github-starred/open-webui
2
0
Fork 0
mirror of https://github.com/open-webui/open-webui.git synced 2026-07-11 10:34:13 -05:00
Code Issues 2.4k Packages Projects Releases 129 Wiki Activity

[GH-ISSUE #21159] issue: Model Description Image not updating for mmproj models. #106397

New Issue
Closed
opened 2026-05-18 04:42:41 -05:00 by GiteaMirror · 14 comments
GiteaMirror commented 2026-05-18 04:42:41 -05:00
Owner
Copy Link

Originally created by @JChover on GitHub (Feb 4, 2026).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/21159

Check Existing Issues

  • I have searched for any existing and/or related issues.
  • I have searched for any existing and/or related discussions.
  • I have also searched in the CLOSED issues AND CLOSED discussions and found no related items (your issue might already be addressed on the development branch!).
  • I am using the latest version of Open WebUI.

Installation Method

Docker

Open WebUI Version

v0.7.2 (latest)

Ollama Version (if applicable)

No response

Operating System

Ubuntu Server 22.04

Browser (if applicable)

Safari/Chrome

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

When selecting an image for a Model Entry in Admin Panel / Settings / Models for a specific model it should display it in the Model Selection Bar and other places in the UI.

Example of Model Working:

Image

Changes are applied and visible in the rest of the UI:

Image

Actual Behavior

When selecting an image for a Model Entry in Admin Panel / Settings / Models for a specific model is not working for some models and the changes are not visible in the Model Selection Bar and other places in the UI.

Example of Model NOT Working:

Image

Changes are NOT applied and NOT visible in the rest of the UI:

Image

Steps to Reproduce

OS: Ubuntu Server LTS 24.04
Backend: llama.cpp (latest)
OWUI: v0.7.2 (latest) - docker

Serve models through llama-server using Router Mode and .ini config file.

Example of .ini Config File Model entry that reproduces the error:

[ Devstral Small 2 - 24B Q6_K 400K kv_q8 + Vision] 
model = /home/USER/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/Devstral-Small-2-24B-Instruct-2512-Q6_K.gguf
mmproj = /home/USER/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/mmproj-BF16.gguf
ctx-size = 393216
n-gpu-layers = all
no-kv-offload = true
cache-type-k = q8_0
cache-type-v = q8_0
temp = 0.15
min-p = 0.01

This issue only appears for Vision Models that have an --mmproj flag and file in the configuration.

Logs & Screenshots

Things that I tried and didn't solve the bug:

  1. Using different browsers: Chrome / Safari.
  2. Hard Refresh Browser: Cmd + Shift + R
  3. Opening OWUI in Incognito.
  4. Cleaning cache from the browsers.
  5. Restarting OWUI Docker Container.
  6. Deleting the "cache" folder at /var/lib/docker/volumes/open-webui/_data.
  7. Accessing OWUI through IP:PORT instead of Nginx Proxy.
  8. Accessing OWUI from different devices.
  9. Using the same Model Image as one of the cases that is working as expected.
  10. Deleting the Model entry and redoing it from scratch.
  11. Exported Preset template from the models list and importing it again.

Additional Information

As you can see in the images, for some models it's working properly. After spending some time trying to identify the difference between the working ones and the others, here are my observations and conclusions:

List of Models:

Image

The Models that are not working and updating are the ones that are using an --mmproj flag and *.gguf file in my llama-server .ini config file.

Image

From the exported .json config File from the Models List in Admin Panel / Settings / Models:

Image

profile_image_url is well defined and even manually setting the value of an existing working image like the one from OpenAI seen in the above images and then Importing the .json back is not making the Model Image to work in the rest of the UI.

