Skip to content

vMLX 1.5.55

Choose a tag to compare

View all tags
@jjang-ai jjang-ai released this 05 Jun 04:26
· 1077 commits to main since this release
v1.5.55
1afe210

vMLX 1.5.55 emergency runtime release.

Highlights:

  • Fixes Qwen3.6 35B MXFP8 MTP gated-delta-net gdn_sink crash.
  • Fixes packaged video processing fallback for Gemma 4, Qwen MTP, Step 3.7 Flash, and Nemotron Omni paths.
  • Preserves deterministic Step 3.7 video contact-sheet cache reuse for exact repeat media+prompt.
  • Includes release-gate fixes for deferred objective rows and Developer ID entitlement parsing.

Verification:

  • Packaged release gate passed on v1.5.55 Sequoia app with Gemma 4 12B MXFP4: OpenAI Chat, Responses, Anthropic, Ollama, multi-turn recall, prefix cache hit, cache stats, and JIT soft wake.
  • Packaged video/MTP matrix passed for Qwen3.6 27B/35B MTP, Gemma 4 12B MXFP4/MXFP8/JANG_4M, Step 3.7 Flash JANG_2L, and Nemotron Omni JANGTQ4.
  • Both Sequoia and Tahoe DMGs are Developer ID signed, notarized, stapled, Gatekeeper accepted, and verified.
Assets 6
Loading