Deep Learning with TensorFlow

JIT compilation

So we mentioned just-in-time compilation, so what's it all about?

The main thing about JIT compilation is that the programs you get the program to be built at runtime so when you type a TensorFlow expression and press enter, you don't want to sit for a while to figure out how to compile so you will also get low-overhead compilation. As you saw in the previous example that you go the assembly instructions generated just by pressing Enter.

Another advantage of JIT is that you can bind variables later in the code.

For example, you don't have to specify the batch size at the very beginning of your code. You want to able to specify the batch size right when you find out what you want it to be.

So if we looked at the basic TensorFlow ...