Originally created by @JChover on GitHub (Feb 4, 2026). Original GitHub issue: https://github.com/open-webui/open-webui/issues/21159 ### Check Existing Issues - [x] I have searched for any existing and/or related issues. - [x] I have searched for any existing and/or related discussions. - [x] I have also searched in the CLOSED issues AND CLOSED discussions and found no related items (your issue might already be addressed on the development branch!). - [x] I am using the latest version of Open WebUI. ### Installation Method Docker ### Open WebUI Version v0.7.2 (latest) ### Ollama Version (if applicable) _No response_ ### Operating System Ubuntu Server 22.04 ### Browser (if applicable) Safari/Chrome ### 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 When selecting an image for a Model Entry in Admin Panel / Settings / Models for a specific model it should display it in the Model Selection Bar and other places in the UI. **Example of Model Working:** <img width="810" height="545" alt="Image" src="https://github.com/user-attachments/assets/e6f72066-8b85-4af1-b84c-4550b45c5112" /> **Changes are applied and visible in the rest of the UI:** <img width="876" height="791" alt="Image" src="https://github.com/user-attachments/assets/2e960de6-9b5c-4a03-a9c5-72f7f348d163" /> ### Actual Behavior When selecting an image for a Model Entry in Admin Panel / Settings / Models for a specific model is not working for some models and the changes are not visible in the Model Selection Bar and other places in the UI. **Example of Model NOT Working:** <img width="774" height="521" alt="Image" src="https://github.com/user-attachments/assets/3ef68c12-6fb9-421a-8a99-bcdac3307a4a" /> **Changes are NOT applied and NOT visible in the rest of the UI:** <img width="946" height="747" alt="Image" src="https://github.com/user-attachments/assets/6ceb9f18-f54d-4ceb-afde-fe3bca01876e" /> ### Steps to Reproduce **OS:** Ubuntu Server LTS 24.04 **Backend:** llama.cpp (latest) **OWUI:** v0.7.2 (latest) - docker Serve models through llama-server using Router Mode and .ini config file. **Example of .ini Config File Model entry that reproduces the error:** ``` [ Devstral Small 2 - 24B Q6_K 400K kv_q8 + Vision] model = /home/USER/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/Devstral-Small-2-24B-Instruct-2512-Q6_K.gguf mmproj = /home/USER/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/mmproj-BF16.gguf ctx-size = 393216 n-gpu-layers = all no-kv-offload = true cache-type-k = q8_0 cache-type-v = q8_0 temp = 0.15 min-p = 0.01 ``` This issue only appears for Vision Models that have an --mmproj flag and file in the configuration. ### Logs & Screenshots **Things that I tried and didn't solve the bug:** 1. Using different browsers: Chrome / Safari. 2. Hard Refresh Browser: Cmd + Shift + R 3. Opening OWUI in Incognito. 4. Cleaning cache from the browsers. 5. Restarting OWUI Docker Container. 6. Deleting the "cache" folder at /var/lib/docker/volumes/open-webui/_data. 7. Accessing OWUI through IP:PORT instead of Nginx Proxy. 8. Accessing OWUI from different devices. 9. Using the same Model Image as one of the cases that is working as expected. 10. Deleting the Model entry and redoing it from scratch. 11. Exported Preset template from the models list and importing it again. ### Additional Information As you can see in the images, for some models it's working properly. After spending some time trying to identify the difference between the working ones and the others, here are my observations and conclusions: **List of Models:** <img width="799" height="998" alt="Image" src="https://github.com/user-attachments/assets/0561cb0c-3dc8-4f34-b266-9e75762d8618" /> The Models that are not working and updating are the ones that are using an --mmproj flag and *.gguf file in my llama-server .ini config file. <img width="445" height="641" alt="Image" src="https://github.com/user-attachments/assets/b0bc2941-e7a3-4207-84a8-36f767522d73" /> ------------ **From the exported .json config File from the Models List in Admin Panel / Settings / Models:** <img width="603" height="564" alt="Image" src="https://github.com/user-attachments/assets/0d621adc-f910-4834-8b1a-c43e5780482c" /> `profile_image_url` is well defined and even manually setting the value of an existing working image like the one from OpenAI seen in the above images and then Importing the .json back is not making the Model Image to work in the rest of the UI.
GiteaMirror added the bug label 2026-05-18 04:42:41 -05:00
GiteaMirror commented 2026-05-18 04:42:46 -05:00
Author
Owner
Copy Link

@owui-terminator[bot] commented on GitHub (Feb 4, 2026):

🔍 Similar Issues Found

I found some existing issues that might be related to this one. Please check if any of these are duplicates or contain helpful solutions:

  1. #20824 issue: Response from model does not appear until page refresh
    by NTShop • Jan 20, 2026 • bug

  2. #19610 issue: Models not appearing for non-admin users
    by westbrook-ai • Nov 30, 2025 • bug

  3. #19711 issue: Editing function for models broken
    by skleffmann • Dec 03, 2025 • bug

  4. #19788 issue: Every time I update OWUI my models get deleted
    by M4th12 • Dec 06, 2025 • bug

💡 Tips:

  • If this is a duplicate, please consider closing this issue and adding any additional details to the existing one
  • If you found a solution in any of these issues, please share it here to help others

This comment was generated automatically by a bot. Please react with a 👍 if this comment was helpful, or a 👎 if it was not.

