Error messages

model runner has unexpectedly stopped, this may be due to resource limitations or an internal error, check ollama server logs for details

model runner has unexpectedly stopped, this may be due to resource limitations or an internal error, check ollama server logs for details

The model loads, you type a prompt, and partway in (sometimes before a single word appears) Ollama prints:

Error: model runner has unexpectedly stopped, this may be due to resource limitations or an internal error, check ollama server logs for details

That's the full line as reported in issue #12472. The wording has stayed the same since then, and 61 issues in the Ollama repo quote it as of September 26, 2026.

Where it comes from

Ollama doesn't run the model itself. It starts a helper, llama-server, and talks to it over a local network connection. We read that code in Ollama 0.34.4. When Ollama sends your prompt to the helper and the connection fails outright, it returns this message. So the helper was fine a moment earlier (the model loaded) and then went away or stopped responding.

If the helper had printed an error line before dying, Ollama tacks it onto the end after a colon, and that tail is the part worth searching for. When there's nothing after details, the helper went down without saying why.

A crash during loading gets a different message, llama-server process has terminated. This one means you got further than that.

What the log usually shows

The message says "resource limitations or an internal error", and the threads split roughly that way.

Memory is the first thing to rule out. A model that just fits can load and then run short once the prompt and its working memory grow, and a long prompt makes that worse.

Graphics drivers come up a lot too. In #12472, on an AMD Radeon 8060S under Linux, the log had HW Exception by GPU node-1 ... reason :GPU Hang. An Ollama collaborator read it as a driver problem: "I don't think it's an ollama issue, looks like a kernel/driver issue." He later posted the kernel settings that had stopped the hangs on his own machine, which is about as far as that thread got.

Then there are brand-new models. When the qwen3.5 family arrived, reports piled into #14419 within a day, and the collaborator's first answer was "0.17.1 for qwen3.5 support". An older Ollama can pull a model it can't run properly yet.

What to do

Find the server log (the troubleshooting page lists it per system: ~/.ollama/logs/server.log on a Mac, journalctl -u ollama on Linux with systemd, %LOCALAPPDATA%\Ollama\server.log on Windows). Scroll to the time of the failure and read upward. A line with CUDA error, ROCm error, GGML_ASSERT or a GPU fault tells you which of the three stories you're in.

Then work through the cheap fixes. Update Ollama, especially if the model came out recently. Try a smaller size tag of the same model, or a shorter context, and close other programs using the graphics card. If only one model does it and others run fine, search the issues for that model's name with "unexpectedly stopped", since a bad model release tends to collect a dozen reports fast.

If you file an issue, attach the log. "Server logs will aid in debugging" is the collaborator's standard first reply, and this message on its own doesn't give anyone enough to go on.