Scott McKay

/azp run Windows CPU CI Pipeline,Windows GPU CI Pipeline,Windows GPU TensorRT CI Pipeline,ONNX Runtime Web CI Pipeline,Linux CPU CI Pipeline,Linux CPU Minimal Build E2E CI Pipeline,Linux GPU CI Pipeline,Linux GPU...

/azp run MacOS CI Pipeline,orttraining-amd-gpu-ci-pipeline,orttraining-linux-ci-pipeline,orttraining-linux-gpu-ci-pipeline,orttraining-ortmodule-distributed,onnxruntime-python-checks-ci-pipeline,onnxruntime-binary-size-checks-ci-pipeline

> I see 4 pipelines failing with many complaining of permission errors for needed file paths. I suspect this is not related to my commit Haven't seen that error before...

Seems like your runtime error comes from a vendor extensions issue, which is outside of ORT. https://source.android.com/devices/neural-networks/vendor-extensions Currently the ORT NNAPI execution provider does not have support for converting the...

We could certainly look at adding support for LSTM to the NNAPI EP in the next release, but that will be a few months away. Contributions are welcome if you...

As the ORT NNAPI EP doesn't have handling for LSTM currently it will be using the CPU execution provider for LSTM operations on all ARM platforms. If there's a crash...

I will add a check that the bias/scale are a valid size in a separate PR so there's no overrun, but what's the real-world use case where bias and scale...

Unfortunately the NNAPI Sigmoid (ANEURALNETWORKS_LOGISTIC) only supports up to 4 dimensions. https://developer.android.com/ndk/reference/group/neural-networks We can update the ORT NNAPI EP to add a check so we don't attempt to use NNAPI...

That performance would be as expected. Looking at [an onnx model in the release from that repo](https://github.com/WongKinYiu/yolov7/releases/download/v0.1/yolov7-nms-640.onnx) there's a Sigmoid node after every Conv. If NNAPI can't run the Sigmoid...

Ah ok - missed those 3. Didn't expect it to become 5D at that point. Those switches could potentially cause a lot of overhead as you have to move data...