<!-- gh-comment-id:3847301604 --> @owui-terminator[bot] commented on GitHub (Feb 4, 2026): 🔍 **Similar Issues Found** I found some existing issues that might be related to this one. Please check if any of these are duplicates or contain helpful solutions: 1. [#20824](https://github.com/open-webui/open-webui/issues/20824) **issue: Response from model does not appear until page refresh** *by NTShop • Jan 20, 2026 • `bug`* 2. [#19610](https://github.com/open-webui/open-webui/issues/19610) **issue: Models not appearing for non-admin users** *by westbrook-ai • Nov 30, 2025 • `bug`* 3. [#19711](https://github.com/open-webui/open-webui/issues/19711) **issue: Editing function for models broken** *by skleffmann • Dec 03, 2025 • `bug`* 4. [#19788](https://github.com/open-webui/open-webui/issues/19788) **issue: Every time I update OWUI my models get deleted** *by M4th12 • Dec 06, 2025 • `bug`* --- 💡 **Tips:** - If this is a duplicate, please consider closing this issue and adding any additional details to the existing one - If you found a solution in any of these issues, please share it here to help others *This comment was generated automatically by a bot.* Please react with a 👍 if this comment was helpful, or a 👎 if it was not.
GiteaMirror commented 2026-05-18 04:42:47 -05:00
Author
Owner
Copy Link

@Classic298 commented on GitHub (Feb 4, 2026):

very very likely a cache issue. does it happen still if you completely delete cache in your browser AND access open webui directly without a reverse proxy?

<!-- gh-comment-id:3847342696 --> @Classic298 commented on GitHub (Feb 4, 2026): very very likely a cache issue. does it happen still if you completely delete cache in your browser AND access open webui directly without a reverse proxy?
GiteaMirror commented 2026-05-18 04:42:47 -05:00
Author
Owner
Copy Link

@Classic298 commented on GitHub (Feb 4, 2026):

cannot reproduce anyways.

<!-- gh-comment-id:3847343421 --> @Classic298 commented on GitHub (Feb 4, 2026): cannot reproduce anyways.
GiteaMirror commented 2026-05-18 04:42:48 -05:00
Author
Owner
Copy Link

@JChover commented on GitHub (Feb 4, 2026):

very very likely a cache issue. does it happen still if you completely delete cache in your browser AND access open webui directly without a reverse proxy?

TESTED: Deleted cache from all the ends + access through IP:PORT, still having the same behavior.

I also thought it should be a cache issue but idk, very strange...

<!-- gh-comment-id:3847759020 --> @JChover commented on GitHub (Feb 4, 2026): > very very likely a cache issue. does it happen still if you completely delete cache in your browser AND access open webui directly without a reverse proxy? **TESTED:** Deleted cache from all the ends + access through IP:PORT, still having the same behavior. I also thought it should be a cache issue but idk, very strange...
GiteaMirror commented 2026-05-18 04:42:48 -05:00
Author
Owner
Copy Link

@JChover commented on GitHub (Feb 4, 2026):

UPDATE: For testing purposes I duplicated the entry for the exact same model file in the same directory without the mmproj file arg and as I was suspecting it worked without any issue.

Image

This may have something to do with the Router Mode being quite recent in llama.cpp? Idk

Here is the code for both Model entries in my config.ini file, the only difference being the mmproj arg as commented above.

