14:58:08 RRSAgent has joined #webmachinelearning 14:58:09 logging to https://www.w3.org/2022/12/01-webmachinelearning-irc 14:58:10 RRSAgent, make logs Public 14:58:12 please title this meeting ("meeting: ..."), anssik 14:58:28 Meeting: WebML WG Teleconference – 1 December 2022 14:58:33 Chair: Anssi 14:58:39 Agenda: https://github.com/webmachinelearning/meetings/blob/main/telcons/2022-12-01-wg-agenda.md 14:58:44 Scribe: Anssi 14:58:48 scribeNick: anssik 14:58:54 Present+ Anssi_Kostiainen 14:59:00 Regrets+ Dominique_Hazael-Massieux 14:59:14 ghurlbot, this is webmachinelearning/webnn 14:59:14 anssik, OK 14:59:18 RRSAgent, draft minutes 14:59:18 I have made the request to generate https://www.w3.org/2022/12/01-webmachinelearning-minutes.html anssik 15:00:53 ningxin_hu has joined #webmachinelearning 15:01:07 Present+ Zoltan_Kis 15:01:14 Present+ Bruce_Dai 15:01:22 Present+ Ningxin_Hu 15:01:29 bruce_dai has joined #webmachinelearning 15:03:59 zkis has joined #webmachinelearning 15:05:49 Topic: Announcements 15:05:55 Subtopic: webnn-polyfill npm package published 15:06:04 anssik: webnn-polyfill published as an npm package: 15:06:14 https://www.npmjs.com/package/@webmachinelearning/webnn-polyfill 15:06:14 https://github.com/webmachinelearning -> @webmachinelearning 15:06:28 ... thank you to all the polyfill contributors! 15:10:00 Topic: WebNN API Candidate Recommendation open issues 15:10:06 -> Current CR issues https://github.com/webmachinelearning/webnn/labels/cr 15:10:14 Subtopic: Web platform tests 15:10:28 anssik: I want to discuss & resolve any blockers for w-p-t & webnn-baseline reference impl. 15:11:00 ... webnn-baseline is the pure JS double-precision baseline implementation of WebNN operations for testing purpose without 3rd party deps 15:11:06 anssik: webnn-baseline remaining implementation work is documented in: 15:11:11 -> webnn-baseline implementation open issues https://github.com/webmachinelearning/webnn-baseline/issues 15:11:27 Implement softsign 15:11:27 Implement softplus 15:11:27 Implement resample2d 15:11:27 Implement others reduction operations 15:11:27 Implement l2Pool2d 15:11:27 Implement pad 15:11:29 Implement linear 15:11:29 Implement instanceNormalization 15:11:29 Implement hardSwish 15:11:30 Implement hardSigmoid 15:11:30 Implement convTranspose2d 15:11:30 New feature: Support float16 15:11:59 anssik: next, let's discuss the status of the wpt open PRs 15:12:09 -> wpt open PRs https://github.com/web-platform-tests/wpt/labels/wg-s_webmachinelearning 15:12:31 ... I bucketed the wpt PRs into: 15:12:36 ... 1) Ready to merge? 15:12:41 ... 2) Ready to close? 15:12:48 ... 3) Review comments to be addressed 15:12:55 ... 4) Pending reviewers 15:13:15 ... let's look at each of these buckets. Bruce, et al. please correct me needed. 15:13:25 ... 1) Ready to merge? 15:13:30 -> https://github.com/web-platform-tests/wpt/pull/36782 [wpt] concat tests 15:14:42 ... 36782 is reviewer approved but fails auto merge due to wpt-chrome-dev-stability failure (sink-task is a check that reports a failure if any of the taskcluster checks fails) 15:14:48 ... is this due to the WebNN API being under a build-time flag in Chromium? 15:15:33 https://github.com/web-platform-tests/wpt/pull/36908 15:16:05 Bruce: #36782 blocked by #36908 15:16:05 https://github.com/webmachinelearning/webnn/issues/36782 -> Issue 36782 [not found] 15:16:05 https://github.com/webmachinelearning/webnn/issues/36908 -> Issue 36908 [not found] 15:16:55 RafaelCintron has joined #webmachinelearning 15:17:21 anssik: I could ping Dom to accelerate the merge of this PR 15:17:26 Bruce: that'd be good 15:17:33 Present+ Rafael_Cintron 15:17:59 ... 