Ollama: MLX Memory Cuts and Tool-Call Fixes
Apple Silicon memory work headlined the week with the merged Qwen 3.8 Flash Next efficiency fix, alongside a cluster of tool-parsing and scheduler reliability patches. Launch routing, Intel graphics defaults, and backend accounting also tightened.
Duration: PT2M39S
Episode overview
This episode is a short developer briefing from Ollama.
It explains recent repository work in plain language.
- Show: Ollama
- Published: 2026-11-07T14:00:33Z
- Audio duration: PT2M39S
Transcript excerpt
This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.
Hello, this is Ollama developer news for Saturday, November seventh, twenty twenty-six.
Lead: The big signal this week is reliability under load — large models using less memory on Mac, tool calls surviving strict validation, and the scheduler no longer lying about freed video memory.
First, M L X efficiency. Merged PR 18078 reworks Qwen 3.8 Flash Next quantization after blanket N V F P 4 hurt long generations. Sensitive attention projections and the non-expert draft path stay in B F 16, other weights move to M X F P 8, and the 26-gigabyte lookup table stays memory-mapped on the host with only…
Second, tool calling correctness. PR 19569 accepts whole-number floats like 30-thousand point zero for integer arguments, fixing loops against strict tools like Open Code's bash timeout. PR 19588 names the tool a tool result answers on the Anthropic endpoint, matching the chat completions behavior. Complementary…
Third, launch and scheduling. Merged PR 19587 scopes Claude Code routing to the child process and surfaces cloud model retirement errors before launch. Open work tightens Chat G P T restore prompts and Pi sign-in metadata. On the backend, PR 19581 keeps an unstoppable runner in the…
Wha…
Nearby episodes from Ollama
- 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
- Launcher Return, API Fixes and Inference Reliability
- MLX Upgrades and OpenAI Compatibility Fixes
- Launch Hub Expands, Tokenizer APIs Proposed
- Reliability Fixes and Rockchip NPU Support