# NOT WORKING CASE
[🧑🏻‍💻 Devstral Small 2 - 24B Q6_K 400K kv_q8 + Vision] 
model = /home/USER/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/Devstral-Small-2-24B-Instruct-2512-Q6_K.gguf
mmproj = /home/USER/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/mmproj-BF16.gguf
ctx-size = 393216
n-gpu-layers = all
no-kv-offload = true
cache-type-k = q8_0
cache-type-v = q8_0
temp = 0.15
min-p = 0.01

# WORKING CASE
[🧑🏻‍💻 Devstral Small 2 - 24B TEST] 
model = /home/USER/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/Devstral-Small-2-24B-Instruct-2512-Q6_K.gguf
ctx-size = 393216
n-gpu-layers = all
no-kv-offload = true
cache-type-k = q8_0
cache-type-v = q8_0
temp = 0.15
min-p = 0.01

CONCLUSION: Removing the mmproj arg from the config.ini file for the llama-server Router Mode makes the image work as expected, indicating as suspected that this may not be a cache issue @Classic298

<!-- gh-comment-id:3847909343 --> @JChover commented on GitHub (Feb 4, 2026): **UPDATE:** For testing purposes I duplicated the entry for the exact same model file in the same directory without the mmproj file arg and as I was suspecting it worked without any issue. <img width="586" height="408" alt="Image" src="https://github.com/user-attachments/assets/07f5ed71-3370-4c66-901f-325254e02f0b" /> This may have something to do with the Router Mode being quite recent in llama.cpp? Idk Here is the code for both Model entries in my config.ini file, the only difference being the mmproj arg as commented above. ``` # NOT WORKING CASE [🧑🏻‍💻 Devstral Small 2 - 24B Q6_K 400K kv_q8 + Vision] model = /home/USER/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/Devstral-Small-2-24B-Instruct-2512-Q6_K.gguf mmproj = /home/USER/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/mmproj-BF16.gguf ctx-size = 393216 n-gpu-layers = all no-kv-offload = true cache-type-k = q8_0 cache-type-v = q8_0 temp = 0.15 min-p = 0.01 # WORKING CASE [🧑🏻‍💻 Devstral Small 2 - 24B TEST] model = /home/USER/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/Devstral-Small-2-24B-Instruct-2512-Q6_K.gguf ctx-size = 393216 n-gpu-layers = all no-kv-offload = true cache-type-k = q8_0 cache-type-v = q8_0 temp = 0.15 min-p = 0.01 ``` **CONCLUSION:** Removing the `mmproj` arg from the config.ini file for the llama-server Router Mode makes the image work as expected, indicating as suspected that this may not be a cache issue @Classic298
GiteaMirror commented 2026-05-18 04:42:49 -05:00
Author
Owner
Copy Link

@silentoplayz commented on GitHub (Feb 4, 2026):

I couldn’t reproduce the issue on the latest dev commit. No reverse proxy and accessing via http://localhost:1337.

Here’s exactly what I did:

  1. Disabled the wired connection from the quick-settings panel.
Image
  1. Went to Admin → Models, picked a vision model that had no avatar, and uploaded an image.
  2. Saved, returned to the list, and hard-refreshed with Ctrl + F5.
  3. Scrolled down—the new avatar appears.
  4. Examine model selector—the new avatar appears.

Environment

  • OS: Ubuntu 24.04.3 LTS
  • Browsers: Firefox 147.0.2 (Snap) & Chrome 144.0.7559.132
  • Backend: ghcr.io/ggml-org/llama.cpp:full-vulkan Docker image
Image Image Image

Here is the entry for this particular model in my config.ini file for llama-server:

[hf.co/unsloth/InternVL3-14B-Instruct-GGUF:Q4_K_M]
m = /mnt/Samsung_990_EVO_Plus_4TB/native_gguf/hf.co_unsloth_InternVL3-14B-Instruct-GGUF_Q4_K_M.gguf
mmproj = /mnt/Samsung_990_EVO_Plus_4TB/native_gguf/hf.co_unsloth_InternVL3-14B-Instruct-GGUF_Q4_K_M.mmproj
jinja = on

