github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 16:40:08 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #8274] Ollama hangs without timeout, Ollama model is consuming full CPU or GPU #31052

New Issue
Closed
opened 2026-04-22 11:10:46 -05:00 by GiteaMirror · 7 comments
GiteaMirror commented 2026-04-22 11:10:46 -05:00
Owner
Copy Link

Originally created by @ttww on GitHub (Dec 31, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/8274

What is the issue?

Doing Ollama connections with langchain the API hangs, depending on the input.
No information is loggend on the Ollama serve side with debug option 3.

I have attached a test case (Python program, test image, and README) to reproduce it.
ollama_hang.tgz

Changing the prompt, may change the situation (see c't Forum)

OS

Linux, macOS

GPU

Intel, Apple

CPU

AMD, Apple

Ollama version

0.5.4

Originally created by @ttww on GitHub (Dec 31, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/8274 ### What is the issue? Doing Ollama connections with langchain the API hangs, depending on the input. No information is loggend on the Ollama serve side with debug option 3. I have attached a test case (Python program, test image, and README) to reproduce it. [ollama_hang.tgz](https://github.com/user-attachments/files/18281727/ollama_hang.tgz) Changing the prompt, may change the situation (see [c't Forum](https://www.heise.de/forum/heise-online/Kommentare/Wie-eine-lokale-KI-die-Fotosammlung-auf-dem-NAS-verschlagworten-kann/Haengt-sich-bei-jedem-2-3-Bild-auf/thread-7564685/#posting_43932391)) ### OS Linux, macOS ### GPU Intel, Apple ### CPU AMD, Apple ### Ollama version 0.5.4
GiteaMirror added the bug label 2026-04-22 11:10:46 -05:00
GiteaMirror commented 2026-04-22 11:10:48 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Dec 31, 2024):

ollama seems to work fine.

$ ollama run llava:v1.6 Please find very precise keywords and separate them with commas. ./IMG_8240.JPG 
Added image './IMG_8240.JPG'
 insect, bug, aquatic, microscope, pest, nature, water, tiny, creature, monitor, image, screen, photo, health, environment, ecosystem, biodiversity, science, observation, species
<!-- gh-comment-id:2566509907 --> @rick-github commented on GitHub (Dec 31, 2024): ollama seems to work fine. ```console $ ollama run llava:v1.6 Please find very precise keywords and separate them with commas. ./IMG_8240.JPG Added image './IMG_8240.JPG' insect, bug, aquatic, microscope, pest, nature, water, tiny, creature, monitor, image, screen, photo, health, environment, ecosystem, biodiversity, science, observation, species ```
GiteaMirror commented 2026-04-22 11:10:49 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Dec 31, 2024):

--- ollama_hang/ollama_hang.py	2024-12-31 23:08:01.000000000 +1000
+++ ollama_hang.8274/ollama_hang.py	2025-01-01 01:04:02.098282867 +1000
@@ -27,7 +27,7 @@
 
 def main():
     # Open connection
-    llava_model = Ollama(model="llava:v1.6", base_url=OLLAMA_BASE_URL, temperature=0, verbose=True)
+    llava_model = Ollama(model="llava:v1.6", base_url=OLLAMA_BASE_URL, temperature=0, verbose=True, num_predict=200)
     prompt = "Please find very precise keywords and separate them with commas."
 
