issue: The search tool encountered an error: Torch not compiled with CUDA enabled #5729

New Issue

GiteaMirror · 2025-11-11T16:31:49-06:00

GiteaMirror commented

2025-11-11 16:31:49 -06:00

Originally created by @awesomez on GitHub (Jul 9, 2025).

Check Existing Issues

I have searched the existing issues and discussions.
I am using the latest version of Open WebUI.

Installation Method

Other

Open WebUI Version

v0.6.15

Ollama Version (if applicable)

No response

Operating System

Windows 10

Browser (if applicable)

Firefox

Confirmation

I have read and followed all instructions in README.md.
I am using the latest version of both Open WebUI and Ollama.
I have included the browser console logs.
I have included the Docker container logs.
I have provided every relevant configuration, setting, and environment variable used in my setup.
I have clearly listed every relevant configuration, custom setting, environment variable, and command-line option that influences my setup (such as Docker Compose overrides, .env values, browser settings, authentication configurations, etc).
I have documented step-by-step reproduction instructions that are precise, sequential, and leave nothing to interpretation. My steps:
Start with the initial platform/version/OS and dependencies used,
Specify exact install/launch/configure commands,
List URLs visited, user input (incl. example values/emails/passwords if needed),
Describe all options and toggles enabled or changed,
Include any files or environmental changes,
Identify the expected and actual result at each stage,
Ensure any reasonably skilled user can follow and hit the same issue.

Expected Behavior

Use LLM Web Search tool (from OWUI repository) with "jan-nano-128k" model (with full context) via LM Studio (as the v1 model provider) and SearXNG (via official docker image) to perform a web search for a topic, then an analysis of the found web content should be performed as per the script/tool execution.

Actual Behavior

"The search tool encountered an error: Torch not compiled with CUDA enabled"

Then the model proceeded to think, outputting its response, but no web search had been performed.

Steps to Reproduce

SearXNG works fine with the standard Web Search tool, and it also worked for the FIRST time the web search was performed using "LLM Web Search" tool. (If it helps, the Qwen3 embeddings 0.6 model were in the embeddings folder the LLM Web Search tool requested in its Valves, as I did not know that LLM Web Search tool downloaded its own embeddings).

The second time I performed a web search (and all subsequent times after closing/restarting OpenWebUI, adding or deleting tools, etc), I recieved the error above.

Logs & Screenshots

See this post:

https://github.com/open-webui/open-webui/discussions/8170

This describes the exact scenario I was experiencing.

Additional Information

https://github.com/open-webui/open-webui/discussions/8170

I've provided further information in that post as to particulars of my setup, but these are all bog-standard and the only difference was that I was using SearXNG for the standard Web Search tool (successfully) in OWUI and decided to use a faster one that uses SearXNG (LLM Web Search) - both were using SearXNG via the standard docker image running in Docker Desktop on Windows 10.

Also, sometimes even the standard Web Search tool seems to hang up for a while. Lastly, I also get lots of uvicorn 303 responses in the terminal, assuming that is not a red-herring. These seem to be related to tools.

