github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 16:40:08 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #13495] Token generation speed #34658

New Issue
Closed
opened 2026-04-22 18:24:15 -05:00 by GiteaMirror · 7 comments
GiteaMirror commented 2026-04-22 18:24:15 -05:00
Owner
Copy Link

Originally created by @Eb7CAPJi on GitHub (Dec 16, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/13495

Could you explain why the token generation speed changes from version to version for the same model and the same question? I have a whole collection of these distributions and it's quite noticeable. Specifically, this version has, to put it mildly, a low token generation speed compared to the previous one.
ollama version is 0.13.4

Originally created by @Eb7CAPJi on GitHub (Dec 16, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/13495 Could you explain why the token generation speed changes from version to version for the same model and the same question? I have a whole collection of these distributions and it's quite noticeable. Specifically, this version has, to put it mildly, a low token generation speed compared to the previous one. ollama version is 0.13.4
GiteaMirror added the feature requestneeds more info labels 2026-04-22 18:24:15 -05:00
GiteaMirror commented 2026-04-22 18:24:18 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Dec 16, 2025):

Data?

<!-- gh-comment-id:3659811065 --> @rick-github commented on GitHub (Dec 16, 2025): Data?
GiteaMirror commented 2026-04-22 18:24:18 -05:00
Author
Owner
Copy Link

@pdevine commented on GitHub (Dec 16, 2025):

There is a bench command in cmd/bench which you can run which will give token generation speeds. We do have integration tests to check performance before we release.

Is there a particular model that you're seeing a slow down with, and can you describe the system you're using?

<!-- gh-comment-id:3661702919 --> @pdevine commented on GitHub (Dec 16, 2025): There is a `bench` command in `cmd/bench` which you can run which will give token generation speeds. We do have integration tests to check performance before we release. Is there a particular model that you're seeing a slow down with, and can you describe the system you're using?
GiteaMirror commented 2026-04-22 18:24:19 -05:00
Author
Owner
Copy Link

@Eb7CAPJi commented on GitHub (Dec 17, 2025):

I installed a new version of the Ollama program alongside the old one.
When I run a simple prompt, “Describe yourself,” the newer model version
generates tokens slowly, whereas the older version is very fast—so there
isn’t enough time to read the output before it finishes.

вт, 16 дек. 2025 г. в 20:46, Patrick Devine @.***>:

pdevine left a comment (ollama/ollama#13495)
https://github.com/ollama/ollama/issues/13495#issuecomment-3661702919

There is a bench command in cmd/bench which you can run which will give
token generation speeds. We do have integration tests to check performance
before we release.

Is there a particular model that you're seeing a slow down with, and can
you describe the system you're using?

—
Reply to this email directly, view it on GitHub
https://github.com/ollama/ollama/issues/13495#issuecomment-3661702919,
or unsubscribe
https://github.com/notifications/unsubscribe-auth/ADHCCD4G56GPAUTYBKDV5KD4CBANPAVCNFSM6AAAAACPFSLKFWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTMNRRG4YDEOJRHE
.
You are receiving this because you authored the thread.Message ID:
@.***>

<!-- gh-comment-id:3665991478 --> @Eb7CAPJi commented on GitHub (Dec 17, 2025): I installed a new version of the Ollama program alongside the old one. When I run a simple prompt, “Describe yourself,” the newer model version generates tokens slowly, whereas the older version is very fast—so there isn’t enough time to read the output before it finishes. вт, 16 дек. 2025 г. в 20:46, Patrick Devine ***@***.***>: > *pdevine* left a comment (ollama/ollama#13495) > <https://github.com/ollama/ollama/issues/13495#issuecomment-3661702919> > > There is a bench command in cmd/bench which you can run which will give > token generation speeds. We do have integration tests to check performance > before we release. > > Is there a particular model that you're seeing a slow down with, and can > you describe the system you're using? > > — > Reply to this email directly, view it on GitHub > <https://github.com/ollama/ollama/issues/13495#issuecomment-3661702919>, > or unsubscribe > <https://github.com/notifications/unsubscribe-auth/ADHCCD4G56GPAUTYBKDV5KD4CBANPAVCNFSM6AAAAACPFSLKFWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTMNRRG4YDEOJRHE> > . > You are receiving this because you authored the thread.Message ID: > ***@***.***> >
GiteaMirror commented 2026-04-22 18:24:19 -05:00
Author
Owner
Copy Link

@Eb7CAPJi commented on GitHub (Dec 17, 2025):

another caveat is that token generation has been moved to the CPU instead
of the gpu. This was evident from the PROCESSOR recycling schedule.

