Contents

Ollama FAQ: Troubleshoot Common Errors and Issues

How to view logs

You can find the cause of most issues in the logs:

System	Log location
Windows	`%LOCALAPPDATA%\Ollama\server.log` (paste this into the File Explorer address bar to open it)
macOS	`~/.ollama/logs/server.log`
Linux	`journalctl -u ollama --no-pager -n 200`
Docker	`docker logs ollama`

For more detailed information, set the OLLAMA_DEBUG=1 environment variable and restart.

Connection issues

could not connect to ollama app, is it running?

The Ollama service is not running.

Windows / macOS: Open the Ollama app.
Linux: sudo systemctl start ollama.
Test: curl http://localhost:11434 should return Ollama is running.

Can't connect from another computer or Docker container

By default, Ollama listens only on 127.0.0.1. Set OLLAMA_HOST to 0.0.0.0:11434, restart, and make sure your firewall allows port 11434. See Environment variables and configuration.

Error: listen tcp 127.0.0.1:11434: bind: address already in use

An Ollama service is already running (usually the app or a systemd service), so you don't need to run ollama serve again.

Download issues

Downloads are slow or interrupted

Run ollama pull again. It resumes the download where it stopped.

Not enough disk space

ollama list          # List model sizes
ollama rm model-you-dont-need

Or move the models to another drive and set OLLAMA_MODELS.

Memory and performance

It runs slowly

Check the PROCESSOR field with ollama ps. If it doesn't show 100% GPU, the model may be too large or the GPU may not be in use. For full troubleshooting, see GPU and performance tuning.

model requires more system memory than is available

You don't have enough memory. Switch to a smaller model or a quantized version, shorten the context length, or close other programs.

It takes a long time to start answering every time

The model is being reloaded each time. Increase OLLAMA_KEEP_ALIVE (for example, to 30m), or check whether switching between multiple models is causing them to compete for memory and get unloaded.

Answer quality

Answers are in Simplified Chinese

Set the system prompt to: “Always respond in Traditional Chinese as used in Taiwan,” or use a Modelfile to create a model that defaults to Traditional Chinese. You can also switch to a model with better Chinese-language capabilities.

The answer is cut off partway through

Check whether num_predict is set.
Input that is too long and exceeds the context length can also cause problems. Increase num_ctx.

Long documents are summarized only from the second half

The context length is too short, so the beginning was cut off. Increase the context length, split the document into sections, or use RAG.

Answers are nonsensical or repetitive

Smaller models have limited capabilities. Switch to a larger model.
Adjust the parameters: lower temperature and raise repeat_penalty (for example, to 1.1).
If you imported the GGUF model yourself, its prompt template (TEMPLATE) may be incorrect.

The model can't use tools or understand images

Check whether the model supports the feature: the Capabilities field in ollama show model-name lists items such as tools, vision, and thinking.