Troubleshooting
Fixes for the most common problems.
A model won't load or is slow
- Pick a model marked Runs well for your device, or a smaller size.
- Close other heavy apps, then try again. Lumen lowers the context length automatically when memory is short.
- In Settings › Models & generation, try a shorter context length, keep Compact memory on, and use Benchmark to compare models.
A download fails
Tap Retry: downloads resume where they stopped. Check that there's enough free storage (the model's size plus about 0.5 GB). If a file is damaged, Lumen deletes it and asks you to download it again. Some models need a license accepted on Hugging Face and can't be downloaded directly.
Answers are cut off or repetitive
Raise Max response in Settings › Models & generation, or tap Continue. For repetition, a slightly higher repeat penalty or a larger model helps.
The model forgets earlier messages
Its context window is full. Use Compress conversation from the chat menu, start a new chat, or raise the context length.
A reply stopped when I left the app
iOS limits background work. Lumen keeps replying for as long as iOS allows and notifies you; tap Continue in the chat to finish. Check Settings › Notifications › Keep replying in the background.
Lumen Pro isn't unlocked
Open Settings › Lumen Pro › Restore purchases while online, signed in with the Apple Account you bought with.
Something else
Open Settings › About Lumen › Share diagnostics (it contains no chats) and send it with your message through the support form.