github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-11-10 15:02:13 -06:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #2782] Why the Gemma performs that bad with some simple questions? #48191

New Issue
Closed
opened 2026-04-28 07:06:38 -05:00 by GiteaMirror · 4 comments
GiteaMirror commented 2026-04-28 07:06:38 -05:00
Owner
Copy Link
Copy Source

Originally created by @brightzheng100 on GitHub (Feb 27, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/2782

Just tried Gemma model but not sure why it performed that bad.

ollama pull gemma:2b
ollama run gemma:2b

So is it the model issue or the model hosted here is different?

For example - this should be a common-sense question but it has no idea:

>>> Do you think whether human can fly? Tell me why with step-by-step reasoning.
The context does not provide any information about whether humans can fly, so I cannot generate an answer based on the context.

While this one is interesting, the answer is wrong while the reasoning process is correct:

>>> Michael is 10 years old. Rob is 2 years older than Amy, while Amy is 5 years younger than Michael. How old is Rob?
Rob is 12 years old.

We know that Amy is 5 years younger than Michael, so she is 5 - 10 = 5 years old. Since Rob is 2 years older than Amy, he is 5 + 2 = 7 years old.

ollama

Originally created by @brightzheng100 on GitHub (Feb 27, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/2782 Just tried `Gemma` model but not sure why it performed that bad. ```sh ollama pull gemma:2b ollama run gemma:2b ``` **So is it the model issue or the model hosted here is different?** For example - this should be a common-sense question but it has no idea: ``` >>> Do you think whether human can fly? Tell me why with step-by-step reasoning. The context does not provide any information about whether humans can fly, so I cannot generate an answer based on the context. ``` While this one is interesting, the answer is wrong while the reasoning process is correct: ``` >>> Michael is 10 years old. Rob is 2 years older than Amy, while Amy is 5 years younger than Michael. How old is Rob? Rob is 12 years old. We know that Amy is 5 years younger than Michael, so she is 5 - 10 = 5 years old. Since Rob is 2 years older than Amy, he is 5 + 2 = 7 years old. ``` ![ollama](https://github.com/ollama/ollama/assets/1422425/8dd16e62-1caf-4356-b23f-d1820c883124)
GiteaMirror commented 2026-04-28 07:06:43 -05:00
Author
Owner
Copy Link
Copy Source

@furquanh commented on GitHub (Feb 27, 2024):

This a 2b parameter model. I don't see a use case for chatting with it. Maybe you could perform very basic logical reasoning task with it.

<!-- gh-comment-id:1967505739 --> @furquanh commented on GitHub (Feb 27, 2024): This a 2b parameter model. I don't see a use case for chatting with it. Maybe you could perform very basic logical reasoning task with it.
GiteaMirror commented 2026-04-28 07:06:46 -05:00
Author
Owner
Copy Link
Copy Source

@elmoBG8 commented on GitHub (Feb 27, 2024):

Just tried Gemma model but not sure why it performed that bad.

ollama pull gemma:2b
ollama run gemma:2b

So is it the model issue or the model hosted here is different?

For example - this should be a common-sense question but it has no idea:

>>> Do you think whether human can fly? Tell me why with step-by-step reasoning.
The context does not provide any information about whether humans can fly, so I cannot generate an answer based on the context.

While this one is interesting, the answer is wrong while the reasoning process is correct:

>>> Michael is 10 years old. Rob is 2 years older than Amy, while Amy is 5 years younger than Michael. How old is Rob?
Rob is 12 years old.

We know that Amy is 5 years younger than Michael, so she is 5 - 10 = 5 years old. Since Rob is 2 years older than Amy, he is 5 + 2 = 7 years old.

ollama ollama

You're not the only one experimenting very bad answers from Gemma... Watch the end of this post: https://didyouknowbg8.wordpress.com/2024/02/27/testing-some-large-language-model-locally-with-ollama/
It's astonishing how bad it performs (both models: 2b and 7b) in comparison with the other open source models...