Proof the model works as expected (for vision support):
Image

<!-- gh-comment-id:3847919367 --> @silentoplayz commented on GitHub (Feb 4, 2026): I couldn’t reproduce the issue on the latest `dev` commit. No reverse proxy and accessing via http://localhost:1337. Here’s exactly what I did: 1. Disabled the wired connection from the quick-settings panel. <img width="393" height="559" alt="Image" src="https://github.com/user-attachments/assets/c279caa1-0eaa-4d51-8cb8-4f8360a92137" /> 3. Went to **Admin → Models**, picked a vision model that had no avatar, and uploaded an image. 4. Saved, returned to the list, and hard-refreshed with `Ctrl + F5`. 5. Scrolled down—the new avatar appears. 6. Examine model selector—the new avatar appears. **Environment** - **OS:** Ubuntu 24.04.3 LTS - **Browsers:** Firefox 147.0.2 (Snap) & Chrome 144.0.7559.132 - **Backend:** `ghcr.io/ggml-org/llama.cpp:full-vulkan` Docker image <img width="2311" height="582" alt="Image" src="https://github.com/user-attachments/assets/b85751ad-0e06-4b64-ad40-bc847af6715d" /> <img width="2313" height="1276" alt="Image" src="https://github.com/user-attachments/assets/42aba77d-0649-402f-8b52-397bad46badd" /> <img width="498" height="566" alt="Image" src="https://github.com/user-attachments/assets/75cec774-b005-4019-99ba-c860c7283e69" /> Here is the entry for this particular model in my config.ini file for llama-server: ```ini [hf.co/unsloth/InternVL3-14B-Instruct-GGUF:Q4_K_M] m = /mnt/Samsung_990_EVO_Plus_4TB/native_gguf/hf.co_unsloth_InternVL3-14B-Instruct-GGUF_Q4_K_M.gguf mmproj = /mnt/Samsung_990_EVO_Plus_4TB/native_gguf/hf.co_unsloth_InternVL3-14B-Instruct-GGUF_Q4_K_M.mmproj jinja = on ``` Proof the model works as expected (for vision support): <img width="1347" height="1131" alt="Image" src="https://github.com/user-attachments/assets/2f765ef5-5b89-404c-abf1-8713403471d8" />
GiteaMirror commented 2026-05-18 04:42:50 -05:00
Author
Owner
Copy Link

@Classic298 commented on GitHub (Feb 4, 2026):

why does an EXTERNAL argument in an EXTERNAL software that just serves the model manipulate... whether the profile image properly updates or not? This does not make any sense 🤔

<!-- gh-comment-id:3847991038 --> @Classic298 commented on GitHub (Feb 4, 2026): why does an EXTERNAL argument in an EXTERNAL software that just serves the model manipulate... whether the profile image properly updates or not? This does not make any sense 🤔
GiteaMirror commented 2026-05-18 04:42:50 -05:00
Author
Owner
Copy Link

@JChover commented on GitHub (Feb 4, 2026):

Thanks for the feedback and help, @silentoplayz and @Classic298 .

Swapped from main branch to dev and followed your steps, unfortunately the issue persists from my end.

I'm also not sure how or why the mmproj arg in the config file affects the behavior of OWUI to show or not the model profile image but it is exactly what I'm observing from my end, also in the dev branch.

smh

<!-- gh-comment-id:3848054431 --> @JChover commented on GitHub (Feb 4, 2026): Thanks for the feedback and help, @silentoplayz and @Classic298 . Swapped from `main` branch to `dev` and followed your steps, unfortunately the issue persists from my end. I'm also not sure how or why the `mmproj` arg in the config file affects the behavior of OWUI to show or not the model profile image but it is exactly what I'm observing from my end, also in the `dev` branch. smh
GiteaMirror commented 2026-05-18 04:42:52 -05:00
Author
Owner
Copy Link

@silentoplayz commented on GitHub (Feb 4, 2026):

I couldn’t reproduce the issue on the latest dev commit. No reverse proxy and accessing via http://localhost:1337.

Here’s exactly what I did:

1. Disabled the wired connection from the quick-settings panel.
Image 
3. Went to **Admin → Models**, picked a vision model that had no avatar, and uploaded an image.

4. Saved, returned to the list, and hard-refreshed with `Ctrl + F5`.

5. Scrolled down—the new avatar appears.

6. Examine model selector—the new avatar appears.
   **Environment**


* **OS:** Ubuntu 24.04.3 LTS

* **Browsers:** Firefox 147.0.2 (Snap) & Chrome 144.0.7559.132

* **Backend:** `ghcr.io/ggml-org/llama.cpp:full-vulkan` Docker image

Image Image Image

Here is the entry for this particular model in my config.ini file for llama-server:

[hf.co/unsloth/InternVL3-14B-Instruct-GGUF:Q4_K_M]
m = /mnt/Samsung_990_EVO_Plus_4TB/native_gguf/hf.co_unsloth_InternVL3-14B-Instruct-GGUF_Q4_K_M.gguf
mmproj = /mnt/Samsung_990_EVO_Plus_4TB/native_gguf/hf.co_unsloth_InternVL3-14B-Instruct-GGUF_Q4_K_M.mmproj
jinja = on

Proof the model works as expected (for vision support): Image

It appears I didn't have to toggle off the Wired Ethernet connection to test the reported issue here. I did that though because I had this issue mixed up with https://github.com/open-webui/open-webui/issues/21140 in my mind (or at least toggling off the Wired connection part).

