github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-07-17 13:47:08 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #7395] Why is the performance of x/llama3.2-vision for describing images poor? #51213

New Issue
Closed
opened 2026-04-28 18:55:55 -05:00 by GiteaMirror · 7 comments
GiteaMirror commented 2026-04-28 18:55:55 -05:00
Owner
Copy Link

Originally created by @hosea7456 on GitHub (Oct 28, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/7395

What is the issue?

I am experiencing x/llama3.2-vision(11b) using ollama-0.4.0-rc5, but the performance is poor compared to the official version. What causes this result?
x/llama3.2-vision:
image
meta-llama:
image
image:
cars

OS

Linux

GPU

Nvidia

CPU

Intel

Ollama version

0.4.0-rc5

Originally created by @hosea7456 on GitHub (Oct 28, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/7395 ### What is the issue? I am experiencing x/llama3.2-vision(11b) using ollama-0.4.0-rc5, but the performance is poor compared to the [official version](https://deepinfra.com/meta-llama/Llama-3.2-11B-Vision-Instruct). What causes this result? **x/llama3.2-vision:** ![image](https://github.com/user-attachments/assets/c02cf985-59c0-4cd3-bccc-9ac7a3a19efa) **meta-llama:** ![image](https://github.com/user-attachments/assets/e6c0b2dc-f8aa-465d-8bea-1a0b98704daf) **image:** ![cars](https://github.com/user-attachments/assets/d8ac625e-86c7-47f4-a45d-9a56dc5eeb73) ### OS Linux ### GPU Nvidia ### CPU Intel ### Ollama version 0.4.0-rc5
GiteaMirror added the bug label 2026-04-28 18:55:55 -05:00
GiteaMirror commented 2026-04-28 18:55:56 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Oct 28, 2024):

My guess would be that the official version is unquantized. x/llama3.2-vision is experimental, when it's promoted to the main ollama library it should perform better.

<!-- gh-comment-id:2441251412 --> @rick-github commented on GitHub (Oct 28, 2024): My guess would be that the official version is unquantized. x/llama3.2-vision is experimental, when it's promoted to the main ollama library it should perform better.
GiteaMirror commented 2026-04-28 18:55:57 -05:00
Author
Owner
Copy Link

@hosea7456 commented on GitHub (Oct 30, 2024):

My guess would be that the official version is unquantized. x/llama3.2-vision is experimental, when it's promoted to the main ollama library it should perform better.

That should be the reason. Many thanks.

<!-- gh-comment-id:2445674203 --> @hosea7456 commented on GitHub (Oct 30, 2024): > My guess would be that the official version is unquantized. x/llama3.2-vision is experimental, when it's promoted to the main ollama library it should perform better. That should be the reason. Many thanks.
GiteaMirror commented 2026-04-28 18:55:57 -05:00
Author
Owner
Copy Link

@step21 commented on GitHub (Nov 7, 2024):

On a similar issue - I just tried two images from my download folder - for one that is a pencil drawing without anything particularly salacious, it steadfastly refused, always saying sth like "I cannot engage in this topic". When I asked it to elaborate, it also just repeated a similar message. Did anybody else experience sth like that?

Update: it worked when starting a new chat. But seemingly and follow up questions result in refusals to answer.

<!-- gh-comment-id:2462084118 --> @step21 commented on GitHub (Nov 7, 2024): On a similar issue - I just tried two images from my download folder - for one that is a pencil drawing without anything particularly salacious, it steadfastly refused, always saying sth like "I cannot engage in this topic". When I asked it to elaborate, it also just repeated a similar message. Did anybody else experience sth like that? Update: it worked when starting a new chat. But seemingly and follow up questions result in refusals to answer.
GiteaMirror commented 2026-04-28 18:55:58 -05:00
Author
Owner
Copy Link

@pdevine commented on GitHub (Nov 13, 2024):

I think we can close this out since x/llama3.2-vision was just for the RC versions. The model isn't perfect, unfortunately, and still does hallucinate occasionally.

<!-- gh-comment-id:2474926376 --> @pdevine commented on GitHub (Nov 13, 2024): I think we can close this out since `x/llama3.2-vision` was just for the RC versions. The model isn't perfect, unfortunately, and still does hallucinate occasionally.
GiteaMirror commented 2026-04-28 18:55:59 -05:00
Author
Owner
Copy Link

@step21 commented on GitHub (Nov 13, 2024):

@pdevine is there a new version now?

<!-- gh-comment-id:2474935239 --> @step21 commented on GitHub (Nov 13, 2024): @pdevine is there a new version now?
GiteaMirror commented 2026-04-28 18:55:59 -05:00
Author
Owner
Copy Link

@pdevine commented on GitHub (Nov 13, 2024):

@step21 Yes, see the blog post. The models are now called llama3.2-vision (without the x/ on the front). The weights did change, so you'll unfortunately need to pull them again.

<!-- gh-comment-id:2474980723 --> @pdevine commented on GitHub (Nov 13, 2024): @step21 Yes, see the [blog post](https://ollama.com/blog/llama3.2-vision). The models are now called `llama3.2-vision` (without the `x/` on the front). The weights did change, so you'll unfortunately need to pull them again.
GiteaMirror commented 2026-04-28 18:56:00 -05:00
Author
Owner
Copy Link

@step21 commented on GitHub (Nov 13, 2024):

ok, thx. though I just noticed I only used the new one without x/ anyway. Sorry, didn't realize this was for a previous version. It still does the behaviour I described anyway, but still ok to close this thread.

<!-- gh-comment-id:2474983448 --> @step21 commented on GitHub (Nov 13, 2024): ok, thx. though I just noticed I only used the new one without x/ anyway. Sorry, didn't realize this was for a previous version. It still does the behaviour I described anyway, but still ok to close this thread.
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
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/ollama#51213
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?