<!-- gh-comment-id:1967624300 --> @elmoBG8 commented on GitHub (Feb 27, 2024): > Just tried `Gemma` model but not sure why it performed that bad. > > ```shell > ollama pull gemma:2b > ollama run gemma:2b > ``` > > **So is it the model issue or the model hosted here is different?** > > For example - this should be a common-sense question but it has no idea: > > ``` > >>> Do you think whether human can fly? Tell me why with step-by-step reasoning. > The context does not provide any information about whether humans can fly, so I cannot generate an answer based on the context. > ``` > > While this one is interesting, the answer is wrong while the reasoning process is correct: > > ``` > >>> Michael is 10 years old. Rob is 2 years older than Amy, while Amy is 5 years younger than Michael. How old is Rob? > Rob is 12 years old. > > We know that Amy is 5 years younger than Michael, so she is 5 - 10 = 5 years old. Since Rob is 2 years older than Amy, he is 5 + 2 = 7 years old. > ``` > > ![ollama](https://private-user-images.githubusercontent.com/1422425/308213304-8dd16e62-1caf-4356-b23f-d1820c883124.gif?jwt=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.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.adMScMHFrbJM86ub8pVkVQNwtSbELK4BxItGtNY6UD4) [ ![ollama](https://private-user-images.githubusercontent.com/1422425/308213304-8dd16e62-1caf-4356-b23f-d1820c883124.gif?jwt=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.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.adMScMHFrbJM86ub8pVkVQNwtSbELK4BxItGtNY6UD4) ](https://private-user-images.githubusercontent.com/1422425/308213304-8dd16e62-1caf-4356-b23f-d1820c883124.gif?jwt=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.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.adMScMHFrbJM86ub8pVkVQNwtSbELK4BxItGtNY6UD4) [ ](https://private-user-images.githubusercontent.com/1422425/308213304-8dd16e62-1caf-4356-b23f-d1820c883124.gif?jwt=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.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.adMScMHFrbJM86ub8pVkVQNwtSbELK4BxItGtNY6UD4) You're not the only one experimenting very bad answers from Gemma... Watch the end of this post: https://didyouknowbg8.wordpress.com/2024/02/27/testing-some-large-language-model-locally-with-ollama/ It's astonishing how bad it performs (both models: 2b and 7b) in comparison with the other open source models...
GiteaMirror commented 2026-04-28 07:06:48 -05:00
Author
Owner
Copy Link
Copy Source

@gl-prout commented on GitHub (Mar 1, 2024):

I have tried using gemma 2b and 7b (the ones that you pull using ollama pull), using ollama with docker.

And every time I try to call the model (using API, using Open-WebUI), it crashes the ollama docker, making it restart

Edit: nevermind, had to pull the latest ollama image and now, it responds

<!-- gh-comment-id:1972632442 --> @gl-prout commented on GitHub (Mar 1, 2024): I have tried using gemma 2b and 7b (the ones that you pull using `ollama pull`), using ollama with docker. And every time I try to call the model (using API, using Open-WebUI), it crashes the ollama docker, making it restart Edit: nevermind, had to pull the latest ollama image and now, it responds
GiteaMirror commented 2026-04-28 07:06:49 -05:00
Author
Owner
Copy Link
Copy Source

@jmorganca commented on GitHub (May 10, 2024):

2b can have differing results given it's small size although I'm impressed by it. You may want to try using 7B – let me know if that helps!

<!-- gh-comment-id:2103676952 --> @jmorganca commented on GitHub (May 10, 2024): 2b can have differing results given it's small size although I'm impressed by it. You may want to try using 7B – let me know if that helps!
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 labels
Milestone
No items
No Milestone
Projects
Clear projects
No projects
Assignees
Clear assignees
GiteaMirror ninjasurge (Kaehlen Rog)
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#48191