Ollama: Mac Prefill Gains and Scheduler Fixes
Apple silicon work dominates, with faster sliding-window prefill and a fix for Metal watchdog kills, alongside smarter GPU scheduling and pull throttling. Desktop updates round out the day with auto-restart and launcher compatibility fixes.
Duration: PT2M5S
Episode overview
This episode is a short developer briefing from Ollama.
It explains recent repository work in plain language.
- Show: Ollama
- Published: 2026-10-26T13:01:05Z
- Audio duration: PT2M5S
Transcript excerpt
This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.
Good morning, it's Monday, October 26th, 2026, and this is your Ollama briefing.
The lead pattern is performance and reliability for long prompts on Mac, with two complementary changes landing together.
First, merged change number 19291 bumps the core Mac libraries to use native sliding-window attention. Instead of building a full causal plus sliding mask for prefill chunks, the runner now passes the window size directly to the fast attention path for Gemma 4, Flint and Laguna models. The result is higher prefill…
Second theme is resource fairness on the server side. Proposal 19284 fixes spurious evictions on multi-GPU hosts where the scheduler could evict a CPU-only embedding model to make room for a GPU load, even though that frees no video memory. The fix ensures CPU-only runners are never evicted to satisfy a graphics…
Finally, desktop polish. Proposal 19285 adds an automatic restart preference after an update downloads, with a database migration and a new checkbox in settings that skips the tray notification when enabled. Proposal 19298 fixes the OpenCode launch for version 2 by keeping the selected model in inline configuration…
What's next: Apple silicon developers should…
Nearby episodes from Ollama
- MLX Performance Push and Compatibility Fixes
- OpenAI Fidelity and Runner Stability
- LFM3 Speech, Video and Reliability Fixes
- Scheduler Memory Fixes and Faster Drafting
- Weekly Recap - Phonon Voice Goes GA and MLX Speed Gains
- Gemma 4 Reliability and Runner Parity
- Tool Call Fixes and Phonon Voice Update
- Phonon Voice Goes Stable and API Failures Get Fixed