Ollama: MLX Goes Parallel
MLX models gain concurrent generation support alongside backend version bumps, while Vulkan memory handling and API correctness fixes land across the server and app. The dominant pattern is making alternative runners production-ready.
Duration: PT2M15S
Episode overview
This episode is a short developer briefing from Ollama.
It explains recent repository work in plain language.
- Show: Ollama
- Published: 2026-11-11T14:01:20Z
- Audio duration: PT2M15S
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 November 11th, 2026 — this is your Ollama briefing.
The lead story is that MLX is now a true multi-request server path. Pull request 19197 lets MLX models handle parallel generation through the parallel request setting, cutting first-token wait by sharing projection work while keeping attention, caches, sampling and streams independent.
That batching work is backed by dependency updates. Pull requests 19647 and 19646 bump the MLX stack and llama dot cpp, keeping the two runner families in step. For developers, the impact is direct: you can raise parallelism from the default of one, but you will need memory for each active request, and the scheduler…
A second theme is hardware reliability and runner clarity. Pull request 19640 switches Vulkan integrated graphics to memory-mapped loading, fixing load failures where CPU tensors and weights competed for the same device budget on shared-memory systems, including pinned-memory errors on Windows AMD. Related, pull…
A third theme is API correctness. Pull request 19648 removes hardcoded model-name checks for thinking controls in the app, reading support from the show endpoint instead. Pull request 19651 fixes dropped tool…
Wh…
Nearby episodes from Ollama
- Tool Schema Fidelity, Desktop Launch, and Download Safety
- Tool Reliability and Mac Inference Tuning
- Faster Concurrent Inference and Unified Desktop Launch
- VRAM Fixes and Launch Polish
- Stricter Model Imports and CLI Output Fixes
- Weekly Recap - MLX Model Expansion and Launch Polish
- Ripple on MLX and Validation Hardening
- MLX Memory Cuts and Tool-Call Fixes