Skip to content

vMLX v1.5.47

Choose a tag to compare

View all tags
@jjang-ai jjang-ai released this 22 May 17:07
· 3217 commits to main since this release
v1.5.47
f9d4685

vMLX 1.5.47

Highlights:

  • Expanded release-regression gates for max output vs max context wiring, Responses output caps, generation defaults, parser parity, cache architecture, native MTP, MCP, VL media cache, and model-family launch policy.
  • Added stronger parser boundaries so non-reasoning families such as Qwen2/Qwen2-VL do not silently inherit reasoning parsers while MiniMax, DSV4, ZAYA, Hy3, Gemma, and GLM parser rows stay covered.
  • Added DSV4-specific release guards for stale native-MTP launch args, cache/prefix policy, and clean JANG source parity.
  • Integrated and credited the safe PR slices for UDS serving, coding-tool path hardening, prefill-loop cleanup, SSM/cache hot-path cleanup, and compatibility contracts.

Downloads:

  • This updater manifest points to the Sequoia-compatible DMG, which is the broadest supported public build.
  • The GitHub release also includes a Tahoe-native DMG for macOS Tahoe users who want native macosx_26 MLX wheels.

Verification:

  • Sequoia and Tahoe DMG containers were submitted for notarization, stapled, and accepted.
  • Mounted apps from both DMGs report 1.5.47 and pass strict codesign verification.
  • Packaged app release gate passed version, twine check, panel request/type tests, TypeScript typecheck, bundled Python/JANG source hash checks, packaged imports, pycache cleanup, Developer ID signature, strict codesign, and spctl.

Known follow-up:

  • DSV4 Flash long full-output/code-generation quality remains a documented release exception pending separate exact-code/runtime-quality clearance.

DMG SHA256:

  • Sequoia DMG (vMLX-1.5.47-sequoia-arm64.dmg): b025f7c369561dda70cf566618991f3c158c5b8af5351dc04307f3bbc9aae8f1
  • Tahoe DMG (vMLX-1.5.47-tahoe-arm64.dmg): 8521976a07de9c33ae0ebec60c37cd41d99bce5dd9ef1d983cb4599b8c0536f5
Assets 6
Loading