Ollama: OpenAI Fidelity and Runner Stability
Merged fixes preserve reasoning, refusals, and images when history is resent through the Responses API, with follow-ups tightening embedding and content validation. Runner work addresses a Mac Metal crash, build deadlocks, and a proposal for per-request speculative decoding metrics.
Duration: PT2M29S
Episode overview
This episode is a short developer briefing from Ollama.
It explains recent repository work in plain language.
- Show: Ollama
- Published: 2026-10-29T13:00:59Z
- Audio duration: PT2M29S
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 October 29th, 2026, and this is your Ollama developer briefing.
The throughline today is stopping silent data loss in the OpenAI compatibility layer, backed by a set of runner and build stability fixes.
First, context preservation. Merged PR 19192 keeps readable reasoning, prior refusals, and inline images when apps resend history through the Responses endpoint, and replaces missing file-only images with an explicit notice instead of skipping them. Three open follow-ups extend that same discipline: PR 19384…
Second, runtime reliability and visibility. The team bumped the core inference engine in PR 19360 and added a temporary workaround in PR 19372 for a Metal compiler crash building the unary library on older Mac versions. Continuous integration was unblocked in PR 19373 by pinning the container build tool after…
Finally, small correctness fixes with outsized developer impact: trimming whitespace around comma-separated allow-list entries in PR 19365 to avoid failed origin matches, ignoring null sampler values from generation configs in PR 19382 to avoid forcing greedy decoding, and omitting Modelfile values that cannot…
What's next: review the proposed metrics…
Nearby episodes from Ollama
- Weekly Recap - Sampler Fidelity, Faster Drafting, and Launcher Fixes
- Sampler Defaults Fix and Memory Work
- MLX Memory Safety and Silent Correctness Fixes
- MLX Performance Push and Compatibility Fixes
- LFM3 Speech, Video and Reliability Fixes
- Scheduler Memory Fixes and Faster Drafting
- Mac Prefill Gains and Scheduler Fixes
- Weekly Recap - Phonon Voice Goes GA and MLX Speed Gains