- From: Kostiainen, Anssi <anssi.kostiainen@intel.com>
- Date: Thu, 21 Sep 2023 16:05:39 +0000
- To: "public-webmachinelearning-wg@w3.org" <public-webmachinelearning-wg@w3.org>
Hi All, Thanks for another great discussion. Today’s minutes at: https://www.w3.org/2023/09/21-webmachinelearning-minutes.html Here’s a summary of what was discussed: We agreed to use the following models as the basis for our WebNN v2 work, next advancing to the op breakdown analysis: • 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 had a discussion around Storage APIs for caching large models and observed the currently broadly shipping File System API (aka OPFS), Storage API, Cache API work quite well for caching large models. Sharing a cached model across domains remains an unsolved issue. We also heard updates from W3C TPAC 2023 and Intel Innovation that took place last week and this week. This WG’s work was well received at both the events. Keep up the great work! Thanks, -Anssi (WebML WG chair)
Received on Thursday, 21 September 2023 16:13:41 UTC