issue: Document embedding does not work - when OLLAMA_NEW_ENGINE=1 is enabled on Ollama #6247

New Issue

GiteaMirror · 2025-11-11T16:49:08-06:00

GiteaMirror commented

2025-11-11 16:49:08 -06:00

Originally created by @eXt73 on GitHub (Aug 28, 2025).

Check Existing Issues

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

Installation Method

Git Clone

Open WebUI Version

v0.6.26

Ollama Version (if applicable)

0.11.7

Operating System

Kubuntu Linux 24.04

Browser (if applicable)

Brave 1.81.136

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

The document should be properly embedded, then searched by Qwen3, reranked by bge-reranker-v2-m3 and the fragments prepared in this way are sent to LLM 'on the front end'

With this configuration [ollama 0.11.7], everything works correctly - although in the latest versions of Open WebUI = from .25 each change of the embedding module requires executing: sudo systemctl daemon-reload && sudo systemctl restart ollama.service - otherwise the embedding does not work.

Environment="OLLAMA_FLASH_ATTENTION=1"
Environment="OLLAMA_KV_CACHE_TYPE=q4_0"
Environment="OLLAMA_NEW_ESTIMATES=1"

Actual Behavior

In version v.0.6.25 and .26, document embedding and hybrid search/reranker stopped working when You active [on ollama]: OLLAMA_NEW_ENGINE=1. when you try to run the embedding model = Qwen3-Embedding-4B-GGUF:IQ4_NL you will get an error: 'NoneType' object is not iterable.

With this configuration [Ollma 0.11.7], embedding does not work:

Environment="OLLAMA_NEW_ENGINE=1"
Environment="OLLAMA_FLASH_ATTENTION=1"
Environment="OLLAMA_KV_CACHE_TYPE=q4_0"
Environment="OLLAMA_NEW_ESTIMATES=1"

Steps to Reproduce

I'm skipping the obvious parts:

With the Qwen3 4B embedding model and the hybrid search engine + bge-reranker-v2-m3 reranker active.
Post the document in the chat window.
Ask the main LLM model for information from the document.
The model will respond [if its system prompt is configured] that there is no information or file and an error will be displayed: 'NoneType' object is not iterable.
The response shows that the embedded file is not attached.

Logs & Screenshots

Additional Information

No response

Originally created by @eXt73 on GitHub (Aug 28, 2025). ![Image](https://github.com/user-attachments/assets/01538b06-ba73-4668-8331-7eff6612c676) ### Check Existing Issues - [x] I have searched the existing issues and discussions. - [x] I am using the latest version of Open WebUI. ### Installation Method Git Clone ### Open WebUI Version v0.6.26 ### Ollama Version (if applicable) 0.11.7 ### Operating System Kubuntu Linux 24.04 ### Browser (if applicable) Brave 1.81.136 ### 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 The document should be properly embedded, then searched by Qwen3, reranked by bge-reranker-v2-m3 and the fragments prepared in this way are sent to LLM 'on the front end' With this configuration [ollama 0.11.7], everything works correctly - although in the latest versions of Open WebUI = from .25 each change of the embedding module requires executing: sudo systemctl daemon-reload && sudo systemctl restart ollama.service - otherwise the embedding does not work. Environment="OLLAMA_FLASH_ATTENTION=1" Environment="OLLAMA_KV_CACHE_TYPE=q4_0" Environment="OLLAMA_NEW_ESTIMATES=1" ### Actual Behavior In version v.0.6.25 and .26, document embedding and hybrid search/reranker stopped working when You active [on ollama]: OLLAMA_NEW_ENGINE=1. when you try to run the embedding model = Qwen3-Embedding-4B-GGUF:IQ4_NL you will get an error: 'NoneType' object is not iterable. With this configuration [Ollma 0.11.7], embedding does not work: Environment="OLLAMA_NEW_ENGINE=1" Environment="OLLAMA_FLASH_ATTENTION=1" Environment="OLLAMA_KV_CACHE_TYPE=q4_0" Environment="OLLAMA_NEW_ESTIMATES=1" ### Steps to Reproduce I'm skipping the obvious parts: With the Qwen3 4B embedding model and the hybrid search engine + bge-reranker-v2-m3 reranker active. Post the document in the chat window. Ask the main LLM model for information from the document. The model will respond [if its system prompt is configured] that there is no information or file and an error will be displayed: 'NoneType' object is not iterable. The response shows that the embedded file is not attached. ### Logs & Screenshots ![Image](https://github.com/user-attachments/assets/9c8156aa-812a-4ab5-a12a-4ddf2ab11ece) ### Additional Information _No response_

GiteaMirror added the bug label 2025-11-11 16:49:08 -06:00

GiteaMirror closed this issue

2025-11-11 16:49:08 -06:00

GiteaMirror commented

2025-11-11 16:49:09 -06:00

@oatmealm commented on GitHub (Aug 28, 2025):

It’d be great if there would be a way to notify the user embedding failed following a nontype error. This is such a common scenario at this point.

@oatmealm commented on GitHub (Aug 28, 2025): It’d be great if there would be a way to notify the user embedding failed following a nontype error. This is such a common scenario at this point.

GiteaMirror commented

2025-11-11 16:49:09 -06:00

@rgaricano commented on GitHub (Aug 28, 2025):

Are you tried with OLLAMA_NEW_ENGINE also set?

@rgaricano commented on GitHub (Aug 28, 2025): Are you tried with OLLAMA_NEW_ENGINE also set?

GiteaMirror referenced this issue

2025-11-11 18:01:52 -06:00

[PR #6247] [MERGED] feat: folders #8643

Sign in to join this conversation.