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 #1339] MacOS opens kernel tasks doesn't unload model #47209

New Issue
Closed
opened 2026-04-28 03:26:00 -05:00 by GiteaMirror · 11 comments
GiteaMirror commented 2026-04-28 03:26:00 -05:00
Owner
Copy Link

Originally created by @igorcosta on GitHub (Dec 1, 2023).
Original GitHub issue: https://github.com/ollama/ollama/issues/1339

One of the things that makes me cringe is when swapping between models, it never releases the memory when I'm done using it. It's just piles up and I eventually have to restart my Mac.

Would memory optimisation being a target for next release?

Originally created by @igorcosta on GitHub (Dec 1, 2023). Original GitHub issue: https://github.com/ollama/ollama/issues/1339 One of the things that makes me cringe is when swapping between models, it never releases the memory when I'm done using it. It's just piles up and I eventually have to restart my Mac. Would memory optimisation being a target for next release?
GiteaMirror commented 2026-04-28 03:26:00 -05:00
Author
Owner
Copy Link

@pdevine commented on GitHub (Dec 1, 2023):

@igorcosta the memory for each model is released after 5 minutes, and it should release between loading each model. Can you be more specific about what you're seeing?

<!-- gh-comment-id:1835463392 --> @pdevine commented on GitHub (Dec 1, 2023): @igorcosta the memory for each model is released after 5 minutes, and it should release between loading each model. Can you be more specific about what you're seeing?
GiteaMirror commented 2026-04-28 03:26:01 -05:00
Author
Owner
Copy Link

@tjthejuggler commented on GitHub (Dec 1, 2023):

is there any way to not have to wait 5 minutes? I'm on ubuntu and I have to go in and manually sudo kill ollama in order to get my memory back every time

<!-- gh-comment-id:1835981144 --> @tjthejuggler commented on GitHub (Dec 1, 2023): is there any way to not have to wait 5 minutes? I'm on ubuntu and I have to go in and manually sudo kill ollama in order to get my memory back every time
GiteaMirror commented 2026-04-28 03:26:01 -05:00
Author
Owner
Copy Link

@jakenvac commented on GitHub (Dec 15, 2023):

I'm on macos and came to this issue wondering why there was still memory allocated after I ended the ollama run command. I wasn't aware of the 5 minute rule.

I can confirm this works as @pdevine describes and is perfectly acceptable behavior in my mind.

<!-- gh-comment-id:1858056094 --> @jakenvac commented on GitHub (Dec 15, 2023): I'm on macos and came to this issue wondering why there was still memory allocated after I ended the `ollama run` command. I wasn't aware of the 5 minute rule. I can confirm this works as @pdevine describes and is perfectly acceptable behavior in my mind.
GiteaMirror commented 2026-04-28 03:26:02 -05:00
Author
Owner
Copy Link

@realazizk commented on GitHub (Dec 16, 2023):

I use other models like whisper on the side with ollama, so i do this to unload model when i need vram, might be useful to some people.

kill -SIGUSR1 $(pgrep -f ollama)

code:
c9a1ee8f1c

<!-- gh-comment-id:1858813133 --> @realazizk commented on GitHub (Dec 16, 2023): I use other models like whisper on the side with ollama, so i do this to unload model when i need vram, might be useful to some people. ``` kill -SIGUSR1 $(pgrep -f ollama) ``` code: https://github.com/mohamed-aziz/ollama/commit/c9a1ee8f1c1e3e464d36d81089b7b7d827675846
GiteaMirror commented 2026-04-28 03:26:02 -05:00
Author
Owner
Copy Link

@Natfan commented on GitHub (Jan 25, 2024):

I actually have the opposite issue, I would like my ollama models to be cached for longer. Is there any way that I can programmatically get the model currently loaded (outside of scanning the log), so I can poll /api/embeddings every few minutes to keep it up and running?

