WebML WG Teleconference – 2 November 2023 - meeting minutes

Hi All,

Thanks for joining us. Today we discussed the latest developments in the transformer ops and were the first to hear the exciting news about a new distilled version of the already best-in-class speech-to-text model (hint: Distil-Whisper, look for the pointer in the minutes!).

We also discussed a number of WebNN API enhancements informed by the diverse implementation experience our participants continue to bring to the group from both web engine and framework implementation efforts. Please keep this implementation experience feedback coming!

Please see the minutes for more details and pointers:

https://www.w3.org/2023/11/02-webmachinelearning-minutes.html


Topics discussed:

• WebNN v2: Review transformer ops spec contributions (continued)
• Enhancements
• 0D scalars
• Softmax axis absent
• split() into sizes not widely supported
• Clarify the restriction for minValue and maxValue of MLClampOptions

Thanks,

-Anssi (WebML WG chair)