Ollama: Weekly Recap - Hardening the Runtime, One Edge Case at a Time
This week's activity centered on closing correctness and reliability gaps across the model-serving path — from GPU scheduling and memory estimation to chat template safety and the agent runtime — with fifty pull requests and two dozen supporting commits.
Duration: PT3M30S
Episode overview
This episode is a short developer briefing from Ollama.
It explains recent repository work in plain language.
- Show: Ollama
- Published: 2026-07-20T09:09:03Z
- Audio duration: PT3M30S
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 the Ollama weekly recap for July 13th through July 20th.
This week: fifty pull request activity items, twenty-four additional commits. The dominant pattern is a hardening pass — a lot of bug reports from real deployments turning into fixes across GPU scheduling, memory estimation, and chat template handling.
Let's start with GPU and scheduling reliability. Multiple pull requests, 17,191 and 17,194, tackled a Vulkan multi-GPU issue where a smaller card got zero layers assigned, forcing CPU fallback despite plenty of VRAM elsewhere. The fix orders multi-GPU groups by free memory first. Alongside that, PR 17,201 improves…
Second theme: chat template and tokenization safety. Two independent pull requests, 17,262 and 17,263, addressed the same underlying class of bug — literal special-token strings like the "end of think" tag showing up in user messages and corrupting prompt structure for models like GLM and DeepSeek. Related fixes…
Third theme: the agent runtime. Parth Sareen shipped a cluster of related changes — a new skills system in PR 17,203, a broader semantics and code-quality cleanup in PR 17,212, and unlimited tool rounds by default for cloud models in PR…
Fourt…
Nearby episodes from Ollama
- Thinking Models Leak Control Tokens
- Closing the Gaps Between API Layers
- Cleaning Up the Agent Loop and Widening Model Support
- Agent Package Gets a Major Consolidation
- Agent Hardening and a Path Traversal Fix
- Resource Leaks and GPU Placement Get a Clean-Up Pass
- Cleaning Up Memory Planning and the Launch Experience
- Fixing What Defaults Got Wrong