WebML WG Teleconference – 14 November 2024 - 15:00-16:00 UTC

Latest version: https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-11-14-wg-agenda.md


WebML WG Teleconference – 14 November 2024 - 15:00-16:00 UTC
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-11-14-wg-agenda.md#webml-wg-teleconference--14-november-2024---1500-1600-utc>
See the timezone table ...
San Francisco   Thu, 14 November 2024   07:00
Boston  Thu, 14 November 2024   10:00
London  Thu, 14 November 2024   15:00
Berlin  Thu, 14 November 2024   16:00
Helsinki        Thu, 14 November 2024   17:00
Shanghai        Thu, 14 November 2024   23:00
Tokyo   Fri, 15 November 2024   00:00
UTC     Thu, 14 November 2024   15:00 UTC


Other locations: https://www.timeanddate.com/worldclock/fixedtime.html?iso=20241114T15<https://www.timeanddate.com/worldclock/fixedtime.html?iso=20241114T15>


Note


The meeting starts 1 hour later in locations that do not change their clocks, see Daylight Saving Time<https://www.timeanddate.com/time/dst/events.html>. The timezone table has been adjusted accordingly.

Logistics

<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-11-14-wg-agenda.md#logistics>

  *   Chair: Anssi
  *   Scribe: ? (see howto<https://github.com/webmachinelearning/meetings/blob/main/scribe-howto.md>, volunteers welcome!)
  *   IRC: irc://irc.w3.org:6667/#webmachinelearning

  *   IRC web client: https://irc.w3.org/?channels=#webmachinelearning

  *   Joining instructions: https://lists.w3.org/Archives/Member/internal-webmachinelearning/2023Jun/0000.html

  *   Minutes: https://www.w3.org/2024/11/14-webmachinelearning-minutes.html


Agenda

ℹ️ Call for review: WebML Community Group Charter update

<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-11-14-wg-agenda.md#%E2%84%B9%EF%B8%8F-call-for-review-webml-community-group-charter-update>

Important

This concerns the Community Group (CG) charter, a group that incubates new web spec proposals. This does not suggest changes to the Working Group (WG) scope.


  *   Call for review: https://lists.w3.org/Archives/Public/public-webmachinelearning/2024Nov/0000.html

  *   Joining the CG: https://webmachinelearning.github.io/community/#join


ℹ️ Device selection abstractions
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-11-14-wg-agenda.md#%E2%84%B9%EF%B8%8F-device-selection-abstractions>


Continue discussion<https://www.w3.org/2024/10/31-webmachinelearning-minutes.html#f739> on the generalized device selection proposal (MLOpSupportLimits), considered alternatives (mint more device-agnostic deviceType terms?), security and privacy considerations (enumeration fingerprinting), use cases.

  *   ⨀ webmachinelearning/webnn#749<https://github.com/webmachinelearning/webnn/issues/749>
  *   ☰ Initiate work on an explainer.

ℹ️ MLTensor
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-11-14-wg-agenda.md#%E2%84%B9%EF%B8%8F-mltensor>


The group is gathering implementation experience on the MLTensor design to inform an upcoming specification update. The explainer is considered the source of truth in this prototyping phase.

  *   ☰ Discuss open questions<https://github.com/webmachinelearning/webnn/blob/main/mltensor-explainer.md#open-questions> as required to unblock progress.

ℹ️ Core op set & MLIR Linalg mapping
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-11-14-wg-agenda.md#%E2%84%B9%EF%B8%8F-core-op-set--mlir-linalg-mapping>


Review mapping from WebNN ops to MLIR Linalg Dialect and discuss key findings.

  *   ⨀ webmachinelearning/webnn#573<https://github.com/webmachinelearning/webnn/issues/573>
  *   ☰ MLIR Linalg Dialect<https://mlir.llvm.org/docs/Dialects/Linalg/>
  *   See also: 📁 TPAC slides<https://lists.w3.org/Archives/Public/www-archive/2024Sep/att-0007/Tensor_Primitive_Ops_Proposal_-_TPAC.pdf>

🏷️ feature request<https://github.com/webmachinelearning/webnn/labels/feature%20request> issues
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-11-14-wg-agenda.md#%EF%B8%8F-feature-request-issues>

  *

Device Memory Management Primitives

     *   ⨀ webmachinelearning/webnn#780<https://github.com/webmachinelearning/webnn/issues/780>
     *   ☰ Early proposal to reduce GPU/NPU memory, memory copy overhead, enable pipelining; all feedback welcome.
  *   Report non-fatal errors from the WebNN timeline

     *   ⨀ webmachinelearning/webnn#778<https://github.com/webmachinelearning/webnn/issues/778>
     *   ☰ Review proposal, address open questions.
  *   Support reverse operator

     *   ⨀ webmachinelearning/webnn#773<https://github.com/webmachinelearning/webnn/issues/773>
     *   ☰ Reverses the order of the input tensor along specified axes, improves performance of PyTorch models.
  *   Support strides option for slice operator

     *   ⨀ webmachinelearning/webnn#772<https://github.com/webmachinelearning/webnn/issues/772>
     *   ☰ Strides option with performance benefit, include also negative strides?

🏷️ operator specific<https://github.com/webmachinelearning/webnn/labels/operator%20specific> issues


  *   Support block-wise quantization

     *   ⨀ webmachinelearning/webnn#779<https://github.com/webmachinelearning/webnn/issues/779>
     *   ☰ Allows input tensors be divided into smaller independently quantized blocks, used by SLMs.
  *   Remove or redesign triangular

     *   ⨀ webmachinelearning/webnn#768<https://github.com/webmachinelearning/webnn/issues/768>
     *   ☰ All backends emulate triangular, discuss specific mask behaviour.

🏷️ interop<https://github.com/webmachinelearning/webnn/labels/interop> issues

<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-11-14-wg-agenda.md#%EF%B8%8F-interop-issues>

  *   Remove pool2d MLRoundingType
     *   ⨀ webmachinelearning/webnn#324<https://github.com/webmachinelearning/webnn/issues/324>
     *   ⛙ webmachinelearning/webnn#770<https://github.com/webmachinelearning/webnn/pull/770>
     *   ☰ Review PR.