Ollama: Cleaning Up Memory Planning and the Launch Experience
Ollama consolidated backend memory planning into one place and simplified the launch and integration surface, while several smaller fixes addressed chat template correctness, streaming behavior, and crash safety in model conversion. The changes are mostly maintainability and correctness work, not new features.
Duration: PT2M45S
Episode overview
This episode is a short developer briefing from Ollama.
It explains recent repository work in plain language.
- Show: Ollama
- Published: 2026-07-14T13:00:49Z
- Audio duration: PT2M45S
Transcript excerpt
This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.
Good day. It's July 14th, and here's what moved in Ollama.
The headline change is a consolidation of backend load planning. Daniel Hiltgen's PR 17165 centralizes memory policy that had been scattered across scheduler preflight, request setup, and runner startup. That fragmentation was papering over a real bug: Ollama was adding extra padding for image projectors even though…
Second theme: the launch and integration experience got a real cleanup. Parth Sareen renamed the Codex App integration to ChatGPT in PR 17161, reflecting that OpenAI now ships Codex inside the ChatGPT desktop app, with old names kept as aliases so nothing breaks for existing users. Alongside that, PR 17159…
A few standalone fixes are worth flagging. PR 17173 corrects Gemma 4 chat template rendering, aligning tool-calling and turn-closure behavior with Google's updated reference templates. PR 17166 changes the OpenAI-compatible streaming API so the assistant role is only emitted once per response instead of on every…
On infrastructure, PR 17160 restores missing Metal NAX kernels after a macOS deployment target mismatch, and PR 14796 makes MLX model load timeouts configurable.
What to remember: if you touch memory…
Nearby episodes from Ollama
- Fixing What Defaults Got Wrong
- Weekly Recap - Model Correctness and Runtime Hardening
- Tightening Up the Serving Layer
- Thinking Model Output Is Getting a Real Fix
- Stability Sweep Across Cloud, GPU, and Agent Tools
- Qwen3.5 Gets Untangled, and Small Bugs With Big Blast Radius
- Model Behavior Bugs and Blob Security Hardening
- Launch Gets Smarter, Model Support Gets Wider