Skip to content

vMLX 1.5.64

Choose a tag to compare

View all tags
@jjang-ai jjang-ai released this 18 Jun 13:50
· 58 commits to main since this release
v1.5.64
3e79bf5

vMLX 1.5.64

Highlights:

  • MiniMax-M3 compatibility update with stricter exact-output coverage for live UI, Chat Completions, Responses, Anthropic-compatible, Ollama-compatible, and streaming tool/image paths.
  • MiniMax-M3 startup keeps prefix cache enabled, paged cache disabled, disk prompt cache enabled, JIT disabled, and native MSA/Lightning sparse cache storage with generic TurboQuant KV disabled.
  • Gemma 4 E2B/E4B/12B/26B/31B JANG_4M VL and 26B/31B MXFP4 visual rows were refreshed with live UI/API/cache coverage.
  • Streaming reasoning-only responses now get a bounded visible-answer pass for supported reasoning models instead of appearing blank when the model emits hidden reasoning only.
  • Model-owned generation defaults continue to be surfaced in the UI and applied to API/chat startup paths.

Verification:

  • vMLX main source commit: cb4c768df.
  • vMLX release tag source commit: 701411284.
  • Post-DMG scoped MM3/Gemma preflight: status=pass, failures=[].
  • MM3 signed-app stress covered 10 UI turns, reasoning off/on/auto, exact tool calls, image input, long-context recall, prefix-cache hits, Chat/Responses/Anthropic/Ollama, and streaming tool/image exactness.
  • MM3 clean-start proof confirmed enablePrefixCache=true, usePagedCache=false, enableDiskCache=true, enableJit=false, toolCallParser=minimax_m3, reasoningParser=minimax_m3, and native minimax_m3_msa_v1 cache telemetry.
  • MM3 lifecycle proof confirmed abort/stop ended streaming and no extra assistant messages appeared during quiet windows.
  • Gemma 4 E2B/E4B/12B/26B/31B JANG_4M VL and 26B/31B MXFP4 visual signed-app stress covered 10 UI turns, reasoning off/on/auto, image input, tool calls, cache hits, Chat/Responses/Anthropic/Ollama, and streaming rows.
  • Sequoia and Tahoe DMGs are Developer ID signed, Apple notarized, stapled, and Gatekeeper accepted.

Notarization:

  • Sequoia notary id: e4181a30-a25e-4b30-92ef-5d026e2df62f.
  • Tahoe notary id: 35400c4d-be69-484d-bdb8-947cb988229f.

Artifacts:

  • Sequoia DMG SHA256: cb818bfc4082b65e5469bf04a3d4e4d4864938cbaf373c8ddc1a5ee0eb04b882
  • Tahoe DMG SHA256: 12a34c1e9b7e3ffc7c5dfe31899b469d305033c15d15d64f0b5601d14b5a3010

Scope:

  • This is a scoped MM3 + Gemma 4 VL compatibility release. Gemma audio remains out of this release gate. Broader model-family matrix work continues in the next release line.
Assets 4
Loading