15:01:36 RRSAgent has joined #webmachinelearning 15:01:36 logging to https://www.w3.org/2019/10/31-webmachinelearning-irc 15:01:42 Zakim has joined #webmachinelearning 15:01:47 RRSAgent, make logs public 15:03:11 jdarpinian has joined #webmachinelearning 15:03:38 baul has joined #webmachinelearning 15:03:53 Meeting: WebML CG Teleconference – 31 October 2019 15:04:01 Chair: Anssi 15:04:18 Agenda: https://github.com/webmachinelearning/meetings/blob/master/telcons/2019-10-31-agenda.md 15:04:25 scribeNick: anssik 15:04:32 Present+ Anssi_Kostiainen 15:04:53 Present+ James_Darpinian 15:05:04 dennis has joined #webmachinelearning 15:06:23 Present+ Baul_Eun 15:06:33 Present+ Dennis_Yurkevich 15:07:52 RRSAgent, draft minutes v2 15:07:52 I have made the request to generate https://www.w3.org/2019/10/31-webmachinelearning-minutes.html anssik 15:09:09 Present+ Ningxin_Hu 15:09:12 RRSAgent, draft minutes v2 15:09:12 I have made the request to generate https://www.w3.org/2019/10/31-webmachinelearning-minutes.html anssik 15:09:33 GabeEsteves has joined #webmachinelearning 15:09:55 Present+ Gabe_Esteves 15:10:24 Happy Halloween! 🎃 It seems we lost a bunch of folks due to halloween celebrations. 👻 15:10:43 TOPIC: WebNN interop investigation 15:11:11 anssik: Review POC requirements for buffer sharing between dedicated ML hardware and GPU 15:11:20 ... Ningxin has done investigation with Paul 15:12:04 ningxinhu: no slides to share, but can update on the status and next steps 15:12:27 ... we need to investigate buffer sharing and performance 15:12:57 ... we reported at F2F the results from WebGPU 15:13:27 ... use case is to share WebGPU buffer, and run shader on GPU device and interact with ML accelerator 15:13:41 ... we have Chromium WebNN POC implementing the foundation API 15:14:00 ... this is based on native platform APIs, decided to use DirectML as a platform to do this investigation 15:14:36 ... with DirectML we have WebGPU with D3D12 backend 15:14:44 jdarpinian: at least Dawn backend is confirmed 15:15:08 ningxinhu: WebGPU support with Chromium fork rebased to 79 so will check WebGPU support status 15:15:36 ... in our fork WebNN works on top of DirectML, also supports Intel's VPUs on some devices 15:16:10 ... used AI devkit, a small PC with VPU, PCIe interface for the investigation 15:16:24 ... got help from Paul McDaniel of Msft 15:17:19 https://docs.microsoft.com/en-us/windows/win32/dxcore/dxcore-enum-adapters 15:17:32 ... DirectML only in Windows insider builds, so testing with those preview releases, API enumerating accelerators only works on GPU devices, new interface dxcore in the insider builds only 15:19:07 ningxinhu: to summarize, we get GPU device to work on WebNN DirectML backend via DXCore, preliminary testing demonstrated it works 15:20:04 ... VPU functionality confirmed, interop testing with WebGPU needs more work, is a next step 15:20:43 ... can WebGPU compute shaders run on dedicated accelerators, VPU? 15:21:57 ... feedback requested from community on the interop testing approches 15:23:08 anssik: can you open an issue to document the interop testing results to date and proposed next steps? 15:23:11 ningxinhu: can do that 15:23:42 jdarpinian: no plans to support VPUs via WebGPU I suspect, focused on GPU 15:24:41 ningxinhu: VPU just an example, WebGPU built on top of D3D12, so can WebGPU shaders run on compute-only devices? 