Ollama: Tool Call Reliability and Runner Efficiency
Tool calling edge cases dominate activity, with three PRs fixing lost or misreported calls, while a merged runner fix stops repeated model reloads. An Apple Silicon speculative decoding optimization is also in review.
Duration: PT2M17S
Episode overview
This episode is a short developer briefing from Ollama.
It explains recent repository work in plain language.
- Show: Ollama
- Published: 2026-11-19T14:02:01Z
- Audio duration: PT2M17S
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 19th, 2026.
The lead pattern today is fragile tool calling. Three separate changes tackle cases where valid tool intent never reaches your client as a structured call.
First, the Qwen coder parser. Two open proposals address opposite ends of the same problem. PR 19764 fixes calls missing the opening tool call tag, which happens when the model writes a sentence first and then emits a complete function block with a closer. Today the parser ignores it and returns text. PR 19776…
Second, runner efficiency. Merged PR 19748, fixing issue 18950, stops a reload on every request when clients alternate context shift settings. The server now launches with shift when the model supports it and only reloads if a request needs shift the runner lacks. Requests asking for no shift are enforced by capping…
In housekeeping, merged PR 19746 stabilizes a flaky speech slow-reader test by blocking the writer explicitly, and three proposals add Coati and Meta-Capp to community integrations.
What's next: watch whether the tool call fixes merge together without parser conflicts, and whether the shift change alters long-context behavior in production.
Nearby episodes from Ollama
- Silent Path Fixes and Runner Cleanup
- Inkling on Mac Gains and Stream Reliability Fixes
- Adaptive Memory on Macs, Tighter Server Security
- Runner Reuse, Decision Models, and Thinking UI Fixes
- MLX Concurrency and Memory Tuning
- GPU Load Fixes and Clearer Errors
- Weekly Recap - Parallel Inference, Tool Fidelity, and Launcher Polish
- Qwen Tool-Call Reliability Fix