WebML WG Teleconference – 5 September 2024 - 14:00-15:00 UTC

Latest version: https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-09-05-wg-agenda.md


WebML WG Teleconference – 5 September 2024 - 14:00-15:00 UTC
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-09-05-wg-agenda.md#webml-wg-teleconference--5-september-2024---1400-1500-utc>
See the timezone table ...

Logistics
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-09-05-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/2024/09/05-webmachinelearning-minutes.html


Agenda
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-09-05-wg-agenda.md#agenda>

TPAC F2F agenda published
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-09-05-wg-agenda.md#tpac-f2f-agenda-published>


TPAC F2F agenda has been published. We'll do fine-tuning and gardening: confirm our topic leads, timeboxes, discuss group contributions that'd be welcome by F2F.


  *   🗓️ #25<https://github.com/webmachinelearning/meetings/issues/25>

ℹ️ Device selection abstractions
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-09-05-wg-agenda.md#%E2%84%B9%EF%B8%8F-device-selection-abstractions>


📌 We agreed to evolve MLContextOptions and other API controls for device selection informed by further implementation experience and new use cases from the wider web community.


Discuss how frameworks and backends approach configuring inference sessions, understand framework-level use cases and requirements. Distill learnings to help evolve MLContextOptions.


  *   ONNX Runtime Web

     *   Session Options<https://onnxruntime.ai/docs/tutorials/web/env-flags-and-session-options.html#session-options> / API<https://onnxruntime.ai/docs/api/js/interfaces/InferenceSession.SessionOptions.html>
     *   WebNN EP Session Options<https://onnxruntime.ai/docs/tutorials/web/ep-webnn.html#how-to-use-webnn-ep-in-onnx-runtime-web> / API<https://onnxruntime.ai/docs/api/js/types/InferenceSession.WebNNExecutionProviderOption.html>
     *   Other EPs<https://github.com/microsoft/onnxruntime/blob/main/js/common/lib/inference-session.ts>
  *   WebKit / CoreML

     *   ⨀ webmachinelearning/webnn#749<https://github.com/webmachinelearning/webnn/issues/749>
  *   WebNN API

     *   Current device selection mechanism: device selection<https://www.w3.org/TR/webnn/#programming-model-device-selection>, MLContextOptions<https://www.w3.org/TR/webnn/#dictdef-mlcontextoptions>
     *   🏷️ All device selection<https://github.com/webmachinelearning/webnn/labels/device%20selection> issues

ℹ️ MLTensor
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-09-05-wg-agenda.md#%E2%84%B9%EF%B8%8F-mltensor>


📌 MLTensor is the new MLBuffer! Review the explainer PR with focus on the overall design, open questions. Spec changes including IDL will come in separate PRs. Any IDL is tentative and is subject to change. Close webgpu interop issues that have been addressed.


  *   ⛙ webmachinelearning/webnn#754<https://github.com/webmachinelearning/webnn/pull/754>
  *   🏷️ All webgpu interop<https://github.com/webmachinelearning/webnn/labels/webgpu%20interop> issues

ℹ️ MLConstantOperand
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-09-05-wg-agenda.md#%E2%84%B9%EF%B8%8F-mlconstantoperand>


📌 An MLConstantOperand is an MLOperand which represents constant weight data, allows for optimizations such as weight preprocessing. Discuss design considerations: operand vs. isConstant, enforce weights as MLConstantOperand, backend and framework limitations.

PR expected to evolve after settling on the design.


  *   ⨀ webmachinelearning/webnn#668<https://github.com/webmachinelearning/webnn/issues/668>
  *   ⛙ webmachinelearning/webnn#747<https://github.com/webmachinelearning/webnn/pull/747>

ℹ️ Open issues and PRs

<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-09-05-wg-agenda.md#%E2%84%B9%EF%B8%8F-open-issues-and-prs>

We'll discuss new issues, debrief the group on merged PRs and review open PRs since our last meeting:


  *   ⨀ All open issues<https://github.com/webmachinelearning/webnn/issues>
  *   ⛙ All open pull requests<https://github.com/webmachinelearning/webnn/pulls>
  *   ⛙ Recently merged PRs<https://github.com/webmachinelearning/webnn/pulls?q=is%3Apr+is%3Amerged>
  *

Priority is given to issues and PRs proposed for discussion in the following issue:


  *   🆕 #28<https://github.com/webmachinelearning/meetings/issues/28>


Time permitting, we'll pick from the following list of issues that have received your attention recently:


🏷️ feature request<https://github.com/webmachinelearning/webnn/labels/feature%20request>
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-09-05-wg-agenda.md#%EF%B8%8F-feature-request>

  *   LocalResponseNormalization (LRN) operation

     *   ⨀ webmachinelearning/webnn#228<https://github.com/webmachinelearning/webnn/issues/228>
     *   ☰ Discuss prototyping findings.
  *   WebNN should support int8 quantized models

     *   ⨀ webmachinelearning/webnn#128<https://github.com/webmachinelearning/webnn/issues/128>
     *   ☰ Discuss backend differences for proposed new quant ops.

🏷️ question<https://github.com/webmachinelearning/webnn/labels/question>
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-09-05-wg-agenda.md#%EF%B8%8F-question>

  *   Don't transfer input ArrayBuffers

     *   ⨀ webmachinelearning/webnn#566<https://github.com/webmachinelearning/webnn/issues/566>
     *   ☰ Discuss BYOB readBuffer() implementation experience.
  *   lstm and gru CoreML implementation feedback

     *   ⨀ webmachinelearning/webnn#751<https://github.com/webmachinelearning/webnn/issues/751>
     *   ☰ Review the proposed changes on how to pass bias, weights, activations.

🏷️ operator specific<https://github.com/webmachinelearning/webnn/labels/operator%20specific>
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2024-09-05-wg-agenda.md#%EF%B8%8F-operator-specific>

  *   Consider adding int64/uint64 data type support for some reduce operators

     *   ⨀ webmachinelearning/webnn#694<https://github.com/webmachinelearning/webnn/issues/694>
     *   ⛙ webmachinelearning/webnn#695<https://github.com/webmachinelearning/webnn/pull/695>
     *   ☰ See PR discussion. Feedback on conditional support for data types?
  *   Limited support for pad on Core ML backend

     *   ⨀ webmachinelearning/webnn#739<https://github.com/webmachinelearning/webnn/issues/739>
     *   ☰ Discuss Core ML backend constraints & address questions.