github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 16:11:34 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #12360] Gemma3 outputs gibberish on Apple silicon #8212

New Issue
Closed
opened 2026-04-12 20:41:12 -05:00 by GiteaMirror · 5 comments
GiteaMirror commented 2026-04-12 20:41:12 -05:00
Owner
Copy Link

Originally created by @zachrattner on GitHub (Sep 20, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/12360

What is the issue?

Last week or so, I was using gemma3:12b-it-qat and everything worked fine.

With the latest Ollama v0.12.0, it appears that Gemma 3 outputs gibberish. I'm on M4 Pro Mac mini.

Has anyone else run into this?

>> ollama run gemma3:12b-it-qat 
>>> hi
Hi, I'm3

Hi there!
Hi there! How are you do you do you doing?

<end_
end_of_turn_of_of_turn>
I
<start>
Hi,I'm doingend_do you doing.
I'm
end>I'm doend_?
<turn>

I

I'm doing_I I'm
end
<turn
I'm doing
end

Howare_to the
Hiend>end
</
I
doingm
I'I end_you end
am<start>

are
doing end
I
end
doing.
end
to doing>
^C

Relevant log output

OS

macOS

GPU

Apple

CPU

Apple

Ollama version

0.12.0

Originally created by @zachrattner on GitHub (Sep 20, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/12360 ### What is the issue? Last week or so, I was using `gemma3:12b-it-qat` and everything worked fine. With the latest Ollama v0.12.0, it appears that Gemma 3 outputs gibberish. I'm on M4 Pro Mac mini. Has anyone else run into this? ``` >> ollama run gemma3:12b-it-qat >>> hi Hi, I'm3 Hi there! Hi there! How are you do you do you doing? <end_ end_of_turn_of_of_turn> I <start> Hi,I'm doingend_do you doing. I'm end>I'm doend_? <turn> I I'm doing_I I'm end <turn I'm doing end Howare_to the Hiend>end </ I doingm I'I end_you end am<start> are doing end I end doing. end to doing> ^C ``` ### Relevant log output ```shell ``` ### OS macOS ### GPU Apple ### CPU Apple ### Ollama version 0.12.0
GiteaMirror added the bug label 2026-04-12 20:41:12 -05:00
GiteaMirror commented 2026-04-12 20:41:16 -05:00
Author
Owner
Copy Link

@zachrattner commented on GitHub (Sep 20, 2025):

I also upgraded to macOS Tahoe so I suppose that could also be causing the issue. Might not be purely ollama 0.12.0's fault

<!-- gh-comment-id:3315300953 --> @zachrattner commented on GitHub (Sep 20, 2025): I also upgraded to macOS Tahoe so I suppose that could also be causing the issue. Might not be purely ollama 0.12.0's fault
GiteaMirror commented 2026-04-12 20:41:17 -05:00
Author
Owner
Copy Link

@Haervwe commented on GitHub (Sep 20, 2025):

i have the same problem on linux, incoherent poorly formated outputs

<!-- gh-comment-id:3315311004 --> @Haervwe commented on GitHub (Sep 20, 2025): i have the same problem on linux, incoherent poorly formated outputs
GiteaMirror commented 2026-04-12 20:41:18 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Sep 20, 2025):

https://github.com/ollama/ollama/pull/12348

<!-- gh-comment-id:3315356710 --> @rick-github commented on GitHub (Sep 20, 2025): https://github.com/ollama/ollama/pull/12348
GiteaMirror commented 2026-04-12 20:41:19 -05:00
Author
Owner
Copy Link

@pdevine commented on GitHub (Sep 21, 2025):

Sorry guys. The problem happened because the official QAT weights from Google specified the wrong rope scaling factor, and we made a change which picked up that value. The fix has been merged and will be available in 0.12.1. The non-QAT weights are working as a work-around.

I'll go ahead and close the issue.

<!-- gh-comment-id:3315475197 --> @pdevine commented on GitHub (Sep 21, 2025): Sorry guys. The problem happened because the official QAT weights from Google specified the wrong rope scaling factor, and we made a change which picked up that value. The fix has been merged and will be available in `0.12.1`. The non-QAT weights are working as a work-around. I'll go ahead and close the issue.
GiteaMirror commented 2026-04-12 20:41:21 -05:00
Author
Owner
Copy Link

@zachrattner commented on GitHub (Sep 21, 2025):

Sorry guys. The problem happened because the official QAT weights from Google specified the wrong rope scaling factor, and we made a change which picked up that value. The fix has been merged and will be available in 0.12.1. The non-QAT weights are working as a work-around.

I'll go ahead and close the issue.

No sweat, thank you!!

<!-- gh-comment-id:3316070955 --> @zachrattner commented on GitHub (Sep 21, 2025): > Sorry guys. The problem happened because the official QAT weights from Google specified the wrong rope scaling factor, and we made a change which picked up that value. The fix has been merged and will be available in `0.12.1`. The non-QAT weights are working as a work-around. > > I'll go ahead and close the issue. No sweat, thank you!!
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#8212
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?