Ollama: Thinking Separation Fix and GPU Visibility
Three small targeted changes with no single sweeping theme. The highest-impact fix corrects reasoning output leaking into responses in the generate API, alongside a Docker Vulkan enablement and new GPU monitoring docs.
Duration: PT2M10S
Episode overview
This episode is a short developer briefing from Ollama.
It explains recent repository work in plain language.
- Show: Ollama
- Published: 2026-09-21T13:02:03Z
- Audio duration: PT2M10S
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, September 21st, 2026, and this is your Ollama developer briefing.
Today's activity is scattered rather than thematic, with three narrow changes, and the highest-impact fix corrects how reasoning traces leak into user-facing responses.
First, cleaner thinking separation. In pull request 18561, linked to issue 18554, the generate endpoint could leave reasoning tags in the main response when using a built-in parser without thinking support. The fix composes that built-in parser with the generic tag extractor, so thinking and response stay separated…
Second, better graphics handling and visibility. Pull request 18562, fixing issue 17549, adds the graphics capability to the Docker image's NVIDIA driver setting. The image already shipped the Vulkan backend, but without that flag the container runtime wouldn't mount host graphics drivers, blocking Vulkan…
What's next: if you rely on reasoning output, verify thinking fields after this change, and if you run containers on NVIDIA hardware, expect Vulkan to initialize where it previously failed. The new monitor is also worth a look for diagnosing partial GPU loads.
That's it for today, thanks for listening.
Nearby episodes from Ollama
- Mac Polish, MLX Reliability, and Compatibility Fixes
- GPU Efficiency Fix and MLX Hardening
- Single-Pass Structured Outputs and MLX Speedups
- Agent Reliability Fixes and Offline Model Transfer
- Weekly Recap - MLX Graduation, Reasoning Fixes, and Onboarding Polish
- Compatibility Fixes and Mac Efficiency
- Closing the Gaps Between Model Metadata and Client Compatibility
- Hardening Pass Across Network, Memory, and Release Paths