13:59:08 RRSAgent has joined #webmachinelearning 13:59:08 logging to https://www.w3.org/2022/08/25-webmachinelearning-irc 13:59:10 RRSAgent, make logs Public 13:59:11 please title this meeting ("meeting: ..."), anssik 13:59:15 Meeting: WebML WG Teleconference – 25 August 2022 13:59:22 Chair: Anssi 13:59:27 Agenda: https://github.com/webmachinelearning/meetings/blob/main/telcons/2022-08-25-wg-agenda.md 13:59:33 Scribe: Anssi 13:59:41 scribeNick: anssik 13:59:46 scribe+ dom 14:00:08 ghurlbot, this is webmachinelearning/webnn 14:00:09 anssik, OK 14:00:10 ningxin_hu has joined #webmachinelearning 14:00:22 Present+ Anssi_Kostiainen 14:00:31 Present+ Bruce_Dai 14:00:48 RRSAgent, draft minutes 14:00:48 I have made the request to generate https://www.w3.org/2022/08/25-webmachinelearning-minutes.html anssik 14:01:02 Present+ Jan_Wang 14:01:20 Present+ Dominique_Hazael-Massieux 14:01:28 Present+ Ningxin_Hu 14:01:56 AramZS has joined #webmachinelearning 14:02:13 bruce_dai has joined #webmachinelearning 14:02:58 jan-www has joined #webmachinelearning 14:03:58 RRSAgent, draft minutes 14:03:58 I have made the request to generate https://www.w3.org/2022/08/25-webmachinelearning-minutes.html anssik 14:04:15 Topic: Meeting logistics update 14:04:21 Present+ Rafael_Cintron 14:04:33 anssik: Teleconferences: Given good overlap of participants across WG+CG, we plan to repurpose monthly Sydney afternoon CG slots (Wed 04:00-05:00 UTC) for WG+CG topics when there's demand. This slot will be in addition to the normal WG bi-weekly (Thu 14:00-15:00 UTC) and caters specifically to our participants from APAC time zones. 14:04:40 RafaelCintron has joined #webmachinelearning 14:06:23 Present+ Jonathan_Bingham 14:06:26 anssik: TPAC: The Web Machine Learning Working Group's hybrid TPAC meeting is canceled due to key participants' inability to attend the in-person meeting. 14:06:32 q+ 14:06:33 jonathan has joined #webmachinelearning 14:07:06 ack ningxin_hu 14:07:51 ningxin_hu: re the APAC-friendly meeting time being scheduled on-demand - what will be the criteria? will it be a regular event? 14:08:13 anssik: when the Sydney people feels the call would be useful, we should schedule such a call 14:08:40 ... we could make it a recurring call at some point if that proves useful 14:08:49 zkis has joined #webmachinelearning 14:09:00 Present+ Zoltan_Kis 14:09:29 Topic: WebNN API Candidate Recommendation open issues 14:09:51 anssik: Review and discuss the current CR issues, work out a plan to address the issues prior to the expected CR publication. 14:09:53 -> https://github.com/webmachinelearning/webnn/issues?q=is%3Aissue+is%3Aopen+label%3Acr Current CR issues 14:10:25 anssi: beyond wide reviews, we have 5 open issues which makes me feel that we're on track 14:10:38 Subtopic: Remove the support of negative dimension values 14:10:44 anssik: issue #279 and PR #281 14:10:44 https://github.com/webmachinelearning/webnn/issues/279 -> Issue 279 Remove the support of negative dimension values (huningxin) cr 14:10:44 https://github.com/webmachinelearning/webnn/issues/281 -> Pull Request 281 Drop dynamic tensor support (huningxin) 14:12:16 ningxin_hu: I think we're good to merge 14:12:23 anssik: most recent change is the dfn of the byte length of an MLOperandDescriptor: 14:12:27 -> byte length of an MLOperandDescriptor https://pr-preview.s3.amazonaws.com/huningxin/webnn/pull/281.html#mloperanddescriptor-byte-length 14:12:33 ... this dfn is referenced from within the following algos in the spec: 14:12:33 14:12:33 ... - Syncronous Execution 14:12:33 ... - Asyncronous Execution 14:12:33 ... - Dispatch Execution Commands 14:12:38 ... one highlight: this is based on implementation experience as we're implementing WebNN in Chromium 14:12:51 ... this feature is not widely supported in native ML APIs we're targeting 14:12:58 ... that's the reason we propose to drop this 14:13:19 ... this feature can still be supported by frameworks 14:13:45 ... they can do shape inference and support dynamic tensor at their level 14:13:47 RRSAgent, draft minutes 14:13:47 I have made the request to generate https://www.w3.org/2022/08/25-webmachinelearning-minutes.html anssik 14:13:58 ... we would limit WebNN to static shapes 14:14:21 ... that makes for a good decomposition given WebNN's role as a backend for frameworks 14:14:44 ... WebNN Graph are made ready in two stage: compilation/build, and then compute 14:14:56 ... when a graph is a built, it should be ready for performant compute 14:15:28 ... if we were to support dynamic tensor shape through emulation, it