WebML WG Teleconference – 5 October 2023 - 14:00-15:00 UTC

Latest version: https://github.com/webmachinelearning/meetings/blob/main/telcons/2023-10-05-wg-agenda.md


WebML WG Teleconference – 5 October 2023 - 14:00-15:00 UTC<https://github.com/webmachinelearning/meetings/blob/main/telcons/2023-10-05-wg-agenda.md#webml-wg-teleconference--5-october-2023---1400-1500-utc>
See the timezone table ...
Logistics<https://github.com/webmachinelearning/meetings/blob/main/telcons/2023-10-05-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/2023/10/05-webmachinelearning-minutes.html


Agenda<https://github.com/webmachinelearning/meetings/blob/main/telcons/2023-10-05-wg-agenda.md#agenda>
ℹ️ WebNN v2: Review op breakdown for proposed model targets<https://github.com/webmachinelearning/meetings/blob/main/telcons/2023-10-05-wg-agenda.md#%E2%84%B9%EF%B8%8F-webnn-v2-review-op-breakdown-for-proposed-model-targets>

The WG has identified the following as its v2 model targets:

  *   Text-to-image: Stable Diffusion unet/VAE/text encoder
  *   Image segmentation: Segment Everything decoder
  *   Speech-to-text: Whisper Tiny
  *   Text-to-text generation (encoder-decoder): t5 and m2m100
  *   Text-generation (decoder-only): llama

We will review the initial op breakdown to better understand what is common across these architectures to inform WebNN API v2 priorities.

Please provide your feedback via webmachinelearning/webnn#375<https://github.com/webmachinelearning/webnn/issues/375>

ℹ️ Security considerations<https://github.com/webmachinelearning/meetings/blob/main/telcons/2023-10-05-wg-agenda.md#%E2%84%B9%EF%B8%8F-security-considerations>

  *   Computation control-flow attack based on weights / constants change
     *    issue: webmachinelearning/webnn#443<https://github.com/webmachinelearning/webnn/issues/443>

ℹ️ New features<https://github.com/webmachinelearning/meetings/blob/main/telcons/2023-10-05-wg-agenda.md#%E2%84%B9%EF%B8%8F-new-features>

  *   Allow checking whether operators/types are supported for a backend before creating a graph
     *    issue: webmachinelearning/webnn#463<https://github.com/webmachinelearning/webnn/issues/463>

ℹ️ Enhancements<https://github.com/webmachinelearning/meetings/blob/main/telcons/2023-10-05-wg-agenda.md#%E2%84%B9%EF%B8%8F-enhancements>

  *   split() into sizes not widely supported
     *    issue: webmachinelearning/webnn#392<https://github.com/webmachinelearning/webnn/issues/392>
  *   Clarify the restriction for minValue and maxValue of MLClampOptions
     *    issue: webmachinelearning/webnn#396<https://github.com/webmachinelearning/webnn/issues/396>
  *   Type of parameters should match the input data type
     *    issue: webmachinelearning/webnn#442<https://github.com/webmachinelearning/webnn/issues/442>