<!-- gh-comment-id:3848149586 --> @silentoplayz commented on GitHub (Feb 4, 2026): > I couldn’t reproduce the issue on the latest `dev` commit. No reverse proxy and accessing via http://localhost:1337. > > Here’s exactly what I did: > > 1. Disabled the wired connection from the quick-settings panel. > > > <img alt="Image" width="393" height="559" src="https://private-user-images.githubusercontent.com/50341825/545045699-c279caa1-0eaa-4d51-8cb8-4f8360a92137.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.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.FZ3FYDqVJTbLfcZfQcnMhIqSVAzm_daBTw3YExlLUto"> > > 3. Went to **Admin → Models**, picked a vision model that had no avatar, and uploaded an image. > > 4. Saved, returned to the list, and hard-refreshed with `Ctrl + F5`. > > 5. Scrolled down—the new avatar appears. > > 6. Examine model selector—the new avatar appears. > **Environment** > > > * **OS:** Ubuntu 24.04.3 LTS > > * **Browsers:** Firefox 147.0.2 (Snap) & Chrome 144.0.7559.132 > > * **Backend:** `ghcr.io/ggml-org/llama.cpp:full-vulkan` Docker image > > > <img alt="Image" width="2000" height="582" src="https://private-user-images.githubusercontent.com/50341825/545045390-b85751ad-0e06-4b64-ad40-bc847af6715d.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.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.qOwIEDjis574rPWh3xme_FY1JnRDtC1VtaxOcqmxExo"> <img alt="Image" width="2000" height="1276" src="https://private-user-images.githubusercontent.com/50341825/545045358-42aba77d-0649-402f-8b52-397bad46badd.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.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.g2bjAv37B8cx_Mx_MYg4UP4FB5JeDVxj2DrfVgbWdaE"> <img alt="Image" width="498" height="566" src="https://private-user-images.githubusercontent.com/50341825/545046262-75cec774-b005-4019-99ba-c860c7283e69.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.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.kQlLCHe2-rhD6iSU-ixuYPMpO1tYnMs9kTOtE0HmNQA"> > > Here is the entry for this particular model in my config.ini file for llama-server: > > [hf.co/unsloth/InternVL3-14B-Instruct-GGUF:Q4_K_M] > m = /mnt/Samsung_990_EVO_Plus_4TB/native_gguf/hf.co_unsloth_InternVL3-14B-Instruct-GGUF_Q4_K_M.gguf > mmproj = /mnt/Samsung_990_EVO_Plus_4TB/native_gguf/hf.co_unsloth_InternVL3-14B-Instruct-GGUF_Q4_K_M.mmproj > jinja = on > > Proof the model works as expected (for vision support): <img alt="Image" width="1347" height="1131" src="https://private-user-images.githubusercontent.com/50341825/545054901-2f765ef5-5b89-404c-abf1-8713403471d8.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.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.SbrTud9Ueto99wQt9MBYNizgOI6nUFI-d5OMhUtu0mI"> It appears I didn't have to toggle off the Wired Ethernet connection to test the reported issue here. I did that though because I had this issue mixed up with https://github.com/open-webui/open-webui/issues/21140 in my mind (or at least toggling off the Wired connection part).
GiteaMirror commented 2026-05-18 04:42:53 -05:00
Author
Owner
Copy Link

@JChover commented on GitHub (Feb 4, 2026):

Okay, finally got it after three days of banging my head. 🥲

Comparing your setup with mine @silentoplayz , I've noticed the following:

You serve the mmproj file as the exact same name of the *.gguf model file ended with .mmproj file extension whereas myself I was just using the default mmproj-BF16.gguf

I duplicated the mmproj-BF16.gguf file and renamed it to exactsamenameofmodel.mmproj and now it is working!

