Life cycle of ML program(Compilaition to execution)

Hello People,

I’m trying to understand the compilation/execution flow of a ML program(using TensorFlow). Particularly interested in understanding how TensorFlow graphs are lowered to MLIR using XLA dialect and lowering to a CPU/GPU via LLVM.

It would be great, if someone can give an overview on it. Thanks.

Hi,
I think this is something that can help you on that front: https://google.github.io/iree/design-docs/simple-ir-walkthrough.

1 Like

This looks great, Thanks.