github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 08:30:05 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #1855] I miss option to specify num of gpu layers as model parameter #47571

New Issue
Closed
opened 2026-04-28 04:15:32 -05:00 by GiteaMirror · 5 comments
GiteaMirror commented 2026-04-28 04:15:32 -05:00
Owner
Copy Link

Originally created by @JoseConseco on GitHub (Jan 8, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/1855

The 2 most used parameters for gguf models are IMO: temp, and number of gpu layers for mode to use.
But number of gpu layers is 'baked' into ollama model template file.
This means we have to create new model, with new num of gpu layer - jut to change it.

yes I understand number of gpu layers is not something that can be changed after model was loaded. But still, creating new modelfile just to change gpu layer offloading parameter is overkill imo.

Originally created by @JoseConseco on GitHub (Jan 8, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/1855 The 2 most used parameters for gguf models are IMO: temp, and number of gpu layers for mode to use. But number of gpu layers is 'baked' into ollama model template file. This means we have to create new model, with new num of gpu layer - jut to change it. yes I understand number of gpu layers is not something that can be changed after model was loaded. But still, creating new modelfile just to change gpu layer offloading parameter is overkill imo.
GiteaMirror commented 2026-04-28 04:15:35 -05:00
Author
Owner
Copy Link

@BruceMacD commented on GitHub (Jan 8, 2024):

Thanks for the feedback @JoseConseco, as of the last few versions of Ollama you can actually specify this in the interactive mode.

ollama run llama2
>>> /set parameter num_gpu 12
Set parameter 'num_gpu' to '12'

>>>

Does that help your use-case?

<!-- gh-comment-id:1881719430 --> @BruceMacD commented on GitHub (Jan 8, 2024): Thanks for the feedback @JoseConseco, as of the last few versions of Ollama you can actually specify this in the interactive mode. ``` ollama run llama2 >>> /set parameter num_gpu 12 Set parameter 'num_gpu' to '12' >>> ``` Does that help your use-case?
GiteaMirror commented 2026-04-28 04:15:36 -05:00
Author
Owner
Copy Link

@JoseConseco commented on GitHub (Jan 8, 2024):

awesome. It was one of the most annoying thing about ollama (having to create custom model, to change gpu layers. )
While /set parameter num_gpu 12 works - model is reloaded after next prompt, after setting gpu-layes.
Will have to test if this helps, if model is to big to load into vram. I suppose in that case ollama will just error out, and I wont be able to /set parameter num_gpu 12 right? In that case user will have to create new modelfile...
Is that similar option to set gpu-layer from the begining - like :
ollama run model.xyz -gpu-layer n ?

<!-- gh-comment-id:1881794494 --> @JoseConseco commented on GitHub (Jan 8, 2024): awesome. It was one of the most annoying thing about ollama (having to create custom model, to change gpu layers. ) While `/set parameter num_gpu 12` works - model is reloaded after next prompt, after setting gpu-layes. Will have to test if this helps, if model is to big to load into vram. I suppose in that case ollama will just error out, and I wont be able to `/set parameter num_gpu 12` right? In that case user will have to create new modelfile... Is that similar option to set gpu-layer from the begining - like : `ollama run model.xyz -gpu-layer n ` ?
GiteaMirror commented 2026-04-28 04:15:37 -05:00
Author
Owner
Copy Link

@BruceMacD commented on GitHub (Jan 9, 2024):

@JoseConseco setting it as a flag isn't an option right now, however this is a lot of work going on right now to load the optimal number of layers by default when a model is run.

<!-- gh-comment-id:1883427235 --> @BruceMacD commented on GitHub (Jan 9, 2024): @JoseConseco setting it as a flag isn't an option right now, however this is a lot of work going on right now to load the optimal number of layers by default when a model is run.
GiteaMirror commented 2026-04-28 04:15:40 -05:00
Author
Owner
Copy Link

@yangmingming commented on GitHub (Jun 20, 2024):

We can set the value of num_gpu through the following parameters. I didn't find how to get the current value? And what indicators can be referenced for setting this value?

/set parameter num_gpu
<!-- gh-comment-id:2180272988 --> @yangmingming commented on GitHub (Jun 20, 2024): We can set the value of num_gpu through the following parameters. I didn't find how to get the current value? And what indicators can be referenced for setting this value? ``` /set parameter num_gpu ```
GiteaMirror commented 2026-04-28 04:15:44 -05:00
Author
Owner
Copy Link

@dhiltgen commented on GitHub (Jul 24, 2024):

Ultimately the goal is users shouldn't have to adjust the num_gpu setting, and Ollama should load the optimal number of layers given the available VRAM. If there are bugs that cause us to load an incorrect number of layers, this setting can be used to workaround those bugs until we get them fixed. The setting can be specified via CLI, API, or within the Modelfile. I'm also planning to add another mechanism to workaround memory prediction bugs in #5922

<!-- gh-comment-id:2248948890 --> @dhiltgen commented on GitHub (Jul 24, 2024): Ultimately the goal is users shouldn't have to adjust the num_gpu setting, and Ollama should load the optimal number of layers given the available VRAM. If there are bugs that cause us to load an incorrect number of layers, this setting can be used to workaround those bugs until we get them fixed. The setting can be specified via CLI, API, or within the Modelfile. I'm also planning to add another mechanism to workaround memory prediction bugs in #5922
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
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#47571
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?