2) Ready to close? 15:18:05 -> https://github.com/web-platform-tests/wpt/pull/31771 [wpt] polyfill as an idlharness.js dep 15:18:31 anssik: 31771 proposed to add the webnn-polyfill as an idlharness.js dep, but per review comments injecting scripts into idlharness.js is not allowed 15:18:47 anssik: can we close this PR #31771? 15:18:48 https://github.com/webmachinelearning/webnn/issues/31771 -> Issue 31771 [not found] 15:18:51 Bruce: can close this 15:19:10 -> https://github.com/web-platform-tests/wpt/pull/33857 [wpt] idlharness.js fix 15:19:29 anssik: ... 33857 is superceded by now merged https://github.com/web-platform-tests/wpt/pull/35790 15:20:13 anssik: can we close #33857? 15:20:14 https://github.com/webmachinelearning/webnn/issues/33857 -> Issue 33857 [not found] 15:21:42 Bruce: #33857 can be closed 15:21:58 ... 3) Review comments to be addressed 15:22:09 -> https://github.com/web-platform-tests/wpt/pull/34287 [wpt] first-wave ops tests (part 1) 15:22:30 anssik: comments from Dwayne to be addressed prior to merging 15:23:05 Bruce: this big PR can be closed, now separated in multiple smaller PRs 15:23:17 ... #34287 can be closed? 15:23:24 s/closed?/closed 15:23:31 -> https://github.com/web-platform-tests/wpt/pull/37254 [wpt] softmax tests 15:24:17 anssik: looks like very fresh PR that is being worked on 15:24:28 anssik: 4) Pending reviewers: 15:24:38 -> https://github.com/web-platform-tests/wpt/pull/37222 [wpt] x() + xSync() naming 15:24:41 https://github.com/webmachinelearning/webnn/issues/34287 -> Issue 34287 [not found] 15:26:01 Bruce: submitted this IDL manually, this naming PR is OK 15:27:01 -> https://github.com/web-platform-tests/wpt/pull/36202 [wpt] first-wave ops tests (part 2) 15:27:37 Bruce: #36202 is separated into smaller PRs, can be closed 15:27:37 https://github.com/webmachinelearning/webnn/issues/36202 -> Issue 36202 [not found] 15:27:48 -> https://github.com/web-platform-tests/wpt/pull/37067 [wpt] clamp tests 15:27:48 -> https://github.com/web-platform-tests/wpt/pull/37068 [wpt] gemm tests 15:27:48 -> https://github.com/web-platform-tests/wpt/pull/37069 [wpt] matmul tests 15:27:48 -> https://github.com/web-platform-tests/wpt/pull/37070 [wpt] reshape tests 15:27:49 -> https://github.com/web-platform-tests/wpt/pull/37071 [wpt] relu tests 15:27:49 -> https://github.com/web-platform-tests/wpt/pull/37072 [wpt] slice tests 15:28:47 Bruce: Dwayne, Wanming and Ningxin are reviewing these PRs, concat actively being reviewed now 15:29:22 anssik: thank Bruce and Dwayne, others for your work on wpt! Contributions from everyone is welcome even informal reviews for the above-mentioned PRs to get them into a merge-able state. 15:29:32 ... We have a good testing story developing here for CR publication purposes 15:29:54 ... I'm confident we will be able to satisfy the "must document how adequate implementation experience will be demonstrated" CR requirement even if some of these wpt tests would be still be in review while we transition to CR. 15:30:15 RRSAgent, draft minutes 15:30:15 I have made the request to generate https://www.w3.org/2022/12/01-webmachinelearning-minutes.html anssik 15:30:21 Subtopic: Use modern WebIDL conventions 15:30:57 anssik: to satisfy the CR readiness requirements, the WebNN API spec is expected to specify its normative parts using the modern authoring conventions. This convention is codified in the WebIDL spec. 15:31:46 ... Zoltan has volunteered to make editorial adjustments to the specification. I have discussed this work with Zoltan and we agreed to focus on editorial changes and minimize any normative changes unless they are needed to fix a spec bug, or to clarify an aspect where the spec and implementation disagree. 