Skip to content

vMLX 1.5.39

Choose a tag to compare

View all tags
@jjang-ai jjang-ai released this 17 May 22:58
· 3406 commits to main since this release
v1.5.39
ded7bb3

Release 1.5.39 focuses on native MTP detection and runtime gates.

  • Native MTP health/status now reports depth, scope, tensor counts, and gating reasons.
  • Supported Qwen 3.6 MTP artifacts default to D3; model-local tuning sidecars are opt-in.
  • Qwen 3.6 MTP+VL app detection now keeps artifact-backed affine-JANG MTP+VL bundles multimodal when indexed MTP and vision tensors are present.
  • Source/PyPI runtime patch parity is included for Kimi K2.6 MLA and Gemma 4 vision pixel-value handling.

Verified with Python source sweep, full panel tests/typecheck, bundled app verifier, signed/notarized DMG validation, mounted-DMG runtime probe, and real Qwen MTP artifact metadata checks.

Assets 2
Loading