     # Loop over images

$ docker compose up
[+] Running 1/0
 ✔ Container ollama_hang8274-i8274-1  Created                                                                                                                                                       0.0s 
Attaching to i8274-1
i8274-1  | /app/ollama_hang.py:30: LangChainDeprecationWarning: The class `Ollama` was deprecated in LangChain 0.3.1 and will be removed in 1.0.0. An updated version of the class exists in the :class:`~langchain-ollama package and should be used instead. To use it run `pip install -U :class:`~langchain-ollama` and import as `from :class:`~langchain_ollama import OllamaLLM``.
i8274-1  |   llava_model = Ollama(model="llava:v1.6", base_url=OLLAMA_BASE_URL, temperature=0, verbose=True, num_predict=200)
i8274-1  | Marker scan hit start of image data
i8274-1  | Image 'test_images/IMG_8240.JPG' processing...
i8274-1  | crab,aquatic,animal,screen,monitor,image,computer,mouse,finger,person,hand,water,fish,tank,pet,home,indoor,outdoor,nature,wildlife,insect,shell,claw,antennae,eye,leg,body,color,texture,pattern,habitat,environment,ocean,sea,river,lake,pond,beach,sand,rock,coral,algae,plant,food,diet,survival,adaptation,evolution,biology,ecology,photography,video,technology,device,software,application,program,operatingsystem,userinterface,digital,virtual,reality,3D,imageprocessing,enhancement,zoom,focus,resolution,pixel,display,screensize,aspectratio,colordepth,brightness,contrast,saturation,hue,temperature,humidity,light,shadow,reflection,refraction,lens,camera,photographyequipment,tripod,
i8274-1  | ollama_hang.py
i8274-1  | save to 'test_images_results/IMG_8240_meta.jpg'
i8274-1 exited with code 0
<!-- gh-comment-id:2566522678 --> @rick-github commented on GitHub (Dec 31, 2024): ```diff --- ollama_hang/ollama_hang.py 2024-12-31 23:08:01.000000000 +1000 +++ ollama_hang.8274/ollama_hang.py 2025-01-01 01:04:02.098282867 +1000 @@ -27,7 +27,7 @@ def main(): # Open connection - llava_model = Ollama(model="llava:v1.6", base_url=OLLAMA_BASE_URL, temperature=0, verbose=True) + llava_model = Ollama(model="llava:v1.6", base_url=OLLAMA_BASE_URL, temperature=0, verbose=True, num_predict=200) prompt = "Please find very precise keywords and separate them with commas." # Loop over images ``` ```console $ docker compose up [+] Running 1/0 ✔ Container ollama_hang8274-i8274-1 Created 0.0s Attaching to i8274-1 i8274-1 | /app/ollama_hang.py:30: LangChainDeprecationWarning: The class `Ollama` was deprecated in LangChain 0.3.1 and will be removed in 1.0.0. An updated version of the class exists in the :class:`~langchain-ollama package and should be used instead. To use it run `pip install -U :class:`~langchain-ollama` and import as `from :class:`~langchain_ollama import OllamaLLM``. i8274-1 | llava_model = Ollama(model="llava:v1.6", base_url=OLLAMA_BASE_URL, temperature=0, verbose=True, num_predict=200) i8274-1 | Marker scan hit start of image data i8274-1 | Image 'test_images/IMG_8240.JPG' processing... i8274-1 | crab,aquatic,animal,screen,monitor,image,computer,mouse,finger,person,hand,water,fish,tank,pet,home,indoor,outdoor,nature,wildlife,insect,shell,claw,antennae,eye,leg,body,color,texture,pattern,habitat,environment,ocean,sea,river,lake,pond,beach,sand,rock,coral,algae,plant,food,diet,survival,adaptation,evolution,biology,ecology,photography,video,technology,device,software,application,program,operatingsystem,userinterface,digital,virtual,reality,3D,imageprocessing,enhancement,zoom,focus,resolution,pixel,display,screensize,aspectratio,colordepth,brightness,contrast,saturation,hue,temperature,humidity,light,shadow,reflection,refraction,lens,camera,photographyequipment,tripod, i8274-1 | ollama_hang.py i8274-1 | save to 'test_images_results/IMG_8240_meta.jpg' i8274-1 exited with code 0 ```
GiteaMirror commented 2026-04-22 11:10:50 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Dec 31, 2024):

--- ollama_hang/ollama_hang.py	2024-12-31 23:08:01.000000000 +1000
+++ ollama_hang.8274/ollama_hang.py	2025-01-01 01:12:57.593001024 +1000
@@ -27,7 +27,7 @@
 
 def main():
     # Open connection
-    llava_model = Ollama(model="llava:v1.6", base_url=OLLAMA_BASE_URL, temperature=0, verbose=True)
+    llava_model = Ollama(model="llava:v1.6", base_url=OLLAMA_BASE_URL, verbose=True)
     prompt = "Please find very precise keywords and separate them with commas."
 
