Any doc updates on mlir-hlo?

[孙鹏]

Hi there,

I'm new to mlir-hlo community. As a newcomer, I found it really hard to bootstrap. For example, I can't do some simple tasks like compile a single node GraphDef to binary that can run on GPU. And I've read the project README doc, however, the most relevant `End-to-End pipeline` part remains `TODO`. :-)

May I ask if there are any recommended docs / tutorials for newcomers please?

Thanks!

[Mehdi AMINI]

Hi,

The best source of information and examples for "end-to-end" pipeline with MLIR in this context is likely gonna be found on IREE these days: https://google.github.io/iree/

Cheers,

-- 

Mehdi

--
You received this message because you are subscribed to the Google Groups "MLIR" group.
To unsubscribe from this group and stop receiving emails from it, send an email to mlir+uns...@tensorflow.org.
To view this discussion on the web visit https://groups.google.com/a/tensorflow.org/d/msgid/mlir/7134b34a-b249-4952-b117-85d4d0dafe7cn%40tensorflow.org.

[Liwei Dai]

Hello Mehdi,

Thanks for the info, but the link doesn't work now, maybe you mean https://github.com/iree-org/iree/tree/main/integrations/tensorflow nowadays.

But is there like any simple tutorials on how to enable and unittest MLIR in original pip install tensorflow?

As an end user, what I'm looking for is doing some test to see how MLIR is working in tensorflow.

For example, I was doing simple op unittest by running for example an "relu" operation on GPU and CPU with the same input numpy arrays, and check if the results match.

Meanwhile I can also do simple profilings using nvprof or NSight tools.

I can also run models, for example, a Resnet50 on imagenet, on both GPU and CPU, and check results or do profiling.

Now with XLA, I can simple set some environment variables without changing the python code to run the same operation and model, and do the same thing.

But how do I verify the effect of MLIR in tensorflow? I thought it was like XLA, just setting some environment variables, but I didn't find any resources.