Candidate Recommendation readiness tracker

[anssiko]

ℹ️ The requirements and status have been updated to reflect Q1'24 CR Snapshot target.

This is a meta issue to track Candidate Recommendation (CR) publication readiness for the Web Neural Network API.

About

A Candidate Recommendation is a document that satisfies the technical requirements of the Working Group that produced it and their dependencies, and has already received wide review. W3C publishes a Candidate Recommendation to

• signal to the wider community that it is time to do a final review
• gather implementation experience

Requirements

Legend:
⚪ Not started
🟡 In progress
🟢 Completed

To publish a Candidate Recommendation, in addition to meeting the requirements for advancement a Working Group:

• must show that the specification has met all Working Group requirements, or explain why the requirements have changed or been deferred,
  • 🟢 The document meets the requirements set forth in the WG Charter and satisfies the Use Cases updated with transformers
  • 🟢 The document is considered complete and fit for purpose
  • 🟢 Completed Use modern WebIDL and Infra standard conventions #210
  • 🟢 Other deliverables: ethical considerations documented
• must document changes to dependencies during the development of the specification,
  • 🟢 Diff in normative references since CR1: no changes
• must document how adequate implementation experience will be demonstrated,
  • 🟢 WPT coverage 93.6%:
    • web-platform-tests / wpt.fyi
    • WIP: gru, gruCell, lstm, lstmCell, resample2d WPT tests tracker / 2 for remaining ops #338
      (https://wpt.fyi/results/webnn?label=experimental&label=master&aligned&view=subtest)
  • 🟢 implementation experience:
    • Google Chrome and Microsoft Edge: XNNPACK CPU backend, DirectML GPU backend
    • ChromeOS: MLService CPU backend
    • JavaScript ML Framework Integrations: ONNX Runtime (release), TensorFlow Lite for TF.js (experimental)

• must specify the deadline for comments, which must be at least 28 days after publication, and should be longer for complex documents,
• must show that the specification has received wide review, and
  • 🟡 Delta wide review tracked in Wide review tracker #239, previously completed in Q1'23 for CR1
  • Implementation-specific reviews of interest (informational, not a complete list):
    • 🟢 Chromium security review for accelerated backends
• may identify features in the document as at risk. These features may be removed before advancement to Proposed Recommendation without a requirement to publish a new Candidate Recommendation.
  • 🟢 WebGPU interoperability needs more implementation experience, SOTD updated accordingly to note work-in-progress MLBuffer Update Status of this document for CR Snapshot #532

[huningxin]

I'd like to propose WebNN baseline implementation (#245) to be tracked by CR readiness. It would be used to establish the baseline for WebNN web-platform-tests.

[wchao1115]

issue #226, #230, and #229 should also be tracked for CR readiness.

[anssiko]

Per WebML WG Teleconference – 13 January 2022, explicitly added to the list the following requirement:

Other deliverables: ethical considerations documented

This work in ongoing in its own repo and is referenced from the WebNN API specification. The plan on how we advance this work was presented in the WebML WG Teleconference – 16 Dec 2021.

[anssiko]

I added webnn-baseline implementation as a new component for the adequate implementation experience requirement. The initial webnn-baseline implementation is currently in review: webmachinelearning/webnn-baseline#1

[anssiko]

The initial webnn-baseline implementation landed. We continue refine this project to help gather implementation experience for Candidate Recommendation purposes. Another ongoing effort that contributes to implementation experience is the Chromium WIP prototype.

[anssiko]

As discussed, our collective goal is to add some final polish and ship the CR during Q1 2023. Thank you for your contributions!

[anssiko]

We have issued a Call for Consensus to publish WebNN API Candidate Recommendation.

[anssiko]

FTR, the Call for Consensus passes.

Related, in preparation for the upcoming Candidate Recommendation publication I've done an issue triage. If you identify any bugs in labelling please let me know and we'll adjust. The labels are considered informative. The purpose is to make the issue list more easy to grasp for the WG as well as for other folks who might peek into our repo on their quest to make a positive impact with concrete contributions. Don't read too much into the semantics of the names of these labels.

[anssiko]

I’m excited to announce the Web Neural Network API (WebNN) has been published as a Candidate Recommendation today:

🚀 https://www.w3.org/TR/2023/CR-webnn-20230330/

[anssiko]

The WG is looking publish a new Candidate Recommendation Snapshot in Q1'24. Delta wide review is in flight #239 (comment) and the Status of this document is staged at #532. Notably, the WG has gathered further implementation experience since its previous CR publication.

[anssiko]

As discussed, with the recent updates to the spec, WPT and implementation status, we're getting very close to CR Snapshot readiness.

We're awaiting a go-ahead signal from the TAG delta review to initiate the transition. Based on my assessment we satisfy all the other requirements. @dontcallmedom can you please cross-check against the latest transition request expectations that we've all set otherwise. Would it be appropriate to issue a CfC now (noting the TAG delta review is in flight)?

Thank you everyone for your major contributions that make this important publication happen!

[dontcallmedom]

yes, starting the CfC seems appropriate; if the TAG review raises new concerns, we would still have the opportunity to revisit our decision prior to submitting the transition request.

(and I confirm we're otherwise ready for the transition request)

[anssiko]

Thanks @dontcallmedom.

Call for Consensus to publish a new WebNN API Candidate Recommendation Snapshot issued.

[anssiko]

The most recent Call for Consensus passes.

[anssiko]

CR Snapshot transition request: w3c/transitions#592 (thanks @dontcallmedom!)

[anssiko]

I’m pleased to announce the Web Neural Network API (WebNN) has been published as a Candidate Recommendation Snapshot today:

🚀 https://www.w3.org/TR/2024/CR-webnn-20240411/

Congratulations to the WG for this major accomplishment. The publication would not have been possible without every WG participants' contributions.