ср, 17 дек. 2025 г., 18:53 Вадим @.***>:

I installed a new version of the Ollama program alongside the old one.
When I run a simple prompt, “Describe yourself,” the newer model version
generates tokens slowly, whereas the older version is very fast—so there
isn’t enough time to read the output before it finishes.

вт, 16 дек. 2025 г. в 20:46, Patrick Devine @.***>:

pdevine left a comment (ollama/ollama#13495)
https://github.com/ollama/ollama/issues/13495#issuecomment-3661702919

There is a bench command in cmd/bench which you can run which will give
token generation speeds. We do have integration tests to check performance
before we release.

Is there a particular model that you're seeing a slow down with, and can
you describe the system you're using?

—
Reply to this email directly, view it on GitHub
https://github.com/ollama/ollama/issues/13495#issuecomment-3661702919,
or unsubscribe
https://github.com/notifications/unsubscribe-auth/ADHCCD4G56GPAUTYBKDV5KD4CBANPAVCNFSM6AAAAACPFSLKFWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTMNRRG4YDEOJRHE
.
You are receiving this because you authored the thread.Message ID:
@.***>

<!-- gh-comment-id:3666431432 --> @Eb7CAPJi commented on GitHub (Dec 17, 2025): another caveat is that token generation has been moved to the CPU instead of the gpu. This was evident from the PROCESSOR recycling schedule. ср, 17 дек. 2025 г., 18:53 Вадим ***@***.***>: > I installed a new version of the Ollama program alongside the old one. > When I run a simple prompt, “Describe yourself,” the newer model version > generates tokens slowly, whereas the older version is very fast—so there > isn’t enough time to read the output before it finishes. > > вт, 16 дек. 2025 г. в 20:46, Patrick Devine ***@***.***>: > >> *pdevine* left a comment (ollama/ollama#13495) >> <https://github.com/ollama/ollama/issues/13495#issuecomment-3661702919> >> >> There is a bench command in cmd/bench which you can run which will give >> token generation speeds. We do have integration tests to check performance >> before we release. >> >> Is there a particular model that you're seeing a slow down with, and can >> you describe the system you're using? >> >> — >> Reply to this email directly, view it on GitHub >> <https://github.com/ollama/ollama/issues/13495#issuecomment-3661702919>, >> or unsubscribe >> <https://github.com/notifications/unsubscribe-auth/ADHCCD4G56GPAUTYBKDV5KD4CBANPAVCNFSM6AAAAACPFSLKFWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTMNRRG4YDEOJRHE> >> . >> You are receiving this because you authored the thread.Message ID: >> ***@***.***> >> >
GiteaMirror commented 2026-04-22 18:24:19 -05:00
Author
Owner
Copy Link

@pdevine commented on GitHub (Dec 17, 2025):

Can you post the logs? What model are you trying to use? What kind of GPU? What is the output of ollama ps?

<!-- gh-comment-id:3667224501 --> @pdevine commented on GitHub (Dec 17, 2025): Can you post the logs? What model are you trying to use? What kind of GPU? What is the output of `ollama ps`?
GiteaMirror commented 2026-04-22 18:24:20 -05:00
Author
Owner
Copy Link

@Eb7CAPJi commented on GitHub (Dec 18, 2025):

The result is independent of the model type. The calculations have shifted
to burden the CPU instead of the GPU.

This is very noticeable on the 34B models. It's the maximum I can afford.

чт, 18 дек. 2025 г. в 00:21, Patrick Devine @.***>:

pdevine left a comment (ollama/ollama#13495)
https://github.com/ollama/ollama/issues/13495#issuecomment-3667224501

Can you post the logs? What model are you trying to use? What kind of GPU?
What is the output of ollama ps?

