Open
Description
Describe the issue
#23026 changes the API to define device
as a Promise that, when awaited, returns the GPUDevice for the backend. However, that promise is never set (if you try to access device
before initializing an inference session, it's just undefined
). Similarly, if you set
the device
, it just gets overwritten. #21553 added code to implement this but it didn't end up getting merged...
Has anyone been looking into implementing the behavior described by the API?
@fs-eire
To reproduce
await ort.env.webgpu.device
at the beginning of your web app's execution.
Urgency
No response
ONNX Runtime Installation
Released Package
ONNX Runtime Version or Commit ID
1.22.0
Execution Provider
'webgpu' (WebGPU)