[GH-ISSUE #2306] Docker image with Ollama installed very slow #28362

Closed
opened 2026-04-25 03:01:32 -05:00 by GiteaMirror · 2 comments
Owner

Originally created by @PietFourie on GitHub (May 16, 2024).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/2306

Hi
My system has a NVDIA 1070 8 GB GPU card running linux mint

I used the latest Docker images as installed on the 15th May

  1. I have installed the Docker image with Ollama installed and it runs very slow.
  2. I then installed the Docker image with Ollama and Open WebUI seperately and it runs the fastest.
  3. I then ran the same prompt directly on the Ollama terminal and it had similar speed.

The prompt was :
Write me a python code for the game centipede
Model LLama 3:lastest

  1. The code generated by Ollama directly and with a seperate Ollama/Open WebUI docker image was the same and used the pygame library. I did not test the code but is used clasees, and more advanced code
  2. The code generated by the combined Olllama/Webui docker image create a very small code using no libraries and I doubt it would have run.

The error is that the combined Ollama/OpenWebUI docker miage creates carbage and is a lot slower.

Error with ModelFilles
Also I coulld not import the modelfiles from the website. It just does nothing. Also when you import the model it looks for ".json" files but the downloaded modelfile have a ".txt" extension. I would presume the error is significant. All model files are on the recommended home page of the website and should be compatible.

Originally created by @PietFourie on GitHub (May 16, 2024). Original GitHub issue: https://github.com/open-webui/open-webui/issues/2306 Hi My system has a NVDIA 1070 8 GB GPU card running linux mint I used the latest Docker images as installed on the 15th May 1) I have installed the Docker image with Ollama installed and it runs very slow. 2) I then installed the Docker image with Ollama and Open WebUI seperately and it runs the fastest. 3) I then ran the same prompt directly on the Ollama terminal and it had similar speed. The prompt was : Write me a python code for the game centipede Model LLama 3:lastest 1) The code generated by Ollama directly and with a seperate Ollama/Open WebUI docker image was the same and used the pygame library. I did not test the code but is used clasees, and more advanced code 2) The code generated by the combined Olllama/Webui docker image create a very small code using no libraries and I doubt it would have run. The error is that the combined Ollama/OpenWebUI docker miage creates carbage and is a lot slower. Error with ModelFilles Also I coulld not import the modelfiles from the website. It just does nothing. Also when you import the model it looks for ".json" files but the downloaded modelfile have a ".txt" extension. I would presume the error is significant. All model files are on the recommended home page of the website and should be compatible.
Author
Owner
<!-- gh-comment-id:2114482299 --> @PietFourie commented on GitHub (May 16, 2024): Here is the code generated [Ollama direct Inference Lama3.txt](https://github.com/open-webui/open-webui/files/15331931/Ollama.direct.Inference.Lama3.txt) [Open WEBUI and Ollama Seperated Docker images.txt](https://github.com/open-webui/open-webui/files/15331932/Open.WEBUI.and.Ollama.Seperated.Docker.images.txt) [OpenWebUI with Ollama InStalled Inference Lama3.txt](https://github.com/open-webui/open-webui/files/15331933/OpenWebUI.with.Ollama.InStalled.Inference.Lama3.txt)
Author
Owner

@PietFourie commented on GitHub (May 16, 2024):

GPU memory used in the combined version was lower than in the separate Ollama docker images. It also used CPU cores.

<!-- gh-comment-id:2114485835 --> @PietFourie commented on GitHub (May 16, 2024): GPU memory used in the combined version was lower than in the separate Ollama docker images. It also used CPU cores.
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/open-webui#28362