15:25:00 jdarpinian: I think eventually yes, but not for the WebGPU v1, not 100% sure though 15:26:02 jdarpinian: any idea of relative perf of VPU and GPU in this machine tested on? 15:26:59 ningxinhu: we can run some models with GPU and VPU, GPU slightly faster than VPU currently, SqueezeNet <10 ms, VPU >10 ms 15:28:11 ningxinhu: the performance is one angle, VPU more power efficient 15:28:44 RRSAgent, draft minutes v2 15:28:44 I have made the request to generate https://www.w3.org/2019/10/31-webmachinelearning-minutes.html anssik 15:30:26 ningxinhu: jdarpinian mentioned WebGPU extensions a while ago, any updates on that? 15:30:45 jdarpinian: WebGPU CG is open to having ML extensions, but they're not going to develop that themselves, would need to be done in this group 15:30:58 s/this group/WebML CG/ 15:31:13 ... I still think it would be a faster, simpler path, but less capable 15:34:16 TOPIC: Op compatibility 15:34:20 -> https://github.com/webmachinelearning/webnn/issues/28 [op compatibility] conv2d 15:34:39 ningxinhu: 4 opens for conv2d to be resolved 15:35:38 -> https://github.com/webmachinelearning/webnn/issues/27 [op compatibility] matMul 15:36:09 anssik, you are breaking up 15:36:16 it is very difficult for me to hear you 15:36:32 ningxinhu: not much bandwidth to help with native mapping for matMul 15:37:15 TOPIC: Explainer 15:37:22 -> https://github.com/webmachinelearning/webnn/blob/master/explainer.md Web Neural Network API Explained 15:37:41 anssik: Explainer is a landing zone for early ideas, when formalized land in spec 15:37:50 ... workflow: new issue -> discussion -> early consensus -> add to explainer and/or spec -> add to spec 15:38:31 ... couple of open PRs waiting to be reviewed: 15:38:31 https://github.com/webmachinelearning/webnn/pulls 15:42:12 GabeEsteves has joined #webmachinelearning 15:42:41 TOPIC: On-device training (exploratory topic) 15:42:57 anssik: Fukuoka F2F had training capability agenda topic, but skipped due to lack of time 15:43:10 ... wanted to discuss key use cases, feasibility, platform support (Core ML 3, ONNX 1.7 ...) 15:43:22 ... currently training out of scope, so need group agreement to work on this 15:44:16 GabeEsteves: I don't have info on platform support, but I have use cases for API support 15:44:51 ... use cases transfer learning and personalization area, customizing the last layers of the network 15:45:06 ... I know e.g. Core ML and TF have had this capability since beginning 15:45:19 ... ONNX 1.7 will also get this, expected to be released this month 15:46:01 ... industry is moving to this direction, so perhaps the group should pay attention and track this area 15:46:26 The APIs in scope of this group will not be tied to any particular platform and will be implementable on top of existing major platform APIs, such as Android Neural Networks API, Windows DirectML, and macOS/iOS Metal Performance Shaders and Basic Neural Network Subroutines. 15:48:46 GabeEsteves: belem has more use cases in mind, specific concrete cases are recommendation engine personalization, speaker recognition, in order the web to be fully featured platform this capability would be useful 15:49:10 ... can put this GH issue for further feedback and comments 15:50:56 anssik: propose the group focuses on inference near-term and keep an eye on training capability developments, related platform APIs etc. 15:51:03 gabe: agree with that positioning 15:52:40 RRSAgent, draft minutes v2 15:52:40 I have made the request to generate https://www.w3.org/2019/10/31-webmachinelearning-minutes.html anssik 15:53:09 TOPIC: Adjourn 15:53:16 RRSAgent, draft minutes v2 15:53:16 I have made the request to generate https://www.w3.org/2019/10/31-webmachinelearning-minutes.html anssik 17:21:13 Zakim has left #webmachinelearning 18:08:19 zkis has joined #webmachinelearning 18:25:24 zkis has joined #webmachinelearning 19:15:58 zkis has joined #webmachinelearning