Perlexica is permanently stuck on "Answer"

[remon-nashid]

Perlexica is permanently stuck on "Answer" loader. As simple as that. Out of tens of attempts it might have worked only two times.

• Using a tiny llama3.2:1b-instruct-q6_K model.
• ollama ps shows that the model gets loaded, but in 4 minutes it gets unloaded without returning an answer.
• The internal SearXNG instance works fine when queried directly on port 4000.
• Using the Perplexica docker image. Updated and rebuilt via docker-compose up --build just yesterday (Oct 18th).
• GPU is ample, and the same question when passed to OpenWebUI, using the search functionality, gets answered in seconds.
• Browser console only shows [DEBUG] opened and [DEBUG] ready.
• Docker logs of backend shows error: Unhandled Rejection at: [object Promise], reason: AxiosError: Request failed with status code 400.
• Docker logs of frontend has no errors.

Screenshots

[remon-nashid]

Well it might be model dependent. Tried gemma2:2b-instruct-q6_K and that seems to work fine :/ . If that's the case then a visible error message that explains the situation might be necessary.

[ItzCrazyKns]

Its not an issue with Perplexica, its rather an issue with the model you're using. 1B model is so small that it doesn't understands the prompt and starts outputting random things. Please use a larger model.