Ollama: Weekly Recap - Sampler Fidelity, Faster Drafting, and Launcher Fixes
Ollama merged fixes to honor packaged sampler defaults, enable LFM3 with faster MLX drafting, and stabilize tool launches. Open proposals point to cache budgets, smarter scheduling, and install reliability next.
Duration: PT3M32S
Episode overview
This episode is a short developer briefing from Ollama.
It explains recent repository work in plain language.
- Show: Ollama
- Published: 2026-11-02T09:07:21Z
- Audio duration: PT3M32S
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 your Ollama recap for October 26th through November 2nd, 2026.
30 pull request activity items, 15 additional commits this week.
This week Ollama focused on correctness and compatibility at the model layer, with meaningful gains in drafting speed and more reliable launches with outside developer tools.
First, model behavior on OpenAI-compatible endpoints. PR 19432 stopped overriding sampler defaults when clients omitted temperature, top P, or penalties. That means models like L F M 3, Qwen and DeepSeek variants now keep their packaged temperature and related settings instead of falling back to generic values,…
Second, speed and new-model support. PR 19149 added D-Flash 2, a block draft approach with local dynamic convolutions and a candidate selector to propose more tokens per round while keeping output exact. PR 19392 followed up by sorting long top K rows in chunks, cutting the cost of large-vocabulary sampling and log…
Third, launcher and app workflows. PR 19310 fixed Open Code version-two launches by using standalone mode so the selected model is honored, with more tolerant version detection and config parsing. PR 19410 raised the output limit for Claude Code to…
Nearby episodes from Ollama
- MLX Upgrades and OpenAI Compatibility Fixes
- Launch Hub Expands, Tokenizer APIs Proposed
- Reliability Fixes and Rockchip NPU Support
- Inference Correctness and Installer Reliability
- Sampler Defaults Fix and Memory Work
- MLX Memory Safety and Silent Correctness Fixes
- MLX Performance Push and Compatibility Fixes
- OpenAI Fidelity and Runner Stability