would create latency in the first inference since the graph would need to be recompiled 14:16:05 ... so this PR would help keep our design goal and align well with the role of WebNN as a backend API 14:16:54 Rafael: I haven't reviewed the details, but I'm supportive of the direction of removing negative offsets 14:17:13 Anssi: ok, so we seem to be good with merging that one and solving the CR issue 14:17:23 q? 14:18:13 Rafael: has anyone looked at what support would look like on Mac? it would be good to include this as a data point in general 14:18:49 Hi, I'm Jan Wang, I have participated in a discussion before. I am now working in Alibaba as a front-end and machine learning engineer. Currently my main job is to build a web machine learning framework. Therefore, I am very interested in W3C's work for webNN and would like to attend your discussions in the future. 14:18:59 ningxin: good point - I can do more investigation on mac APIs and add the relevant information in the issue 14:19:49 Just wonder if there's any docker support for webnn-native project? 14:20:38 AFAIK, no docker support at this stage 14:20:38 https://github.com/webmachinelearning/webnn-native/ 14:21:00 Subtopic: Support asynchronous context creation 14:21:06 anssik: issue #272 and PR #274 14:21:07 https://github.com/webmachinelearning/webnn/issues/274 -> Pull Request 274 Support async context creation and use sync postfix (huningxin) 14:21:07 https://github.com/webmachinelearning/webnn/issues/272 -> Issue 272 Support asynchronous context creation (huningxin) cr 14:21:57 anssi: chai is not 100% happy with that design - I tried to explain some of the challenges in Web API naming 14:23:23 ... I thing these considerations motivated the design (not a complete list): 14:23:27 ... - sync version is for "non-JS developers" for bringing non-JS codebases via worker to the web, suggests async default and use sync postfix 14:24:02 ... - no established naming convention for sync vs async Web APIs due to historical reasons, no design patterns for naming 14:24:28 ... - some Sync suffixed Web APIs exists (FileReaderSync and WebGLSync), no Async suffixed Web APIs 14:24:53 ... - Sync suffix matches Node.js convention 14:25:09 q+ 14:25:14 ack RafaelCintron 14:25:30 s/I thing/I think/ 14:25:47 https://dontcallmedom.github.io/webidlpedia/names/ 14:27:27 https://gpuweb.github.io/gpuweb/#gpu-device 14:28:03 https://dontcallmedom.github.io/webidlpedia/members.html 14:30:00 Anssi: we probably need Chai to progress on this topic 14:30:22 q+ 14:30:27 Rafael: WebGPU uses an Async suffix - so this may be another parameter to consider 14:30:36 ... in particular in light of our interest in working with WebGPU 14:30:50 q? 14:30:55 ack ningxin_hu 14:31:02 WebGPU is the only API that seems to be using this Async suffix FWIW 14:31:28 Ningxin: this may be a good topic for the APAC topic, to get all the people with opinions together 14:31:52 Anssi: it would be useful to document more clearly the two approaches for comparison purposes 14:32:05 ... e.g. with the 2 versions of the IDL 14:32:23 Ningxin: I can; only 3 APIs will be changed 14:33:52 Anssi: is this blocking the landing of implementation patches? presumably this can be updated quickly 14:33:55 q? 14:34:11 Subtopic: Define ULP (unit of least precision) tolerances for testing 14:34:27 anssik: I wanted to check on this issue #265 that is a CR blocker. In Dec 2021 we decided to add ULP-based comparison for testing 14:34:27 https://github.com/webmachinelearning/webnn/issues/265 -> Issue 265 Define ULP (unit of least precision) tolerances for Conformance testing of WebNN API (BruceDai) cr 14:34:51 anssik: Bruce has an open PR for w-p-t that proposes new ULP-based assertions, needs w-p-t reviewers to unblock 14:34:56 -> PR #34287 [w-p-t] Add WebNN API operations tests https://github.com/web-platform-tests/wpt/pull/34287 14:34:56 https://github.com/webmachinelearning/webnn/issues/34287 -> Issue 34287 [not found] 14:35:18 ... Bruce is blocked on w-p-t infra team to know whether testharness.js can be augmented with these new assertions? 14:35:29 ... If that is not possible, can we add these assertions in the WebNN specific test suite and upstream them to testharness.js later? 