github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 16:11:34 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #3009] feat: add "unload model" command/endpoint #63885

New Issue
Closed
opened 2026-05-03 15:19:02 -05:00 by GiteaMirror · 7 comments
GiteaMirror commented 2026-05-03 15:19:02 -05:00
Owner
Copy Link

Originally created by @knoopx on GitHub (Mar 8, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/3009

There's no way to unload a model from VRAM other than killing/restarting ollama and that requires local system access and privileges. Given ollama is mostly used on limited devices, a command/api endpoint would be fantastic.

Originally created by @knoopx on GitHub (Mar 8, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/3009 There's no way to unload a model from VRAM other than killing/restarting ollama and that requires local system access and privileges. Given ollama is mostly used on limited devices, a command/api endpoint would be fantastic.
GiteaMirror added the feature request label 2026-05-03 15:19:02 -05:00
GiteaMirror commented 2026-05-03 15:19:05 -05:00
Author
Owner
Copy Link

@phong-phuong commented on GitHub (Mar 8, 2024):

From what I've seen, the model gets unloaded after a certain amount of time of inactivity, or using a model switch.

<!-- gh-comment-id:1986560967 --> @phong-phuong commented on GitHub (Mar 8, 2024): From what I've seen, the model gets unloaded after a certain amount of time of inactivity, or using a model switch.
GiteaMirror commented 2026-05-03 15:19:05 -05:00
Author
Owner
Copy Link

@commit4ever commented on GitHub (Mar 9, 2024):

ecc133d843/docs/faq.md (L189) use keep_alive param

<!-- gh-comment-id:1986749896 --> @commit4ever commented on GitHub (Mar 9, 2024): https://github.com/ollama/ollama/blob/ecc133d843c8567b27ff3bdc9ff811ecad99281a/docs/faq.md?plain=1#L189 use keep_alive param
GiteaMirror commented 2026-05-03 15:19:05 -05:00
Author
Owner
Copy Link

@knoopx commented on GitHub (Mar 9, 2024):

ecc133d843/docs/faq.md (L189)

use keep_alive param

  • any negative number which will keep the model loaded in memory (e.g. -1 or "-1m")

Does the opposite actually.

<!-- gh-comment-id:1986823344 --> @knoopx commented on GitHub (Mar 9, 2024): > https://github.com/ollama/ollama/blob/ecc133d843c8567b27ff3bdc9ff811ecad99281a/docs/faq.md?plain=1#L189 > > use keep_alive param * any negative number which will keep the model loaded in memory (e.g. -1 or "-1m") Does the opposite actually.
GiteaMirror commented 2026-05-03 15:19:07 -05:00
Author
Owner
Copy Link

@alexandervnuchkov commented on GitHub (Mar 11, 2024):

Would be great if this default behavior could be set via an environment variable (for Docker), if that is possible

<!-- gh-comment-id:1988620250 --> @alexandervnuchkov commented on GitHub (Mar 11, 2024): Would be great if this default behavior could be set via an environment variable (for Docker), if that is possible
GiteaMirror commented 2026-05-03 15:19:08 -05:00
Author
Owner
Copy Link

@commit4ever commented on GitHub (Mar 11, 2024):

I use a set interval in secs and that has worked well.

<!-- gh-comment-id:1989391589 --> @commit4ever commented on GitHub (Mar 11, 2024): I use a set interval in secs and that has worked well.
GiteaMirror commented 2026-05-03 15:19:10 -05:00
Author
Owner
Copy Link

@pdevine commented on GitHub (Mar 12, 2024):

Hey @knoopx You can actually do this by calling curl http://localhost:11434/api/generate -d '{"model": "llama2", "keep_alive": 0}' (not with -1 which will always leave the model loaded). That will immediately unload the model without doing any generation.

You can confirm this with:

lsof -n | grep ollama | grep blobs | awk "{ print \$9 }"

This will show which blob is loaded into memory, and then when you hit the generate endpoint you can see that the blob was unloaded. I'm going to go ahead and close the issue, but feel free to keep commenting.

<!-- gh-comment-id:1992588837 --> @pdevine commented on GitHub (Mar 12, 2024): Hey @knoopx You can actually do this by calling `curl http://localhost:11434/api/generate -d '{"model": "llama2", "keep_alive": 0}'` (not with `-1` which will always leave the model loaded). That will immediately unload the model without doing any generation. You can confirm this with: ``` lsof -n | grep ollama | grep blobs | awk "{ print \$9 }" ``` This will show which blob is loaded into memory, and then when you hit the generate endpoint you can see that the blob was unloaded. I'm going to go ahead and close the issue, but feel free to keep commenting.
GiteaMirror commented 2026-05-03 15:19:12 -05:00
Author
Owner
Copy Link

@knoopx commented on GitHub (Mar 14, 2024):

Hey @knoopx You can actually do this by calling curl http://localhost:11434/api/generate -d '{"model": "llama2", "keep_alive": 0}' (not with -1 which will always leave the model loaded). That will immediately unload the model without doing any generation.

You can confirm this with:

lsof -n | grep ollama | grep blobs | awk "{ print \$9 }"

This will show which blob is loaded into memory, and then when you hit the generate endpoint you can see that the blob was unloaded. I'm going to go ahead and close the issue, but feel free to keep commenting.

nice as a workaround but still loads another model unnecessarily.

<!-- gh-comment-id:1997373117 --> @knoopx commented on GitHub (Mar 14, 2024): > Hey @knoopx You can actually do this by calling `curl http://localhost:11434/api/generate -d '{"model": "llama2", "keep_alive": 0}'` (not with `-1` which will always leave the model loaded). That will immediately unload the model without doing any generation. > > You can confirm this with: > > ``` > lsof -n | grep ollama | grep blobs | awk "{ print \$9 }" > ``` > > This will show which blob is loaded into memory, and then when you hit the generate endpoint you can see that the blob was unloaded. I'm going to go ahead and close the issue, but feel free to keep commenting. nice as a workaround but still loads another model unnecessarily.
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#63885
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?