Getting lots of "WARNING - Timeout on attempt 1/4. Retrying..."

[ColaColin]

I've been trying various models, various providers and two hosts (openrouter & NovitaAI).

For all of them it seems that the LLM calls end up taking many minutes and often timeout.
I don't understand why, models like gemini flash 2 have a e2e latency of maybe 2-3 seconds on openrouter and work fine in other programs.

But here it seems somehow it massively overloads and timeouts happen. This is all with the basic function_min example.

Anyone else has this issue?

EDIT: So it seems no errors show up when using the OpenAI API. I guess something open evolve is doing is not good for other API providers somehow?

[codelion]

The timeout warnings you're seeing may be caused by OpenEvolve's default 60-second timeout being too aggressive for providers like OpenRouter, NovitaAI.

Add this to your config.yaml:

  llm:
    timeout: 300  # 5 minutes instead of default 60s try 600 if it doesn't work

Let me know if this resolves it!

[ColaColin]

Thanks for the reply.

The function_minimization example already has a timeout of 120s. I tried 240, to no avail. Most calls, but not all, still timeout.
Could be that the providers really were that slow. Maybe I got unlucky and they were mostly down when I tried.
Typically when I use openrouter in opencode or other tools they don't hang for 4 minutes before giving a response.