Ollama: Weekly Recap - MLX Hardens as a First-Class Engine
This week's work centered on making Ollama's MLX engine production-grade — stricter error handling, speculative decoding under structured output, and new multimodal support — alongside a cluster of fixes for context length handling and continued build-out of the developer tool launcher ecosystem.
Duration: PT3M11S
Episode overview
This episode is a short developer briefing from Ollama.
It explains recent repository work in plain language.
- Show: Ollama
- Published: 2026-09-07T09:07:58Z
- Audio duration: PT3M11S
Transcript excerpt
This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.
Good week, Ollama listeners. This is your recap for August 31st through September 7th. Fifty pull request activity items and twenty-four additional commits landed this week.
The dominant story is MLX moving from experimental to hardened. Jesse Gross shipped a run of changes tightening the engine's error handling — checking every fallible call into the MLX bindings instead of silently continuing on failure, which previously showed up later as crashes or zeroed-out results. That's PR…
Supporting that push, Daniel Hiltgen's team invested in MLX test infrastructure: wiring MLX unit tests into pull request CI runs, so contributors can validate MLX behavior without a full rebuild, PR seventeen-oh-two-two, plus a follow-up to rebuild test payloads when a release build can't supply them, PR…
The second theme is context length correctness — a recurring bug pattern this week. Multiple independent contributors opened fixes for MLX runners ignoring the configured context and falling back to the architecture maximum, tracked under issue eighteen-one-two-five, with PRs eighteen-one-seven-four,…
Third, model-authored defaults and metrics got attention. PR sixteen-four-seven-one makes Ollama…
Finally,…
Nearby episodes from Ollama
- Tool-Call Parsing Gets a Hardening Pass
- Codex Integration Hardening and Streaming Reliability
- Tightening Up the OpenAI Compatibility Layer
- Closing the Gaps Between What You Configure and What Actually Runs
- Context Handling Gets Serious
- Codex Integration Push and Schema Hardening
- Structured Output Meets Speculative Decoding
- Hardening the MLX Engine and Fixing What Feeds the Scheduler