Skip to content

vMLX 1.6.18

Choose a tag to compare

View all tags
@jjang-ai jjang-ai released this 27 Jul 19:38
v1.6.18
ded3f9f

vMLX 1.6.18

vMLX 1.6.18 is an Electron/Python runtime and packaging checkpoint built from source commit ded3f9f02e324f9aa5df4f3a331b7f7d4df40d64.

Highlights

  • Keeps Auto reasoning aligned with each model bundle's native policy unless a request supplies an explicit supported thinking budget, while keeping reasoning-marker aliases out of visible content.
  • Scopes tool capabilities and authorization to the correct turn and improves multi-tool, cancellation, LFM, and tool-result continuation handling.
  • Accepts MiniMax tool arguments emitted as request-schema XML children without exposing native tool markup as visible answer text.
  • Retains media across Responses tool continuations and permits local attachment playback in the Electron app.
  • Derives fresh-session sampling, parser, output, template, and native MTP settings from the selected model bundle without converting inherited values into hidden saved overrides.
  • Adds a process-wide SSD cache budget with coordinated LRU eviction and off-request-path publication, plus clearer reporting of attempted prefixes, applied reuse, uncached suffixes, reconstruction, fallback prefill, and aggregate disk usage.
  • Prevents recycled paged-cache block IDs from receiving stale prefix credit by validating and atomically pinning the exact ordered cache chain before reuse.
  • Hardens Laguna tool-result history, incomplete terminals, mixed-SWA cache reuse, and affine-JANG JIT policy.
  • Improves KaTeX, Markdown, HTML, XML, currency, code-fence, and in-progress reasoning rendering, and stabilizes localized UI context during development reloads.
  • Preserves Ollama backend failures as failures and releases stale execution and MLX residency state across deep sleep and wake.
  • Strengthens Sequoia/Tahoe packaging so bundled Python, JANG dependencies, release metadata, and artifacts remain tied to one source revision. This release requires JANG 2.5.34.

This checkpoint does not claim exhaustive closure of every model-family, long-context, modality, cache-eviction, gateway/LAN recovery, or JANG conversion scenario. Those broader validation areas remain follow-up work.

Thanks to GitHub @Hornsan1 for reporting many of the runtime, model, UI, and API issues addressed in this checkpoint.

Assets 2
Loading