Meta: Introduce "Interop" label?

[inexorabletash]

While doing triage, I noticed a cluster of issues that relate to inconsistencies between different underlying platforms/backends, e.g. DirectML, CoreML, TFLite, XNNPACK, etc.

• general:
  • Do we need an MLConstantOperand? #668
  • Consider removing MLActivation parameters used for op fusion #658
  • Consider adopting new broadcasting rules #590
  • Clarify the casting behavior from floating-point / signed integers <-> unsigned integers #489
  • Define the maximum number of operand dimensions (maximum rank) #456
  • Allow 0 size dimensions (dimensions containing a 0 in the list of sizes, not a rank of 0 which is valid) #391
• op-specific:
  • Consider dropping the support of uint64/int64 data type for some operators #654
  • Consider changing output type of ArgMax/Argmin to int32, or allow passing output_type #653
  • ArgMax/Min selectLastIndex is not supported on CoreML #652
  • argMax/Min only support scalar axis in TFLite runtime #629
  • Should axes be required parameter for layerNormalization build method? #487
  • Should scale and bias be required inputs for batchNormalization op? #481
  • Zero dilations needs spec clarification #444
  • Clarify the restriction for minValue and maxValue of MLClampOptions #396
  • split into sizes are not widely supported #392
  • Need to restrict the value of alpha to be positive for elu operation #383
  • Simplify the operand layout support of conv2d and pooling 2d operations #324
  • Should dilated pooling be supported #180

Forward progress on resolving these requires considerations such as:

• Does WebNN support only the "lowest common denominator" behavior?
• Does WebNN require support, and implementations must work around backend limitations?
• Should behavior differences be allowed, and WebNN defines what is permitted (e.g. that options may just be a hint)
• Allow checking whether operators/types are supported for a backend before creating a graph #463

This was brought up in the 2024-05-02 WG telecon, @anssiko suggested a meta-issue for follow-up

[anssiko]

Thanks @inexorabletash. I've put up a PR #682 and hearing no concerns I recommend we adopt this label to help contributors navigate the issue list.

It is rather easy to split/merge/rename labels if we find out later we want to e.g. further split these interop issues into specialized chunks such as webgpu interop that coexists with this one.

[anssiko]

Reopened to be closed when my triage team friend @inexorabletash has assigned the labels :-)

[inexorabletash]

@anssiko - it might be easier for you to do it. If you go to https://github.com/webmachinelearning/webnn/issues?q=is%3Aissue+is%3Aopen+654+653+652+629+487+481+444+396+392+383+324+180+668+658+623+590+489+456+391 does your issue list have a checkbox column on the left, allowing you to apply labels to multiple issues at once? Apparently while I can apply labels to one issue at a time I can't bulk-assign labels.

I have that checkbox for repos I'm an Admin on (like https://github.com/w3c/IndexedDB) but not WebNN (where I have Triage)

[inexorabletash]

Never mind - I ended up doing it manually. Closing this!