Hi all,
MLIR/LLVM remains a central place to contribute, but right now the XLA APIs and (M)HLO seem like the best established abstractions to expose target-independent code generation facilities to ML frameworks, and we want to continue to leverage them.
In the coming months we'll reorganize the repository inside TensorFlow to prepare to split XLA into a separate repository on GitHub. This includes also the codebase that is right now under tensorflow/compiler/mlir/hlo and also exported as
https://github.com/tensorflow/mlir-hlo
The preparatory steps involve removing dependencies from tensorflow/compiler/xla/ on the rest of TensorFlow, and one of these step will will be to move tensorflow/compiler/mlir/hlo to tensorflow/compiler/xla/mlir_hlo
Cheers,
--
Mehdi