Ollama: Tool-Call Reliability Fixes
Four parser fixes target lost or malformed tool calls across model families, while cleanup changes reclaim disk and improve model unloading. The merged benchmark update switches to code-like prompts for more realistic performance testing.
Duration: PT2M7S
Episode overview
This episode is a short developer briefing from Ollama.
It explains recent repository work in plain language.
- Show: Ollama
- Published: 2026-09-27T13:00:44Z
- Audio duration: PT2M7S
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 September 27th, 2026 — you're listening to your Ollama developer briefing.
The big signal is tool-call reliability. Four separate fixes address the same failure mode across different models: a valid tool call never reaching the client.
For Gemma 4, in PR 18664, stray characters after an otherwise complete call caused the whole thing to be rejected — now the parser recovers using the JSON boundary. A companion fix for DeepSeek 3, in PR 18682, holds back partial tool tags split across streamed chunks instead of leaking them as assistant text.…
The second thread is resource cleanup. PR 18678 runs blob garbage collection right after model delete, instead of waiting for a server restart, fixing disk left behind when the last reference is gone. That pairs with command-line support for stopping multiple models at once or everything loaded, plus corrections to…
Finally, benchmarking gets more realistic. Merged PR 17480 swaps synthetic word lists for packed HumanEval Python prompts on the chat path, with nonce prefixes to avoid cache hits. In parallel, proposals to build the Go tree without C tooling aim to unblock Windows builds, alongside a macOS menu-bar hide option and…
W…
Nearby episodes from Ollama
- Desktop Fixes and API Correctness
- 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
- Thinking Separation Fix and GPU Visibility
- Weekly Recap - MLX Graduation, Reasoning Fixes, and Onboarding Polish
- Compatibility Fixes and Mac Efficiency