15:32:30 ... Zoltan, please feel free to introduce the editorial changes in PR #301 15:32:30 https://github.com/webmachinelearning/webnn/issues/301 -> Pull Request 301 WiP: editorials (zolkis) 15:33:17 Zoltan: pushed a few editorial changes after discussion 15:33:35 ... in this PR #301 it is mainly trivial changes 15:33:54 ... changed the section titles to include interface/method/dictionary etc. 15:34:11 ... improved the ML.createContext() steps using prose also used in other spec's. 15:34:39 ... discussion on preferred "in parallel" usage in the PR 15:35:19 ... removed the unused internal slot [[implementation]] 15:35:59 ... we should see if we need internal slot for ML context, or we can also refer to it in prose 15:36:12 ... added first stab at MLGraphBuilder constructor steps 15:36:23 ... should we validate context given to the builder? 15:36:28 q? 15:36:52 ... a workaround would be to move builder to context. 15:38:04 ... can create separate issues for these 15:41:29 q+ 15:41:40 ack ningxin_hu 15:42:03 ningxin_hu: questions, this PR has WiP title, are you still working on it, is it ready for review? 15:42:09 Zoltan: ready to review 15:42:23 s/questions/question 15:43:30 Zoltan: prefer smaller atomic PRs 15:44:20 ningxin_hu: I briefly reviewed the 3 commits in the PR, we can separate trivial ones to separate PR to accelerate 15:44:58 ... in Chromium we use big PR to show the overall direction or arch change, get early feedback and then once solidified we separate that for smaller pieces for landing 15:45:11 ... this PR #301 is a collection of multiple editorial changes 15:45:12 https://github.com/webmachinelearning/webnn/issues/301 -> Pull Request 301 WiP: editorials (zolkis) 15:45:48 Zoltan: if you make comments in that PR I will refine the PR with feedback 15:47:32 ningxin_hu: prefer smaller self-contained PRs e.g. proposing to separate the promise usage to a separate PR 15:50:30 Zoltan: I'll work on the editorial change in between the calls and will share with editors PRs for review offline 15:51:28 Zoltan: welcome comments on proposals in #298 15:51:28 https://github.com/webmachinelearning/webnn/issues/298 -> Issue 298 API review, questions, brainstorming (zolkis) 15:52:03 ... specifically: Proposal for exposing context 15:52:19 ... and proposal to include graph builder, command encoder as attributes to MLContext, and make MLGraph an internal slot of MLContext 15:53:34 ... two of the most recent proposals 15:54:03 anssik: it'd be good if these proposal would be positioned as v2 / post-CR to not de-stabilize the ongoing implementation and interop testing efforts during our last mile sprint to CR readiness 15:54:15 Proposal 1: https://github.com/webmachinelearning/webnn/issues/298#issuecomment-1310762948 15:54:27 Proposal 2: https://github.com/webmachinelearning/webnn/issues/298#issuecomment-1331371138 15:54:43 q+ 15:54:48 ack ningxin_hu 15:55:06 ningxin_hu: thanks for this brainstorming happy to discuss 15:55:28 ... as Anssi said, regarding CR target, I also prefer to these be positioned v2 / post-CR and focus on CR blockers 15:56:08 ... another comment regarding implementation, with this implementation we have Dev Trial available in browser and can gather developer feedback and be in a better position on what changes are benefiting the API customers 15:56:27 ... with implementation experience and developer feedback we are in a better position 15:58:20 https://www.w3.org/TR/design-principles/#priority-of-constituencies 16:00:35 ningxin_hu_ has joined #webmachinelearning 16:00:38 Topic: WebML WG Charter 2023-2025 under development 16:00:43 anssik: Web Machine Learning Working Group Charter for 2023-2025 is now under development. 