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 #14965] Start inferencing on CPU while the model weights are being uploaded to GPU VRAM #71683

New Issue
Closed
opened 2026-05-05 02:20:32 -05:00 by GiteaMirror · 1 comment
GiteaMirror commented 2026-05-05 02:20:32 -05:00
Owner
Copy Link

Originally created by @AIWintermuteAI on GitHub (Mar 19, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/14965

As per title - I'm wondering if this is possible.
For more context: my situation might be similar to fair share of people using ollama. I run open webui + ollama locally and use it myself + share with family members. There are multiple models downloaded to choose from, so always keeping one model in VRAM is not desirable.
There is quite a bit of delay where the model being loaded to VRAM. Perhaps, it is possible to utilize CPU for slower inference during that time, so at least the users would see the response being generated?

Originally created by @AIWintermuteAI on GitHub (Mar 19, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/14965 As per title - I'm wondering if this is possible. For more context: my situation might be similar to fair share of people using ollama. I run open webui + ollama locally and use it myself + share with family members. There are multiple models downloaded to choose from, so always keeping one model in VRAM is not desirable. There is quite a bit of delay where the model being loaded to VRAM. Perhaps, it is possible to utilize CPU for slower inference during that time, so at least the users would see the response being generated?
GiteaMirror added the feature request label 2026-05-05 02:20:32 -05:00
GiteaMirror commented 2026-05-05 02:20:34 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Mar 19, 2026):

Generating a single token requires a full forward pass through the model weights, so if inference starts as soon as the first layer is loaded, the user still won't see a response until the model is fully loaded. If you are OK with slower response, you could configure some of the less frequently used models to run only in RAM by setting num_gpu, that way they won't evict the main model from the GPU.

<!-- gh-comment-id:4093480072 --> @rick-github commented on GitHub (Mar 19, 2026): Generating a single token requires a full forward pass through the model weights, so if inference starts as soon as the first layer is loaded, the user still won't see a response until the model is fully loaded. If you are OK with slower response, you could configure some of the less frequently used models to run only in RAM by setting [`num_gpu`](https://github.com/ollama/ollama/issues/6950#issuecomment-2373663650), that way they won't evict the main model from the GPU.
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
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#71683
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?