14:36:11 -> PR #34287 [w-p-t] Add WebNN API operations tests https://github.com/web-platform-tests/wpt/pull/34287 14:36:56 dom: easier to make assertions local to our webnn test directory 14:37:15 ... helps with maintenance, the questions here is the right values for ULP tolerances 14:37:41 https://github.com/webmachinelearning/webnn/issues/265#issuecomment-1168093010 14:38:04 q? 14:38:12 Subtopic: Support for int8 quantized models 14:38:23 anssik: we agreed this feature is in CR scope in March 2022, I want to check we're on track to deliver this feature or whether we need to course-correct 14:38:30 #128 14:38:30 https://github.com/webmachinelearning/webnn/issues/128 -> Issue 128 WebNN should support int8 quantized models (wchao1115) cr 14:39:53 Rafael: I support bringing this feature in; we probably need to defer until Chai is available 14:40:14 ningxin_hu: I know that frameworks have also interest in this 14:40:21 q+ 14:41:03 jonathan: no strong opinion on this one 14:41:31 ... definitely not opposed 14:41:37 ack dom 14:41:54 dom: I wonder if anyone knows how much work this is, in the spec and implementations? 14:41:56 q? 14:42:54 ningxin_hu: based on my experience - given existing support in native APIs, there would be several options: quantize/dequantize operations that frameworks could use within graphs 14:43:10 ... or we could introduce this as an operand descriptor 14:43:23 ... another important aspect is the quantization schema 14:43:52 ... there are variants with different characteristics; we'll need to survey the existing landscape and find common ground 14:44:31 anssik: what would it take to go to the PR stage? 14:44:45 ningxin_hu: chai had mentioned that he had this on his todo list, so we should check with him 14:44:49 q? 14:45:32 Subtopic: Add method steps to operations 14:46:02 ... issue #210 and related issue #211 14:46:02 https://github.com/webmachinelearning/webnn/issues/211 -> Issue 211 Define algorithms for dictionaries with lists as default values (anssiko) 14:46:02 https://github.com/webmachinelearning/webnn/issues/210 -> Issue 210 Add method steps to operations (anssiko) cr 14:46:35 Anssi: Zoltan was interested in helping address this issue and upgrade modern spec conventions 14:46:43 ... including using internal slots 14:46:59 s/upgrade/upgrade WebNN to use/ 14:48:24 zoltan: I'll be happy to help, bringing both spec writing and some AI research backgrounds 14:48:45 anssik: thanks, really happy to get your help on this! 14:49:03 thanks Zoltan! 14:49:07 ... we'll expect PRs from Zoltan going forward 14:49:16 dom:q? 14:49:23 s/dom:q?// 14:49:26 q? 14:49:36 Subtopic: Wide review 14:49:43 #239 14:49:43 https://github.com/webmachinelearning/webnn/issues/239 -> Issue 239 Wide review tracker (anssiko) 14:50:01 anssik: here's how we're doing: 14:50:10 ... Accessibility: Review feedback received, no proposed changes 14:50:18 ... Architecture: Review closed as completed 14:50:26 ... Internationalization: Review feedback received, no proposed changes 14:50:32 ... Privacy: Review feedback received, WG response provided 14:50:44 -> WG response to Privacy review ttps://github.com/webmachinelearning/webnn/issues/280#issuecomment-1212860627 14:51:01 ... Security: Review closed as completed 14:51:12 anssik: overall, we're doing great with wide review, thanks all! 14:51:24 ... the only remaining review we should make progress with is WebGPU Working Group 14:51:30 ... for that our plan of record is: 14:51:35 ... Give WebGPU WG an update on the latest developments since Jan '22 as part of WebGPU wide review, introduce a plan for WebNN-WebGPU interop, consider CommandBuffer usage clarification issue #264 14:51:35 https://github.com/webmachinelearning/webnn/issues/264 -> Issue 264 CommandBuffer usage clarification: internal, external, both? (bbernhar) 14:52:51 q? 14:55:29 Topic: Proposed new features 14:55:35 Subtopic: l2Pool2d algorithm clarifications 14:55:44 #278 14:55:44 https://github.com/webmachinelearning/webnn/issues/278 -> Issue 278 How to define the algorithm of L2_Pool2d? (mingmingtasd) 14:56:55 ... comments? 14:57:08 i can check with mingming offline 14:57:28 Subtopic: Revisit proposed new features discussed earlier 14:57:33 #275 14:57:33 https://github.com/webmachinelearning/webnn/issues/275 -> Issue 275 Should MLBufferView + MLOperandDescriptor be strongly typed (as a MLTensor)? (wacky6) 14:58:46 #270 14:58:47 https://github.com/webmachinelearning/webnn/issues/270 -> Issue 270 Support coordinate transformation modes for Resample2d (Honry) 15:01:06 dom: I think we need to address these two by CR, or confirm we can defer them without creating breaking changes down the line 15:01:56 ... need to understand backwards compat story 15:02:42 RRSAgent, draft minutes 15:02:42 I have made the request to generate https://www.w3.org/2022/08/25-webmachinelearning-minutes.html anssik 17:12:39 Zakim has left #webmachinelearning