<!-- gh-comment-id:1911173965 --> @Natfan commented on GitHub (Jan 25, 2024): I actually have the opposite issue, I would like my ollama models to be cached for longer. Is there any way that I can programmatically get the model currently loaded (outside of scanning the log), so I can poll /api/embeddings every few minutes to keep it up and running?
GiteaMirror commented 2026-04-28 03:26:02 -05:00
Author
Owner
Copy Link

@mxyng commented on GitHub (Jan 25, 2024):

so I can poll /api/embeddings every few minutes to keep it up and running?

Yes, any of /api/embeddings, /api/chat, or /api/generate will work for this

<!-- gh-comment-id:1911184411 --> @mxyng commented on GitHub (Jan 25, 2024): > so I can poll /api/embeddings every few minutes to keep it up and running? Yes, any of `/api/embeddings`, `/api/chat`, or `/api/generate` will work for this
GiteaMirror commented 2026-04-28 03:26:02 -05:00
Author
Owner
Copy Link

@tjthejuggler commented on GitHub (Jan 26, 2024):

@mxyng is there anything that will cancel the ollama model immediately? So i don't have to do a sudo kill and can use the guy again immediately.

<!-- gh-comment-id:1912797597 --> @tjthejuggler commented on GitHub (Jan 26, 2024): @mxyng is there anything that will cancel the ollama model immediately? So i don't have to do a sudo kill and can use the guy again immediately.
GiteaMirror commented 2026-04-28 03:26:03 -05:00
Author
Owner
Copy Link

@mxyng commented on GitHub (Jan 26, 2024):

With #2146, you can set keep_alive: 0 to immediately unload the model once the request is complete

<!-- gh-comment-id:1912822388 --> @mxyng commented on GitHub (Jan 26, 2024): With #2146, you can set `keep_alive: 0` to immediately unload the model once the request is complete
GiteaMirror commented 2026-04-28 03:26:03 -05:00
Author
Owner
Copy Link

@tjthejuggler commented on GitHub (Jan 29, 2024):

With #2146, you can set keep_alive: 0 to immediately unload the model once the request is complete

Thanks so much! How exactly can this be used? I tried it in python code and as an argument in a terminal command and both ways it just doesn't run at all, no error, just no LLMing happening.

<!-- gh-comment-id:1914184726 --> @tjthejuggler commented on GitHub (Jan 29, 2024): > With #2146, you can set `keep_alive: 0` to immediately unload the model once the request is complete Thanks so much! How exactly can this be used? I tried it in python code and as an argument in a terminal command and both ways it just doesn't run at all, no error, just no LLMing happening.
GiteaMirror commented 2026-04-28 03:26:04 -05:00
Author
Owner
Copy Link

@OliChase404 commented on GitHub (Feb 28, 2024):

This might help, from the faq.md file:

How do I keep a model loaded in memory or make it unload immediately?

By default models are kept in memory for 5 minutes before being unloaded. This allows for quicker response times if you are making numerous requests to the LLM. You may, however, want to free up the memory before the 5 minutes have elapsed or keep the model loaded indefinitely. Use the keep_alive parameter with either the /api/generate and /api/chat API endpoints to control how long the model is left in memory.
The keep_alive parameter can be set to:

  • a duration string (such as "10m" or "24h")
  • a number in seconds (such as 3600)
  • any negative number which will keep the model loaded in memory (e.g. -1 or "-1m")
  • '0' which will unload the model immediately after generating a response
    For example, to preload a model and leave it in memory use:
curl http://localhost:11434/api/generate -d '{"model": "llama2", "keep_alive": -1}'

To unload the model and free up memory use:

curl http://localhost:11434/api/generate -d '{"model": "llama2", "keep_alive": 0}'

----------------------------------------------

So to immediately unload a model and free your vram run:

curl http://localhost:11434/api/generate -d '{"model": "MODELNAME", "keep_alive": 0}'

