Ollama: Hardening the MLX Engine and Fixing What Feeds the Scheduler
The MLX engine got a reliability pass, catching silent failures and adding image, audio, and cached-token reporting, while a cluster of scheduler and memory PRs worked to fix inaccurate VRAM and context-length numbers that drive placement decisions.
Duration: PT2M42S
Episode overview
This episode is a short developer briefing from Ollama.
It explains recent repository work in plain language.
- Show: Ollama
- Published: 2026-09-03T13:01:18Z
- Audio duration: PT2M42S
Transcript excerpt
This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.
This is Ollama, your development briefing for September third.
The biggest signal today: two parallel efforts to make Ollama's numbers trustworthy — the numbers the scheduler and users actually rely on.
First, the MLX engine got a serious reliability pass from Jesse Gross. PR 18052 found that MLX bindings were checking almost none of their return codes. A failed call would silently continue with a null result, only to surface later as zeros, skipped evaluations, or a crash somewhere unrelated. The fix, refined…
Second theme: getting resource numbers right. Daniel Hiltgen's PR 17943 adds cached prompt token reporting across native, OpenAI, and Anthropic-compatible APIs, so cache hits no longer inflate prefill rate calculations. Separately, parawanderer's stacked PRs — 18197, 18198, and 18201 — are working through per-device…
A few other notable items: PR 18212 fixes reasoning-budget cutoffs to end on a line break instead of mid-word, and 18205 restores shell tab-completion for model names, closing a gap open since the CLI's earliest days. Routine maintenance included a llama.cpp version bump to b10760 and a documentation cleanup on FAQ…
What to watch: the VRAM and context-length PRs…
Nearby episodes from Ollama
- MLX Grows Up, Model Intent Gets Respected
- Runner Reliability and a Documentation Cleanup Wave
- Proxy Fix and a Documentation Cleanup Sweep
- Weekly Recap - Claude Desktop Integration & MLX Engine Maturity
- Correctness Fixes Under the Hood
- Agent Expansion, Desktop Polish, and a Documentation Sweep
- MLX Engine Grows Up
- Desktop App Stability and the Claude Integration Cleanup