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 #9902] a model run at GPU and another model run at CPU #68540

New Issue
Closed
opened 2026-05-04 14:23:01 -05:00 by GiteaMirror · 5 comments
GiteaMirror commented 2026-05-04 14:23:01 -05:00
Owner
Copy Link

Originally created by @ROBODRILL on GitHub (Mar 20, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/9902

What is the issue?

In docker, I copy the model from deepseek-r1:32b ,Add 2 parameter
PARAMETER num_ctx 131072
PARAMETER num_predict -1
in Modelfile,
named new model deepseek-r1:32b-max-context.

run deepseek-r1:32b-max-context,
then "ollama ps ":
the processors is 100%CPU,
run deepseek-r1:32b,
then "ollama ps ":
the processors is 100%GPU,

Relevant log output

OS

docker

GPU

tesla T4

CPU

No response

Ollama version

0.5.11

Originally created by @ROBODRILL on GitHub (Mar 20, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/9902 ### What is the issue? In docker, I copy the model from deepseek-r1:32b ,Add 2 parameter PARAMETER num_ctx 131072 PARAMETER num_predict -1 in Modelfile, named new model deepseek-r1:32b-max-context. run deepseek-r1:32b-max-context, then "ollama ps ": the processors is 100%CPU, run deepseek-r1:32b, then "ollama ps ": the processors is 100%GPU, ### Relevant log output ```shell ``` ### OS docker ### GPU tesla T4 ### CPU _No response_ ### Ollama version 0.5.11
GiteaMirror added the bug label 2026-05-04 14:23:01 -05:00
GiteaMirror commented 2026-05-04 14:23:07 -05:00
Author
Owner
Copy Link

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

The size of the context is larger than will fit in GPU VRAM, so the model runs in CPU.

<!-- gh-comment-id:2739664293 --> @rick-github commented on GitHub (Mar 20, 2025): The size of the context is larger than will fit in GPU VRAM, so the model runs in CPU.
GiteaMirror commented 2026-05-04 14:23:10 -05:00
Author
Owner
Copy Link

@sieveLau commented on GitHub (Mar 21, 2025):

Your context length is too large. Context length also takes VRAM to store, so try and reduce it to a lower one. In most cases, 32K would be enough for a 32B model, longer ones don't make much difference.

<!-- gh-comment-id:2743203448 --> @sieveLau commented on GitHub (Mar 21, 2025): Your context length is too large. Context length also takes VRAM to store, so try and reduce it to a lower one. In most cases, 32K would be enough for a 32B model, longer ones don't make much difference.
GiteaMirror commented 2026-05-04 14:23:12 -05:00
Author
Owner
Copy Link

@ROBODRILL commented on GitHub (Mar 25, 2025):

The size of the context is larger than will fit in GPU VRAM, so the model runs in CPU.

thankyou,My VARM is 128GB,so the context_size that I can config is ??

<!-- gh-comment-id:2749940726 --> @ROBODRILL commented on GitHub (Mar 25, 2025): > The size of the context is larger than will fit in GPU VRAM, so the model runs in CPU. thankyou,My VARM is 128GB,so the context_size that I can config is ??
GiteaMirror commented 2026-05-04 14:23:13 -05:00
Author
Owner
Copy Link

@ROBODRILL commented on GitHub (Mar 25, 2025):

Your context length is too large. Context length also takes VRAM to store, so try and reduce it to a lower one. In most cases, 32K would be enough for a 32B model, longer ones don't make much difference.

thankyou ,I set the 8K for the context_num,but I don't know what's size to config to match the VARM

<!-- gh-comment-id:2749942766 --> @ROBODRILL commented on GitHub (Mar 25, 2025): > Your context length is too large. Context length also takes VRAM to store, so try and reduce it to a lower one. In most cases, 32K would be enough for a 32B model, longer ones don't make much difference. thankyou ,I set the 8K for the context_num,but I don't know what's size to config to match the VARM
GiteaMirror commented 2026-05-04 14:23:14 -05:00
Author
Owner
Copy Link

@sieveLau commented on GitHub (Mar 25, 2025):

Your context length is too large. Context length also takes VRAM to store, so try and reduce it to a lower one. In most cases, 32K would be enough for a 32B model, longer ones don't make much difference.

thankyou ,I set the 8K for the context_num,but I don't know what's size to config to match the VARM

I think you can add one more parameter num_gpu 999 to force ollama load the model into the VRAM, ignoring its estimation. Then you can increase the num_ctx step by step to test where the limit is. But in this way you should monitor your VRAM through nvidia-smi instead of ollama ps, and watch the systemd ollama service output to see any warnings or errors.

<!-- gh-comment-id:2749961223 --> @sieveLau commented on GitHub (Mar 25, 2025): > > Your context length is too large. Context length also takes VRAM to store, so try and reduce it to a lower one. In most cases, 32K would be enough for a 32B model, longer ones don't make much difference. > > thankyou ,I set the 8K for the context_num,but I don't know what's size to config to match the VARM I think you can add one more parameter `num_gpu 999` to force ollama load the model into the VRAM, ignoring its estimation. Then you can increase the num_ctx step by step to test where the limit is. But in this way you should monitor your VRAM through `nvidia-smi` instead of `ollama ps`, and watch the systemd ollama service output to see any warnings or errors.
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#68540
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?