Quick diagnosis
“Model not found,” a download stalled halfway, and a completed file that will not load are different failures. Keep the exact error and download source, then follow the matching check below.
First three checks
- Copy the exact model name, tag, download address, and error message.
- Confirm that the model exists at its official registry and that your account can access it if access is restricted.
- Check free space on the drive where the app stores models, then retry the same download without deleting your model library.
Check the model identifier
For Ollama, use the tag shown on its model-library page. For example, ollama pull qwen3:4b downloads a specific small model; ollama list should list it after completion. A Hugging Face repository name, a GGUF filename, and an Ollama library tag are not interchangeable identifiers.
In LM Studio, reopen the selected model's download entry and confirm the file and quantization you chose. If a repository requires accepting terms or signing in, resolve access at the repository first. Do not paste access tokens into a public chat or an error report.
Match the error to the fix
| Symptom | Check | Next action |
|---|---|---|
| Not found or invalid manifest | Exact registry, spelling, and tag | Copy an existing tag from that registry |
| Unauthorized, forbidden, or gated model | Account access and model terms | Complete the registry's access process before retrying |
| No space left or write failure | Free space and write permissions on the model-storage drive | Free unrelated storage or change the app's documented storage location |
| Timeout or connection reset | Registry accessibility, VPN, proxy, and network connection | Retry on an allowed stable connection; check proxy configuration |
| Repeated integrity or digest failure | Whether a retry always fails on the same file | Preserve the error and use the app's controls for that incomplete download |
| Download completes but load fails | Installed file and runtime/model compatibility | Follow model-loading troubleshooting |
Allow storage for the final file and download working space. The system drive can fill even when a separate documents drive has room. RAM availability is not the first check for a download that cannot be saved to disk.
Retry without losing working models
In Ollama, rerun the same ollama pull command. In a desktop app, resume or retry the selected download through its download manager. If an incomplete file must be cleared, identify that specific entry first; do not delete shared blob folders or the whole library.
For an HTTPS proxy, use the runtime's documented proxy setting and certificate configuration. Do not disable certificate checks to make a download work. A network administrator may need to allow the model registry on a managed connection.
What success looks like
The download completes, the exact model appears in the installed list, and a short local prompt returns an answer. Test that small model before retrying a much larger download. A successful browser download alone does not establish that the model format is supported by your app.
When to stop
Stop when the file's publisher cannot be identified, access is denied, or repeated integrity failures remain unexplained. Choose a model available through the app's supported download path in install Ollama or install LM Studio.