[Meta] WebNN - ONNX Runtime Issues

[ibelem]

Describe the issue

This is a meta issue to track WebNN / ONNX Runtime relevant open issues.

Feature Request

• [Feature Request] Report error codes that represent critical errors, such as device-removal #26648
• [Feature Request] GetEpDevices() returns a sorted EP devices list #25302
• [Feature Request] Shape inference for GroupQueryAttention Op #23189
• [Feature Request] Create a shared non-CPU allocator that outlives inference sessions #25785

ep:WebGPU

• [Feature Request] [WebGPU] Add support for TinyLlama operators to avoid CPU fallback #25236

Bug

• [OVEP][NPU]compile model will hang all the time and never return when running model florence-2-base-vision-encoder-fp16 with OpenVINO-NPU #26643
• [ORT GPU(WebGPU)][WebNN]Error will happen when running detr-resnet-50 model with ORT-WebGPU backend in WebNN-GPU #26618
• Spec changed for reduction ops when noop_with_empty_axes is True #26288

ep:WebGPU

• [WebGPU] GQA produces unexpected result with FlashAttention applied #26064
• No opset import for domain 'com.microsoft.nchwc' error #25914

ep:DML

• [ORT GPU (DML EP)][WebNN] Handle device-removal error in DML EP #26606
• [ORT CPU][ORT GPU (DML)][WebNN] Fail to run WebNN batchNormalization tests with 1D tensor due to "Dimension mismatch in unification between 6 and 1" #26521
• [ORT CPU][ORT GPU (DML)][WebNN] Wrong results to run some WebNN dequantizeLinear tests #26525
• [ORT CPU][ORT GPU (DML)][WebNN] Error Unrecognized attribute: dilations for operator QLinearAveragePool #26528
• [ORT GPU (DML)][WebNN] Wrong results to run some WebNN dequantizeLinear tests #26530
• [ORT GPU (DML)][WebNN] Fail to run two WebNN batchNormalization tests due to tolerance issues #26532
• [ORT GPU (DML)][WebNN] Wrong results to run one WebNN l2Pool2d test and two WebNN quantizeLinear tests #26533
• [ORT GPU (DML)][WebNN] Exception during initialization when running some WebNN matmul and triangular tests with 5D input #26529
• [Crash] [Blocker] [ORT GPU (DML)][WebNN] Edge Canary browser crashed when running two WebNN QDQ subgraph tests #26531

ORT Default CPU

• [ORT CPU][WebNN] Fail to run some WebNN batchNormalization tests due to tolerance issues #26522
• [ORT CPU][WebNN] Wrong result to run a subgraph test with WebNN batchNormalization default + sigmoid operators #26524

To reproduce

Please refer to links above for more details.

Urgency

No response

ONNX Runtime Installation

Other / Unknown

ONNX Runtime Version or Commit ID

1.24.0

Execution Provider

Other / Unknown