Originally created by @awesomez on GitHub (Jul 9, 2025). ### Check Existing Issues - [x] I have searched the existing issues and discussions. - [x] I am using the latest version of Open WebUI. ### Installation Method Other ### Open WebUI Version v0.6.15 ### Ollama Version (if applicable) _No response_ ### Operating System Windows 10 ### Browser (if applicable) Firefox ### Confirmation - [x] I have read and followed all instructions in `README.md`. - [x] I am using the latest version of **both** Open WebUI and Ollama. - [x] I have included the browser console logs. - [x] I have included the Docker container logs. - [x] I have **provided every relevant configuration, setting, and environment variable used in my setup.** - [x] I have clearly **listed every relevant configuration, custom setting, environment variable, and command-line option that influences my setup** (such as Docker Compose overrides, .env values, browser settings, authentication configurations, etc). - [x] I have documented **step-by-step reproduction instructions that are precise, sequential, and leave nothing to interpretation**. My steps: - Start with the initial platform/version/OS and dependencies used, - Specify exact install/launch/configure commands, - List URLs visited, user input (incl. example values/emails/passwords if needed), - Describe all options and toggles enabled or changed, - Include any files or environmental changes, - Identify the expected and actual result at each stage, - Ensure any reasonably skilled user can follow and hit the same issue. ### Expected Behavior Use LLM Web Search tool (from OWUI repository) with "jan-nano-128k" model (with full context) via LM Studio (as the v1 model provider) and SearXNG (via official docker image) to perform a web search for a topic, then an analysis of the found web content should be performed as per the script/tool execution. ### Actual Behavior "The search tool encountered an error: Torch not compiled with CUDA enabled" Then the model proceeded to think, outputting its response, but no web search had been performed. ### Steps to Reproduce SearXNG works fine with the standard Web Search tool, and it also worked for the FIRST time the web search was performed using "LLM Web Search" tool. (If it helps, the Qwen3 embeddings 0.6 model were in the embeddings folder the LLM Web Search tool requested in its Valves, as I did not know that LLM Web Search tool downloaded its own embeddings). The second time I performed a web search (and all subsequent times after closing/restarting OpenWebUI, adding or deleting tools, etc), I recieved the error above. ### Logs & Screenshots See this post: https://github.com/open-webui/open-webui/discussions/8170 This describes the exact scenario I was experiencing. ### Additional Information https://github.com/open-webui/open-webui/discussions/8170 I've provided further information in that post as to particulars of my setup, but these are all bog-standard and the only difference was that I was using SearXNG for the standard Web Search tool (successfully) in OWUI and decided to use a faster one that uses SearXNG (LLM Web Search) - both were using SearXNG via the standard docker image running in Docker Desktop on Windows 10. Also, sometimes even the standard Web Search tool seems to hang up for a while. Lastly, I also get lots of uvicorn 303 responses in the terminal, assuming that is not a red-herring. These seem to be related to tools. <img width="1635" height="1269" alt="Image" src="https://github.com/user-attachments/assets/5498076e-d45a-4c90-a7bb-a75d6165f69a" />

GiteaMirror added the bug label 2025-11-11 16:31:49 -06:00

GiteaMirror closed this issue

2025-11-11 16:31:49 -06:00

GiteaMirror commented

2025-11-11 16:31:49 -06:00

@awesomez commented on GitHub (Jul 9, 2025):

UPDATE:

It appears that when I turn on "Temporary Chat" and explicitly enable CPU-only, the embedding models load and the web search is performed using the "LLM Web Search" tool. See below:

This is just very slow - I'd prefer NOT to use the CPU-only mode. :(

Otherwise, I get this:

@awesomez commented on GitHub (Jul 9, 2025): UPDATE: It appears that when I turn on "Temporary Chat" and explicitly enable CPU-only, the embedding models load and the web search is performed using the "LLM Web Search" tool. See below: <img width="529" height="1311" alt="Image" src="https://github.com/user-attachments/assets/168dba3a-c6f6-4c54-85ae-f4baadc790d6" /> <img width="1172" height="1307" alt="Image" src="https://github.com/user-attachments/assets/dc735c03-3504-4000-a08d-6ba93da09719" /> This is just very slow - I'd prefer NOT to use the CPU-only mode. :( Otherwise, I get this: <img width="1210" height="1270" alt="Image" src="https://github.com/user-attachments/assets/62c6e605-82b2-4a5f-ac05-983d28e030cb" />

GiteaMirror commented

2025-11-11 16:31:50 -06:00

@tjbck commented on GitHub (Jul 9, 2025):

Custom tools are outside of the scope.

@tjbck commented on GitHub (Jul 9, 2025): Custom tools are outside of the scope.