Release v1.5.0

Added

• ml.threaded option: shorthand for numThreads: 4. On a cross-origin isolated
  host page (COOP: same-origin + COEP: require-corp), this gives roughly
  2x faster ML inference. Without isolation it falls back to running on 1
  thread automatically, so it is safe to request speculatively.

Changed

• scanic-ml now ships a single wasm build (0.2.0), down from 6.9 MB to
  3.5 MB unpacked, roughly half. There used to be two nearly identical wasm
  builds (a dedicated single-thread one and a pthread-capable one); running
  the pthread build on 1 thread costs about 4% versus the dedicated build,
  noise-level, so shipping both cost close to a full extra copy of the wasm
  for no real benefit. If you self-host the scanic-ml assets, update to
  scanic-ml@0.2.0; the file layout changed (dist/ort-wasm-simd.wasm and
  dist/threaded/ are gone).

Installation

npm install scanic@1.5.0

Or via CDN:

<script src="https://unpkg.com/scanic@1.5.0/dist/scanic.js"></script>

Full Changelog: v1.3.0...v1.5.0