Retry stalled Paperless AI suggestions with a different local model
At a glance
A Paperless user reports clearing repeated AI-classification hangs by changing the Ollama model used for suggestions. This workflow is listed under Document Management, Local AI, Troubleshooting.
- Agents / tools
- Paperless-ngx, Ollama, Gemma 3
SourceBuilding a tool for these workflows? See weekly rates.
What they did
1) Start with an existing Paperless-ngx AI integration backed by Ollama. Identify a document whose metadata-suggestion request hangs while Ollama continues generating thousands of tokens. You need administrative access to that integration's model configuration. 2) Make gemma3:4b-it-q4_K_M available in Ollama; the command to fetch that exact model is `ollama pull gemma3:4b-it-q4_K_M`. 3) Change the model selected by the Paperless AI integration to gemma3:4b-it-q4_K_M. 4) Request suggestions again for the troublesome document and check that the loading state resolves and suggestions appear. Compare subsequent requests before deciding whether the change has fixed your installation.
How it went
Commenter SL666 reports quicker responses and no further hangs after switching models.
Worth knowing
This is one user's workaround. The thread also contains a separate llama.cpp tool-call problem, so the model change is not established as a universal fix.