At what level in the algo parallelized?

[aurotripathy]

hello @siboehm,
I read your post on lleaves from 2021.
Very well-written.
I'm new to lightGBM.
Your approach with LLVM to compile the model is something I have a question on.
In a multi-core machine, how you parallelize the work?
Is it that each tree is a separate execution thread?
Can the tree traversal be sped up by distributing the work ?

Thank you
Auro

[siboehm]

see https://github.com/siboehm/lleaves/blob/master/lleaves/lleaves.py#L208, it's data parallel