     # Loop over images

$ docker compose up 
[+] Running 1/0
 ✔ Container ollama_hang8274-i8274-1  Created                                                                                                                0.0s 
Attaching to i8274-1
i8274-1  | /app/ollama_hang.py:30: LangChainDeprecationWarning: The class `Ollama` was deprecated in LangChain 0.3.1 and will be removed in 1.0.0. An updated version of the class exists in the :class:`~langchain-ollama package and should be used instead. To use it run `pip install -U :class:`~langchain-ollama` and import as `from :class:`~langchain_ollama import OllamaLLM``.
i8274-1  |   llava_model = Ollama(model="llava:v1.6", base_url=OLLAMA_BASE_URL, verbose=True)
i8274-1  | Marker scan hit start of image data
i8274-1  | Image 'test_images/IMG_8240.JPG' processing...
i8274-1  | mouse,rat,hamster,computermouse,videocamera,aquarium,fish,snail,crustacean,monitor,microscope
i8274-1  | ollama_hang.py
i8274-1  | save to 'test_images_results/IMG_8240_meta.jpg'
i8274-1 exited with code 0
<!-- gh-comment-id:2566526960 --> @rick-github commented on GitHub (Dec 31, 2024): ```diff --- ollama_hang/ollama_hang.py 2024-12-31 23:08:01.000000000 +1000 +++ ollama_hang.8274/ollama_hang.py 2025-01-01 01:12:57.593001024 +1000 @@ -27,7 +27,7 @@ def main(): # Open connection - llava_model = Ollama(model="llava:v1.6", base_url=OLLAMA_BASE_URL, temperature=0, verbose=True) + llava_model = Ollama(model="llava:v1.6", base_url=OLLAMA_BASE_URL, verbose=True) prompt = "Please find very precise keywords and separate them with commas." # Loop over images ``` ```console $ docker compose up [+] Running 1/0 ✔ Container ollama_hang8274-i8274-1 Created 0.0s Attaching to i8274-1 i8274-1 | /app/ollama_hang.py:30: LangChainDeprecationWarning: The class `Ollama` was deprecated in LangChain 0.3.1 and will be removed in 1.0.0. An updated version of the class exists in the :class:`~langchain-ollama package and should be used instead. To use it run `pip install -U :class:`~langchain-ollama` and import as `from :class:`~langchain_ollama import OllamaLLM``. i8274-1 | llava_model = Ollama(model="llava:v1.6", base_url=OLLAMA_BASE_URL, verbose=True) i8274-1 | Marker scan hit start of image data i8274-1 | Image 'test_images/IMG_8240.JPG' processing... i8274-1 | mouse,rat,hamster,computermouse,videocamera,aquarium,fish,snail,crustacean,monitor,microscope i8274-1 | ollama_hang.py i8274-1 | save to 'test_images_results/IMG_8240_meta.jpg' i8274-1 exited with code 0 ```
GiteaMirror commented 2026-04-22 11:10:50 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Dec 31, 2024):

--- ollama_hang/ollama_hang.py	2024-12-31 23:08:01.000000000 +1000
+++ ollama_hang.8274/ollama_hang.py	2025-01-01 01:25:49.994798877 +1000
@@ -28,7 +28,7 @@
 def main():
     # Open connection
     llava_model = Ollama(model="llava:v1.6", base_url=OLLAMA_BASE_URL, temperature=0, verbose=True)
-    prompt = "Please find very precise keywords and separate them with commas."
+    prompt = "Please find ten very precise keywords and separate them with commas."
 
