Ollama: Closing the Gaps Between What You Configure and What Actually Runs
Today's activity centers on fixing mismatches between user intent and runtime behavior — from context length settings to model names to thinking tokens — plus a llama.cpp version bump adding Spark X2.5 support.
Duration: PT2M23S
Episode overview
This episode is a short developer briefing from Ollama.
It explains recent repository work in plain language.
- Show: Ollama
- Published: 2026-09-07T13:01:05Z
- Audio duration: PT2M23S
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 7th, and here's your Ollama briefing.
The throughline today: several fixes address places where Ollama silently dropped or overrode what users actually asked for.
Start with context length. PR 18285 from aeltawela fixes the MLX runner so that an explicit context size — set through request options, a Modelfile, or the context length environment variable — actually takes precedence over the automatic sizing Ollama uses for VRAM management. Before this, that automatic soft value…
Similar story with PR 18281 from mann1x. When a chat template is designed to replay an assistant's prior thinking block, that thinking field was being parsed from the API but dropped before reaching the model server. The rendered path never had this problem — only the explicit message-building path did. That's now…
And PR 18278 raises the model name length cap from 80 to 96 characters to match HuggingFace's actual repo naming spec. Long, legitimately-named models pulled from HuggingFace were failing with an invalid name error even though the name was valid upstream.
On the reliability side, PR 18280 from ReinisLusis closes out a resume-on-transfer edge case: a partial file that had…
Nearby episodes from Ollama
- Weekly Recap - MLX Hardens as a First-Class Engine
- Context Handling Gets Serious
- Codex Integration Push and Schema Hardening
- Structured Output Meets Speculative Decoding
- Hardening the MLX Engine and Fixing What Feeds the Scheduler
- MLX Grows Up, Model Intent Gets Respected
- Runner Reliability and a Documentation Cleanup Wave
- Proxy Fix and a Documentation Cleanup Sweep