Image

NOT WORKING

[🧑🏻‍💻 Devstral Small 2 - 24B Q6_K 400K kv_q8 + Vision]
model = /home/USER/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/Devstral-Small-2-24B-Instruct-2512-Q6_K.gguf
mmproj = /home/USER/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/mmproj-BF16.gguf

WORKING

[🧑🏻‍💻 Devstral Small 2 - 24B Q6_K 400K kv_q8 + Vision]
model = /home/USER/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/Devstral-Small-2-24B-Instruct-2512-Q6_K.gguf
mmproj = /home/USER/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/Devstral-Small-2-24B-Instruct-2512-Q6_K.mmproj
Image
<!-- gh-comment-id:3848183203 --> @JChover commented on GitHub (Feb 4, 2026): ### **Okay, finally got it after three days of banging my head. 🥲** Comparing your setup with mine @silentoplayz , I've noticed the following: You serve the `mmproj` file as the exact same name of the ***.gguf** model file ended with **.mmproj** file extension whereas myself I was just using the default **mmproj-BF16.gguf** I duplicated the **mmproj-BF16.gguf** file and renamed it to **exactsamenameofmodel.mmproj** and now it is working! <img width="481" height="139" alt="Image" src="https://github.com/user-attachments/assets/f08f085c-1806-4d6b-b0e1-ca6782bc66c8" /> **NOT WORKING** ``` [🧑🏻‍💻 Devstral Small 2 - 24B Q6_K 400K kv_q8 + Vision] model = /home/USER/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/Devstral-Small-2-24B-Instruct-2512-Q6_K.gguf mmproj = /home/USER/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/mmproj-BF16.gguf ``` **WORKING** ``` [🧑🏻‍💻 Devstral Small 2 - 24B Q6_K 400K kv_q8 + Vision] model = /home/USER/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/Devstral-Small-2-24B-Instruct-2512-Q6_K.gguf mmproj = /home/USER/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/Devstral-Small-2-24B-Instruct-2512-Q6_K.mmproj ``` <img width="1201" height="395" alt="Image" src="https://github.com/user-attachments/assets/850f3752-a721-497d-971b-0fb9637fae76" />
GiteaMirror commented 2026-05-18 04:42:55 -05:00
Author
Owner
Copy Link

@Classic298 commented on GitHub (Feb 4, 2026):

yeah i still dont know whats going on but i read this means its solved so i close this

<!-- gh-comment-id:3848356488 --> @Classic298 commented on GitHub (Feb 4, 2026): yeah i still dont know whats going on but i read this means its solved so i close this
GiteaMirror commented 2026-05-18 04:42:56 -05:00
Author
Owner
Copy Link

@silentoplayz commented on GitHub (Feb 4, 2026):

yeah i still dont know whats going on but i read this means its solved so i close this

@Classic298 OP solved his issue organically by looking at my configuration for one of the vision models that I had set up in my config.ini file for llama-server.

@JChover Did the vision models that you had trouble with for setting the avatar image work as intended initially? In other words, could the model correctly describe what it saw in the attached image before you fixed the issue regarding the avatar image using the default mmproj-BF16.gguf?

Your claim: Not working

[🧑🏻‍💻 Devstral Small 2 - 24B Q6_K 400K kv_q8 + Vision]
model = /home/USER/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/Devstral-Small-2-24B-Instruct-2512-Q6_K.gguf
mmproj = /home/USER/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/mmproj-BF16.gguf

You and I can both agree upon this working as expected:

