How to run rools like run_hlo_module and hlo-opt on TPUs?

[SamSz04]

It's feasible to use run_hlo_module and hlo-opt on GPU or CPU through bazel build. I wonder is it possible to use them on Google TPUs? Is there any instruction on this?
Thanks!

[akuegel]

The externally available tools unfortunately don't allow to run on Google TPUs. They would need to link the TPU XLA backend into them, but that is also not available in open source.