Ollama: Streams That Fail Silently, Fixed
Two separate fixes tackle the same class of bug — servers going quiet instead of reporting failure — in the Deepseek tool parser and the OpenAI-compatible streaming endpoint, alongside a registry authentication fix and a batch of documentation cleanup.
Duration: PT2M16S
Episode overview
This episode is a short developer briefing from Ollama.
It explains recent repository work in plain language.
- Show: Ollama
- Published: 2026-08-03T13:01:05Z
- Audio duration: PT2M16S
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 August 3rd, and today's developer briefing centers on one recurring theme: silent failures that look like success.
Two pull requests this cycle fix the same underlying problem in different parts of the stack. PR seventeen-five-two-six patches the Deepseek tool call parser, which was buffering a model's tool call until it saw a closing delimiter. If the stream ended right after a complete call but before that delimiter arrived,…
Second theme: authentication reliability. PR seventeen-five-three-two loosens an overly strict host check during registry authentication. Pulling models from the default registry could fail because the API host and the token-issuing host don't match exactly, even though that's expected behavior. This should reduce a…
Beyond that, it was a heavy documentation day. Fixes landed for missing OpenAPI fields around tool call IDs, a broken GPU docs link, a restored Modelfile parameter for GPU count, a corrected create command example, and a new tip on using OpenAI-compatible gateways with the Ollama client pattern. None of these change…
What to remember: if you're handling streaming responses or tool calls, don't assume a clean stream end means…
That…
Nearby episodes from Ollama
- Weekly Recap - Speed, Speculation, and Scheduler Reliability
- Scheduler Reliability Overhaul
- Speculative Decoding and the Cloud Model Nudge
- API Compatibility and Model Correctness Push
- Speculative Decoding Gains and a Lint Lockdown
- Cleaning Up Concurrency and Cutting Experimental Code
- Trust Your Cache, Fix Your Logs
- Parsing Bugs and Process Hardening