Timeout issues continue...

[rambopreacher]

Ever since I got it running after receiving the unit from my Kickstarter campaign, its performance has been problematic.

My setup:
Latest marketplace version of:
Open WebUI
Qwen3.5 35B A3B UD-Q4 (Ollama)
BGE-M3 (Ollama) (recently replaced Nomic, but the issue is the same regardless of document handler)
Command line at the console says: Ubantu 24.04.03 LTS Olares TTY1, Olares Login:

When typing a request in the WebUI, if I pause for any significant amount of time and come back, it times out, and I get an Oh Snap error in the browser.

If I am able to hit send/enter before it times out, it will also time out at times. If it is a short, easy thing, it may return a result.

I am trying to build a Knowledge library. If I select more than one file to upload, it times out. If I select a single file, it might work, but it usually times out, and I have to start again.

I am the only one connected to the system; no other IDs have even been created. I am on my own LAN. If I restart the Olares One, it works for a little while, but the longer I leave it running, the more I get timeout errors.

I am new to all of this, so I assume I am at fault with the settings and configuration, not the hardware. It's set to performance mode and app exclusive to Qwen.

Suggestions? FIxes?

[Power-One-2025]

Hi @rambopreacher,

Thank you for bringing this to our attention and sorry for the frustrating experience you've had with the timeouts and errors!

Our team has looked into this, and we suspect the issue might be related to how GPU resources are currently allocated in your setup. You mentioned that Qwen is set to "app exclusive" GPU mode, which gives it full access to the GPU. Because of this, BGE-M3 cannot use the GPU and falls back to the CPU. Running BGE-M3 on CPU is slower, especially for larger files, which is likely causing the timeouts and "Oh Snap" errors. Additionally, potential GPU resource competition among multiple apps on v1.12.5 might be contributing to this situation.

Recommendation & Solution

We recommend following these three steps to update your environment and optimize the configuration:

1. Upgrade Olares OS to v1.12.6

Version 1.12.6 introduces a brand new Model Console, per-app CPU/GPU selection, and much better overall GPU resource management for AI apps. Please check out the Release Notes to see the highlights and improvements.

Upgrading takes only a few simple steps. You can follow the quick guide here.

2. Upgrade Open WebUI

Once your OS is updated, please go to the Market and upgrade Open WebUI to the latest version (Chart version 1.0.38, App version 0.11.0).

3. Configure Models & Follow the Tutorial

After upgrading to 1.12.6, we suggest trying the following model setup:

• Qwen3.6-27B (llama.cpp): Install this model app (Chart version 1.0.9, App version unsloth/Qwen3.6-27B-GGUF:Q4_K_M) from the Market.
• EmbeddingGemma: Install this model app (Chart version 1.0.11, App version 0.1.0) from the Market. When installing, select CPU as the hardware accelerator (CPU is sufficient for text embeddings and helps avoid competing with Qwen for GPU memory).
• For a step-by-step guide on putting this together, please refer to our Open WebUI Tutorial, particularly the section on Knowledge Base Setup.

Please try upgrading to 1.12.6 and adjusting these settings, and let us know if this resolves the issue for you. Thank you again for your patience and support!

[rambopreacher]

1. I upgraded to 1.12.6. No issues.
2. I uninstalled the document handler and Qwen that I was running.
3. Updated to the latest Open WebUI.
4. Restarted (to avoid variables in troubleshooting)
5. Installed EmbeddingGemma with no issues (that I can tell so far)
6. After it finished downloading (not something I had to do previously), I was able to copy the link and model and add them to Open WebUI document configuration (no other changes in that configuration.
7. Installed the QWEN version suggested (though I am unsure whey I am not getting the Qwen3.6-35B-A3B version.
8. Waited for the rather large downloads...
9. Then configured that in the AI connections settings in Open WebUI, and it fails when I click the refresh/verify connection. It says Ollama: Network Problem.

EDIT: I thought I was following the instructions I had before, but the AI model goes in OpenAI, not Ollama. Once I made that change, it works. Currently, it is performing as expected (though I am not sure how to build a knowledge base library) - HOWEVER, it always works well after a reboot and on first use.

[Power-One-2025]

Thanks for taking the time to work through the steps and share the detailed update. We are glad to hear that the model connection is now working and the system is performing as expected.

You're correct: the recommended Qwen3.6 model uses an OpenAI-compatible connection, so it should be added under OpenAI, not Ollama. With the recommended configuration, Open WebUI should provide a smooth and reliable experience for everyday chat and knowledge base use. For guidance on creating the knowledge base, you can refer to the Knowledge Base Setup guide.

We have not received similar reports indicating that this is a common issue. If you encounter the same issue again, feel free to let us know, and we'll be happy to investigate it further with you.

Thanks again for your feedback!