—
Reply to this email directly, view it on GitHub
https://github.com/ollama/ollama/issues/13495#issuecomment-3667224501,
or unsubscribe
https://github.com/notifications/unsubscribe-auth/ADHCCD735ZXB6WXGJRRKZWD4CHCNXAVCNFSM6AAAAACPFSLKFWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTMNRXGIZDINJQGE
.
You are receiving this because you authored the thread.Message ID:
@.***>

<!-- gh-comment-id:3669027167 --> @Eb7CAPJi commented on GitHub (Dec 18, 2025): The result is independent of the model type. The calculations have shifted to burden the CPU instead of the GPU. This is very noticeable on the 34B models. It's the maximum I can afford. чт, 18 дек. 2025 г. в 00:21, Patrick Devine ***@***.***>: > *pdevine* left a comment (ollama/ollama#13495) > <https://github.com/ollama/ollama/issues/13495#issuecomment-3667224501> > > Can you post the logs? What model are you trying to use? What kind of GPU? > What is the output of ollama ps? > > — > Reply to this email directly, view it on GitHub > <https://github.com/ollama/ollama/issues/13495#issuecomment-3667224501>, > or unsubscribe > <https://github.com/notifications/unsubscribe-auth/ADHCCD735ZXB6WXGJRRKZWD4CHCNXAVCNFSM6AAAAACPFSLKFWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTMNRXGIZDINJQGE> > . > You are receiving this because you authored the thread.Message ID: > ***@***.***> >
GiteaMirror commented 2026-04-22 18:24:20 -05:00
Author
Owner
Copy Link

@Eb7CAPJi commented on GitHub (Dec 18, 2025):

Honestly, I have neither the time nor the particular desire to dive into
the code and figure out the root causes of the problems.
All I can say is that Ollama is an amazing tool—just wrap it in tests to
minimize the bugs.

чт, 18 дек. 2025 г. в 11:28, Вадим @.***>:

The result is independent of the model type. The calculations have shifted
to burden the CPU instead of the GPU.

This is very noticeable on the 34B models. It's the maximum I can afford.

чт, 18 дек. 2025 г. в 00:21, Patrick Devine @.***>:

pdevine left a comment (ollama/ollama#13495)
https://github.com/ollama/ollama/issues/13495#issuecomment-3667224501

Can you post the logs? What model are you trying to use? What kind of
GPU? What is the output of ollama ps?

—
Reply to this email directly, view it on GitHub
https://github.com/ollama/ollama/issues/13495#issuecomment-3667224501,
or unsubscribe
https://github.com/notifications/unsubscribe-auth/ADHCCD735ZXB6WXGJRRKZWD4CHCNXAVCNFSM6AAAAACPFSLKFWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTMNRXGIZDINJQGE
.
You are receiving this because you authored the thread.Message ID:
@.***>

<!-- gh-comment-id:3669069790 --> @Eb7CAPJi commented on GitHub (Dec 18, 2025): Honestly, I have neither the time nor the particular desire to dive into the code and figure out the root causes of the problems. All I can say is that Ollama is an amazing tool—just wrap it in tests to minimize the bugs. чт, 18 дек. 2025 г. в 11:28, Вадим ***@***.***>: > The result is independent of the model type. The calculations have shifted > to burden the CPU instead of the GPU. > > This is very noticeable on the 34B models. It's the maximum I can afford. > > чт, 18 дек. 2025 г. в 00:21, Patrick Devine ***@***.***>: > >> *pdevine* left a comment (ollama/ollama#13495) >> <https://github.com/ollama/ollama/issues/13495#issuecomment-3667224501> >> >> Can you post the logs? What model are you trying to use? What kind of >> GPU? What is the output of ollama ps? >> >> — >> Reply to this email directly, view it on GitHub >> <https://github.com/ollama/ollama/issues/13495#issuecomment-3667224501>, >> or unsubscribe >> <https://github.com/notifications/unsubscribe-auth/ADHCCD735ZXB6WXGJRRKZWD4CHCNXAVCNFSM6AAAAACPFSLKFWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTMNRXGIZDINJQGE> >> . >> You are receiving this because you authored the thread.Message ID: >> ***@***.***> >> >
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 feature request needs more info
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#34658
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?