Transformer optimization on DirectML not fully working?

[kazssym]

I am using onnxruntime-directml to convert Stable Diffusion models for Olive. Regarding the following line, is DMLExecutionProvider not ready yet or not suitable for optimization in onnxruntime?

onnxruntime/onnxruntime/python/tools/transformers/optimizer.py

Line 93 in 6ea3908

if use_gpu and set(onnxruntime.get_available_providers()).isdisjoint(

[skottmckay]

DMLExecutionProvider is ready and stable.

That script is for optimizing BERT models and I'd guess that when that was being created/tested it was being used on Linux so DML wasn't included in the list. Not sure how that's related to Olive though.

[kazssym]

Thanks.