[🧑🏻‍💻 Devstral Small 2 - 24B Q6_K 400K kv_q8 + Vision]
model = /home/USER/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/Devstral-Small-2-24B-Instruct-2512-Q6_K.gguf
mmproj = /home/USER/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/Devstral-Small-2-24B-Instruct-2512-Q6_K.mmproj
<!-- gh-comment-id:3850044319 --> @silentoplayz commented on GitHub (Feb 4, 2026): > yeah i still dont know whats going on but i read this means its solved so i close this @Classic298 OP solved his issue organically by looking at my configuration for one of the vision models that I had set up in my config.ini file for llama-server. @JChover Did the vision models that you had trouble with for setting the avatar image work as intended initially? In other words, could the model correctly describe what it saw in the attached image *before* you fixed the issue regarding the avatar image using the default `mmproj-BF16.gguf`? ### Your claim: Not working ```ini [🧑🏻‍💻 Devstral Small 2 - 24B Q6_K 400K kv_q8 + Vision] model = /home/USER/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/Devstral-Small-2-24B-Instruct-2512-Q6_K.gguf mmproj = /home/USER/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/mmproj-BF16.gguf ``` ### You and I can both agree upon this working as expected: ```ini [🧑🏻‍💻 Devstral Small 2 - 24B Q6_K 400K kv_q8 + Vision] model = /home/USER/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/Devstral-Small-2-24B-Instruct-2512-Q6_K.gguf mmproj = /home/USER/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/Devstral-Small-2-24B-Instruct-2512-Q6_K.mmproj ```
GiteaMirror commented 2026-05-18 04:42:57 -05:00
Author
Owner
Copy Link

@JChover commented on GitHub (Feb 4, 2026):

@silentoplayz yes, I had the vision models tested and working with the mmproj-BF16.gguf files.

Model wise everything was working as expected, I will share screenshots later.

<!-- gh-comment-id:3850133063 --> @JChover commented on GitHub (Feb 4, 2026): @silentoplayz yes, I had the vision models tested and working with the mmproj-BF16.gguf files. Model wise everything was working as expected, I will share screenshots later.
GiteaMirror commented 2026-05-18 04:42:58 -05:00
Author
Owner
Copy Link

@JChover commented on GitHub (Feb 5, 2026):

Mode entry in the config.ini filel with mmproj-BF16.gguf as the file for the mmproj flag.

[🧑🏻‍💻 Devstral Small 2 - 24B mmproj] # ~30 t/s
model = /home/jchover/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/Devstral-Small-2-24B-Instruct-2512-Q6_K.gguf
mmproj = /home/jchover/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/mmproj-BF16.gguf
ctx-size = 393216
n-gpu-layers = all
no-kv-offload = true
cache-type-k = q8_0
cache-type-v = q8_0
temp = 0.15
min-p = 0.01

Model in OWUI where you can see the Image applied and the mmproj-BF16.gguf loaded as the file for the mmproj flag in the JSON:

Image

Model using the mmproj-BF16.gguf and the model Image not being updated as reported:

Image

I think it should be fairly easy to replicate from your end, just duplicate and rename the model.mmproj file to mmproj-BF16.gguf and create a new entry in your llama-server config.ini

In case you need anything else from my end, please let me know.

@silentoplayz

<!-- gh-comment-id:3852280498 --> @JChover commented on GitHub (Feb 5, 2026): Mode entry in the config.ini filel with `mmproj-BF16.gguf` as the file for the mmproj flag. ``` [🧑🏻‍💻 Devstral Small 2 - 24B mmproj] # ~30 t/s model = /home/jchover/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/Devstral-Small-2-24B-Instruct-2512-Q6_K.gguf mmproj = /home/jchover/models/code/Devstral-Small-2-24B-Instruct-2512-Q6_K/mmproj-BF16.gguf ctx-size = 393216 n-gpu-layers = all no-kv-offload = true cache-type-k = q8_0 cache-type-v = q8_0 temp = 0.15 min-p = 0.01 ``` Model in OWUI where you can see the Image applied and the `mmproj-BF16.gguf` loaded as the file for the mmproj flag in the JSON: <img width="1103" height="1296" alt="Image" src="https://github.com/user-attachments/assets/04961e2b-7b14-4920-a208-591f8d8bafab" /> Model using the `mmproj-BF16.gguf` and the model Image not being updated as reported: <img width="1071" height="399" alt="Image" src="https://github.com/user-attachments/assets/a2acfe13-5430-4bfb-b603-8dcd1f980841" /> -------- I think it should be fairly easy to replicate from your end, just duplicate and rename the `model.mmproj `file to `mmproj-BF16.gguf` and create a new entry in your llama-server config.ini In case you need anything else from my end, please let me know. @silentoplayz
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
bug
confirmed
confirmed issue
core
documentation
enhancement
good first issue
help wanted
non-core
pull-request
Mirrored from GitHub Pull Request
 python
question
testing wanted
No Label bug
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/open-webui#106397
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?