Quick diagnosis
If Ollama reports “model not found,” stops while loading, or closes before answering, first separate a missing download from a stopped service and a memory failure. The error text decides which check comes next.
First three checks
- Run
ollama listand copy the exact installed model name, including its tag. - Run
ollama run MODEL_NAMEwith that name and a short prompt, without a browser interface or document attached. - If the model starts loading but fails, reduce context to 4096 tokens and try a smaller downloaded model before changing drivers.
Replace MODEL_NAME with the name shown in your own list; it is a placeholder, not a model to download.
Check the service and the download
Run ollama --version, then ollama list in Terminal or PowerShell. A version and a model table mean the command is installed and the server responds. An empty table means this server has no downloaded models. Follow install Ollama to download a small model first.
If the message says it cannot connect to Ollama, start the installed Ollama app or service. If you intentionally use the standalone command-line setup, run ollama serve in another terminal. An “address already in use” error means another process already owns the port; identify it instead of starting repeated servers.
If a model name is missing, copy its exact tag from the library and use ollama pull MODEL_NAME. A completed download followed by a matching entry in ollama list is the expected result. Download errors belong in model download failures.
Test one short conversation
Run the installed model directly. At its chat prompt, enter /set parameter num_ctx 4096, then ask “Reply with one short greeting.” Context controls the space for the prompt, conversation, and answer; a short target leaves more memory for the model.
If it answers here but fails in Open WebUI, keep the model unchanged and use the connection checks. If only document chat fails, use the PDF diagnostic.
Read the failure before retrying
| Error or symptom | Next check | Successful result |
|---|---|---|
| Model or manifest not found | Confirm spelling and tag in the library | The exact model appears in the local list |
| Cannot connect to Ollama | Start the app or service on the intended computer | The local list command returns normally |
| Allocation failure, out of memory, or heavy swapping | Close another loaded model, shorten context, then try a smaller file | A short chat runs while the desktop stays responsive |
| Unsupported architecture or runtime error | Check the installed Ollama version and model's required version | A supported model loads on the current stable runtime |
| Process exits without a useful message | Read the server log for the matching time | The log identifies a specific download, allocation, or backend error |
On Windows, open %LOCALAPPDATA%\Ollama in File Explorer and read server.log. On macOS use ~/.ollama/logs/server.log. On a Linux systemd installation use journalctl -u ollama --no-pager -n 100. Remove prompts, paths, and personal information before sharing logs.
When to stop
If a small model also fails, keep the exact error and runtime version and diagnose that failure first. Repeatedly downloading larger models will not repair a stopped service or an unsupported backend.
For memory errors, enter the file size and context in the memory calculator, then follow the out-of-memory guide.