Skip to content

mere.run 0.29.1

Latest
Latest

Choose a tag to compare

View all tags
@a-github-name a-github-name released this 30 Jul 02:35
v0.29.1
8a3fd49
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

mere.run 0.29.1

This release brings native speaker diarization to Apple Metal and NVIDIA CUDA,
and makes the packaged release gate exhaustive across every installed model.

Added

  • Native MLX Streaming Sortformer speaker diarization through
    mere.run speech diarize, with managed model installation, JSON and RTTM
    output, timed speaker segments, and speaker reidentification.
  • gate --all-installed --require-all, a fail-closed packaged-runtime gate that
    discovers the exact installed catalog inventory and requires real inference,
    decoded artifacts, non-silent audio, and semantic output.

Changed

  • Removed Infinity Flash OCR from the catalog, resolver, Studio defaults,
    documentation, and release inventory. Infinity OCR now defaults to the
    validated Pro int8 model.
  • The exhaustive diarization release check now uses the same real A -> B -> A
    speaker fixture as the dedicated Metal and CUDA checkpoints.

Fixed

  • Bundled the SwiftPM resource bundles required by Gemma-family inference in
    the packaged macOS helper.
  • Hardened the release gate against chatty subprocess deadlocks, fast-process
    termination races, incomplete model manifests, silent media, and invalid
    production inputs.

Downloads

  • MereRun-0.29.1.dmg — signed, notarized, and stapled macOS Studio and CLI
    for Apple Silicon, with Sparkle updates.
  • mere-run-0.29.1-linux-arm64.tar.gz and
    mere-run_0.29.1_arm64.deb — native CUDA build for DGX Spark / GB10.
  • mere-run-0.29.1-linux-x86_64-cuda.tar.gz and
    mere-run-cuda_0.29.1_amd64.deb — native CUDA build for x86_64 NVIDIA hosts.
  • SHA256SUMS and platform-specific checksum manifests.

Release validation

  • Exact source tag: v0.29.1, commit
    8a3fd4954b3b9c0e1971d00122fb54c99c2444e4.
  • All applicable GitHub checks passed: 11 successful checks; dependency review
    was skipped outside its pull-request execution context.
  • macOS build 683: disk-image integrity, Apple notarization, stapling,
    Gatekeeper, strict nested code signing, and fresh-mount version checks passed.
  • The exact macOS DMG completed all 59 installed-model packaged-runtime checks
    with no skips or failures.
  • Linux ARM64 on NVIDIA GB10: archive and Debian package checksums, bundled
    library closure, version/catalog checks, real Bonsai binary and ternary image
    generation (95-96% peak GPU), native CUDA Sortformer diarization with two
    speakers and repeated A -> B -> A transitions, and the 0.28.1 -> 0.29.1
    Debian install/upgrade/remove/reinstall lifecycle all passed.
  • Linux x86_64 on an NVIDIA RTX 3080 Ti: archive and Debian package checksums,
    bundled library closure, version/catalog checks, real Krea 2 Turbo image
    generation (83% peak GPU), native CUDA Sortformer diarization with two
    speakers and repeated A -> B -> A transitions, and the 0.28.1 -> 0.29.1
    Debian install/upgrade/remove/reinstall lifecycle all passed.
Assets 11
Loading