wrong please suggest ?

Hi @Akash_Mishra, The approach you are following is correct, but if you want to export your functions to other backend platforms to perform xla_compile refer to this documentation. Thank You.