Replace MODELNAME with the name of the model currently loaded.
<!-- gh-comment-id:1967948376 --> @OliChase404 commented on GitHub (Feb 28, 2024): This might help, from the [faq.md](https://faq.md/) file: ----------------------------------------------------------- ## How do I keep a model loaded in memory or make it unload immediately? By default models are kept in memory for 5 minutes before being unloaded. This allows for quicker response times if you are making numerous requests to the LLM. You may, however, want to free up the memory before the 5 minutes have elapsed or keep the model loaded indefinitely. Use the `keep_alive` parameter with either the `/api/generate` and `/api/chat` API endpoints to control how long the model is left in memory. The `keep_alive` parameter can be set to: * a duration string (such as "10m" or "24h") * a number in seconds (such as 3600) * any negative number which will keep the model loaded in memory (e.g. -1 or "-1m") * '0' which will unload the model immediately after generating a response For example, to preload a model and leave it in memory use: ```shell curl http://localhost:11434/api/generate -d '{"model": "llama2", "keep_alive": -1}' ``` To unload the model and free up memory use: ```shell curl http://localhost:11434/api/generate -d '{"model": "llama2", "keep_alive": 0}' ---------------------------------------------- So to immediately unload a model and free your vram run: curl http://localhost:11434/api/generate -d '{"model": "MODELNAME", "keep_alive": 0}' Replace MODELNAME with the name of the model currently loaded.
GiteaMirror commented 2026-04-28 03:26:04 -05:00
Author
Owner
Copy Link

@AbdulmalekAI commented on GitHub (Aug 6, 2024):

This might help, from the faq.md file:

How do I keep a model loaded in memory or make it unload immediately?

By default models are kept in memory for 5 minutes before being unloaded. This allows for quicker response times if you are making numerous requests to the LLM. You may, however, want to free up the memory before the 5 minutes have elapsed or keep the model loaded indefinitely. Use the keep_alive parameter with either the /api/generate and /api/chat API endpoints to control how long the model is left in memory. The keep_alive parameter can be set to:

  • a duration string (such as "10m" or "24h")
  • a number in seconds (such as 3600)
  • any negative number which will keep the model loaded in memory (e.g. -1 or "-1m")
  • '0' which will unload the model immediately after generating a response
    For example, to preload a model and leave it in memory use:
curl http://localhost:11434/api/generate -d '{"model": "llama2", "keep_alive": -1}'

To unload the model and free up memory use:

curl http://localhost:11434/api/generate -d '{"model": "llama2", "keep_alive": 0}'

----------------------------------------------

So to immediately unload a model and free your vram run:

curl http://localhost:11434/api/generate -d '{"model": "MODELNAME", "keep_alive": 0}'

Replace MODELNAME with the name of the model currently loaded.

I used keep_alive in this command , But not working ollm = OllamaLLM(model="aya",keep_alive=0,temperature=0.1)
Is this parameter woriking for /api/generate and /api/chat API endpoints only ???

<!-- gh-comment-id:2270587439 --> @AbdulmalekAI commented on GitHub (Aug 6, 2024): > This might help, from the [faq.md](https://faq.md/) file: > > ## How do I keep a model loaded in memory or make it unload immediately? > By default models are kept in memory for 5 minutes before being unloaded. This allows for quicker response times if you are making numerous requests to the LLM. You may, however, want to free up the memory before the 5 minutes have elapsed or keep the model loaded indefinitely. Use the `keep_alive` parameter with either the `/api/generate` and `/api/chat` API endpoints to control how long the model is left in memory. The `keep_alive` parameter can be set to: > > * a duration string (such as "10m" or "24h") > * a number in seconds (such as 3600) > * any negative number which will keep the model loaded in memory (e.g. -1 or "-1m") > * '0' which will unload the model immediately after generating a response > For example, to preload a model and leave it in memory use: > > ```shell > curl http://localhost:11434/api/generate -d '{"model": "llama2", "keep_alive": -1}' > ``` > > To unload the model and free up memory use: > > ```shell > curl http://localhost:11434/api/generate -d '{"model": "llama2", "keep_alive": 0}' > > ---------------------------------------------- > > So to immediately unload a model and free your vram run: > > curl http://localhost:11434/api/generate -d '{"model": "MODELNAME", "keep_alive": 0}' > > Replace MODELNAME with the name of the model currently loaded. > ``` I used keep_alive in this command , But not working ollm = OllamaLLM(model="aya",keep_alive=0,temperature=0.1) Is this parameter woriking for /api/generate and /api/chat API endpoints only ???
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#47209
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?