16:00:58 anssik: we expect to have a good draft charter ready EOY 2022 16:01:07 ... - Q4 '22: Charter development 16:01:07 ... - Q1 '23: W3C Advisory Committee review 16:01:07 ... - Q2' 23: Charter approved 16:01:21 Subtopic: [new] OpenXLA coordination 16:01:28 -> OpenXLA coordination https://github.com/w3c/machine-learning-charter/issues/28 16:01:51 anssik: on our last call Google's Eugene Burmako presented on StableHLO that is a portable ML compute operation set being worked on as part of OpenXLA project. We heard StableHLO is being adopted by Google products and projects 16:02:00 –> WebML WG Teleconference – 17 November 2022 https://www.w3.org/2022/11/17-webmachinelearning-minutes.html 16:02:08 anssik: it was discussed we should consider revising the op set compatibility exploration we did earlier in this group 16:02:22 -> WebNN op compatibility exploration https://github.com/webmachinelearning/webnn/blob/main/op_compatibility/first_wave_models.md 16:02:27 anssik: this initial WebNN op compatibility exploration looked at lowering to XLA-HLO in 2020. 16:02:42 ... it was suggested we should coordinate with the OpenXLA project to ensure any synergies can be unlocked between WebNN and StableHLO op sets 16:03:01 ... a proposal was to add StableHLO into the compat study. We heard from Eugene that StableHLO targets feature completeness EOY 2022 and thus a good start for this effort would be ~Q1-Q2'23. 16:03:12 ... External Coordination section of the charter is the place where to note related work in the ecosystem where coordination is expected 16:03:24 ... the current charter notes Khronos and ECMA TC39 in this section 16:03:30 -> External Coordination (from current charter) https://www.w3.org/2021/04/web-machine-learning-charter.html#external-coordination 16:03:35 anssik: here's my proposed addition to the External Coordination of the charter: 16:03:42 OpenXLA Project develops StableHLO, a portable ML compute operation set that makes frameworks easier to deploy across different hardware. WebNN API targets diverse hardware platforms and defines an operation set whose high-level operations can be decomposed to low level primitives that can map to StableHLO operations. Coordination and alignment between these two operation sets is beneficial to the open ML ecosystem. 16:04:21 anssik: feedback welcome via the GH issue 16:04:58 proposed RESOLUTION: Add OpenXLA Project to External Coordination 16:05:15 +1 16:05:21 anssik: any objections or concerns? 16:05:50 PR sgtm 16:05:54 RESOLUTION: Add OpenXLA Project to External Coordination 16:06:16 anssik: I will craft a PR to document the proposal and seek review from the WG 16:07:21 -> Open issues https://github.com/w3c/machine-learning-charter/issues 16:07:37 -> On-device training https://github.com/w3c/machine-learning-charter/issues/27 16:10:41 s/-> On-device training https://github.com/w3c/machine-learning-charter/issues/27// 16:10:46 Subtopic: [new] On-device training 16:10:51 -> On-device training https://github.com/w3c/machine-learning-charter/issues/27 16:10:58 anssik: currently, on-device training is out of scope: 16:10:58 16:10:58 "Training capabilities are out of scope due to limited availability of respective platform APIs." 16:12:05 anssik: Please provide feedback if you think on-device training should be in scope for 2023-2025. 16:12:10 RRSAgent, draft minutes 16:12:10 I have made the request to generate https://www.w3.org/2022/12/01-webmachinelearning-minutes.html anssik 18:11:58 Zakim has left #webmachinelearning