0.5.2rc1

This release candidate brings a full statistics layer to the macOS menu bar with live inference readouts, host CPU/GPU/memory monitoring, and a new Appearance pane, plus a reorganized Models menu and stability fixes across DFlash batching, custom-kernel loading, torch-free Mistral support, and oQ determinism.

Highlights

• Menu bar statistics. New opt-in status items show live, session-average, and all-time inference throughput (PP and TG in tk/s), plus CPU/GPU/MEM usage bars that merge into a single compact item. Each opens a popover with current values and rolling graphs, and a System Stats submenu adds CPU, GPU, and Memory panels with E/P core usage, load average, uptime, thermal state, GPU utilization, and a memory breakdown. Everything is configured from the new Appearance pane, and it is built to be cheap: glyphs re-render only when a reading visibly changes and host sampling runs only while something displays it. (#2303)
• Reorganized Models menu. The menubar Models menu now groups Loaded, Favorites, and Library sections with load state and sizes, and the Appearance pane can limit the list to favorites only. By @faisal in #1882

Bug Fixes

• Torch-free Mistral Small 4 / Devstral loading. The bundled runtime now loads Pixtral and Mistral3 checkpoints without torch, which also fixes Devstral tool calls returning empty messages. Reported by @raucodes in #2263 and @jomei64 in #2291
• DFlash no longer poisons unrelated requests. A failed DFlash arm left class-level hooks installed, corrupting concurrent batched decodes on models that share attention classes. Reported by @theuseless-ai in #2252
• Custom kernels load again on Homebrew installs. The kernel binaries shipped an unresolved @rpath/libmlx.dylib dependency and silently fell back to slow paths. Reported by @zhongyiyang1981 in #2233
• Deterministic oQ/oQe quantization. Calibration subsampling now uses a fixed permutation key, so repeated runs of the same recipe produce identical artifacts. Reported by @zviratko in #2293
• Fixed model swapping with the memory guard disabled: loading a second model now evicts instead of wedging the server. Reported by @raucodes in #2290
• Fixed engine leases accumulating after client disconnects until the server wedged. Reported by @jzupnick in #2260
• Fixed the VLM engine silently ignoring OptiQ vision sidecars on gemma4_unified. Reported by @donkuk in #2298
• Fixed oQe calibration on models that need the compatibility loader. By @seeker-cyber-maker in #2299
• Fixed the desktop UI ignoring a manual "embedding" model type override. Reported by @mveroone in #2270
• External benchmarks now support provider-specific extra body parameters and distinguish endpoint failures from wrong answers. By @earthking11 in #2296 and #2297

Upgrade Notes

• If you had "Show Live Activity in Menu Bar" enabled, it migrates automatically to the new LIV status item; the old inline text next to the menu bar icon is gone. All menu bar items are configured in Settings under Appearance.

New Contributors

Thank you to @faisal, @earthking11, and @seeker-cyber-maker for their first contributions to oMLX in this release.