github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 00:22:43 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #3695] Splitting layers on macOS gives incorrect output #28037

New Issue
Closed
opened 2026-04-22 05:46:08 -05:00 by GiteaMirror · 5 comments
GiteaMirror commented 2026-04-22 05:46:08 -05:00
Owner
Copy Link

Originally created by @sebastiandeutsch on GitHub (Apr 17, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/3695

Originally assigned to: @jmorganca on GitHub.

What is the issue?

When running mixtral:8x22b it won't give any meaningful results:

>>> What is the capital of france? ANSWER:
	


>>> hi
	#" #$

>>>

Just jibberish

What did you expect to see?

When I run

ollama run dolphin-mixtral

the output is

>>> What is the capital of france? ANSWER:
 The capital of France is Paris.

Steps to reproduce

I'm running ollama 0.1.32
on OSX 14.1.1 (M2 Max / 96 GB RAM)

Are there any recent changes that introduced the issue?

No response

OS

macOS

Architecture

arm64

Platform

No response

Ollama version

0.1.32

GPU

Apple

GPU info

M2 MAX / 96GB RAM

CPU

Apple

Other software

No response

Originally created by @sebastiandeutsch on GitHub (Apr 17, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/3695 Originally assigned to: @jmorganca on GitHub. ### What is the issue? When running mixtral:8x22b it won't give any meaningful results: ``` >>> What is the capital of france? ANSWER: >>> hi #" #$ >>> ``` Just jibberish ### What did you expect to see? When I run ``` ollama run dolphin-mixtral ``` the output is ``` >>> What is the capital of france? ANSWER: The capital of France is Paris. ``` ### Steps to reproduce I'm running ollama 0.1.32 on OSX 14.1.1 (M2 Max / 96 GB RAM) ### Are there any recent changes that introduced the issue? _No response_ ### OS macOS ### Architecture arm64 ### Platform _No response_ ### Ollama version 0.1.32 ### GPU Apple ### GPU info M2 MAX / 96GB RAM ### CPU Apple ### Other software _No response_
GiteaMirror added the bug label 2026-04-22 05:46:08 -05:00
GiteaMirror commented 2026-04-22 05:46:09 -05:00
Author
Owner
Copy Link

@tosh commented on GitHub (Apr 17, 2024):

Getting the same result w/ 0.1.32: garbled response w/ hashtags on macOS with Wizard

maybe something about the template? 👋 @sebastiandeutsch

<!-- gh-comment-id:2061122733 --> @tosh commented on GitHub (Apr 17, 2024): Getting the same result w/ 0.1.32: garbled response w/ hashtags on macOS with Wizard maybe something about the template? 👋 @sebastiandeutsch
GiteaMirror commented 2026-04-22 05:46:10 -05:00
Author
Owner
Copy Link

@jmorganca commented on GitHub (Apr 17, 2024):

Thanks for the issue and sorry about this. I think it might be from splitting the model over cpu/gpu layers, in the meantime you try /set parameter num_gpu 0 does the answer become comprehensible? (noting that Mixtral 8x22b is a text completion model)

Also, do you see similar issues with wizardlm2:8x22b?

<!-- gh-comment-id:2061139776 --> @jmorganca commented on GitHub (Apr 17, 2024): Thanks for the issue and sorry about this. I think it might be from splitting the model over cpu/gpu layers, in the meantime you try `/set parameter num_gpu 0` does the answer become comprehensible? (noting that Mixtral 8x22b is a text completion model) Also, do you see similar issues with `wizardlm2:8x22b`?
GiteaMirror commented 2026-04-22 05:46:11 -05:00
Author
Owner
Copy Link

@sebastiandeutsch commented on GitHub (Apr 17, 2024):

/set parameter num_gpu 0 did the trick to get output ❤️

<!-- gh-comment-id:2061316832 --> @sebastiandeutsch commented on GitHub (Apr 17, 2024): `/set parameter num_gpu 0` did the trick to get output ❤️
GiteaMirror commented 2026-04-22 05:46:12 -05:00
Author
Owner
Copy Link

@antonme commented on GitHub (Apr 17, 2024):

I have the same problem with all 8x22 models, num_gpu 0 helps, but makes them very slow (since GPU does not seem to work full throttle in that case)

<!-- gh-comment-id:2061904303 --> @antonme commented on GitHub (Apr 17, 2024): I have the same problem with all 8x22 models, num_gpu 0 helps, but makes them very slow (since GPU does not seem to work full throttle in that case)
GiteaMirror commented 2026-04-22 05:46:12 -05:00
Author
Owner
Copy Link

@jmorganca commented on GitHub (Nov 17, 2024):

Hi there, this was from an issue with Ollama's memory estimation. It should largely be fixed now, but let me know if you're continuing to see the issue.

<!-- gh-comment-id:2481432015 --> @jmorganca commented on GitHub (Nov 17, 2024): Hi there, this was from an issue with Ollama's memory estimation. It should largely be fixed now, but let me know if you're continuing to see the issue.
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#28037
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?