     # Loop over images
     for img_filename in glob.glob(f"{SOURCE_DIR}/*.JPG"):  

$ docker compose up 
[+] Running 1/0
 ✔ Container ollama_hang8274-i8274-1  Created                                                                  0.0s 
Attaching to i8274-1
i8274-1  | /app/ollama_hang.py:30: LangChainDeprecationWarning: The class `Ollama` was deprecated in LangChain 0.3.1 and will be removed in 1.0.0. An updated version of the class exists in the :class:`~langchain-ollama package and should be used instead. To use it run `pip install -U :class:`~langchain-ollama` and import as `from :class:`~langchain_ollama import OllamaLLM``.
i8274-1  |   llava_model = Ollama(model="llava:v1.6", base_url=OLLAMA_BASE_URL, temperature=0, verbose=True)
i8274-1  | Marker scan hit start of image data
i8274-1  | Image 'test_images/IMG_8240.JPG' processing...
i8274-1  | insect,aquatic,microscopic,magnification,close-up,image,monitor,computer,screen,bug
i8274-1  | ollama_hang.py
i8274-1  | save to 'test_images_results/IMG_8240_meta.jpg'
i8274-1 exited with code 0
<!-- gh-comment-id:2566536094 --> @rick-github commented on GitHub (Dec 31, 2024): ```diff --- ollama_hang/ollama_hang.py 2024-12-31 23:08:01.000000000 +1000 +++ ollama_hang.8274/ollama_hang.py 2025-01-01 01:25:49.994798877 +1000 @@ -28,7 +28,7 @@ def main(): # Open connection llava_model = Ollama(model="llava:v1.6", base_url=OLLAMA_BASE_URL, temperature=0, verbose=True) - prompt = "Please find very precise keywords and separate them with commas." + prompt = "Please find ten very precise keywords and separate them with commas." # Loop over images for img_filename in glob.glob(f"{SOURCE_DIR}/*.JPG"): ``` ```console $ docker compose up [+] Running 1/0 ✔ Container ollama_hang8274-i8274-1 Created 0.0s Attaching to i8274-1 i8274-1 | /app/ollama_hang.py:30: LangChainDeprecationWarning: The class `Ollama` was deprecated in LangChain 0.3.1 and will be removed in 1.0.0. An updated version of the class exists in the :class:`~langchain-ollama package and should be used instead. To use it run `pip install -U :class:`~langchain-ollama` and import as `from :class:`~langchain_ollama import OllamaLLM``. i8274-1 | llava_model = Ollama(model="llava:v1.6", base_url=OLLAMA_BASE_URL, temperature=0, verbose=True) i8274-1 | Marker scan hit start of image data i8274-1 | Image 'test_images/IMG_8240.JPG' processing... i8274-1 | insect,aquatic,microscopic,magnification,close-up,image,monitor,computer,screen,bug i8274-1 | ollama_hang.py i8274-1 | save to 'test_images_results/IMG_8240_meta.jpg' i8274-1 exited with code 0 ```
GiteaMirror commented 2026-04-22 11:10:51 -05:00
Author
Owner
Copy Link

@ttww commented on GitHub (Jan 1, 2025):

ollama seems to work fine.

$ ollama run llava:v1.6 Please find very precise keywords and separate them with commas. ./IMG_8240.JPG 
Added image './IMG_8240.JPG'
 insect, bug, aquatic, microscope, pest, nature, water, tiny, creature, monitor, image, screen, photo, health, environment, ecosystem, biodiversity, science, observation, species

Does it work with your changes or only with the original code?
If it works with your changes, please don’t take this the wrong way, but who can guarantee it won’t happen again with a different image?
In my view, the API needs to behave consistently. Ideally, it should show an appropriate error message indicating the cause.
Right now, it feels like luck whether it works or not. It would be best to identify and fix the root cause.

BTW: Using your prompt with the added "ten" works also in my case, but it "feels" wrong to fix it that way :-)

<!-- gh-comment-id:2567071614 --> @ttww commented on GitHub (Jan 1, 2025): > ollama seems to work fine. > > ``` > $ ollama run llava:v1.6 Please find very precise keywords and separate them with commas. ./IMG_8240.JPG > Added image './IMG_8240.JPG' > insect, bug, aquatic, microscope, pest, nature, water, tiny, creature, monitor, image, screen, photo, health, environment, ecosystem, biodiversity, science, observation, species > ``` Does it work with your changes or only with the original code? If it works with your changes, please don’t take this the wrong way, but who can guarantee it won’t happen again with a different image? In my view, the API needs to behave consistently. Ideally, it should show an appropriate error message indicating the cause. Right now, it feels like luck whether it works or not. It would be best to identify and fix the root cause. BTW: Using your prompt with the added "ten" works also in my case, but it "feels" wrong to fix it that way :-)
GiteaMirror commented 2026-04-22 11:10:52 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jan 1, 2025):

The root cause is the imprecise prompting. When you see no response, the model is not hanging, it is generating tokens. If you limit the token generation in any of the depicted ways, the generation will stop and you get a response.

<!-- gh-comment-id:2567075286 --> @rick-github commented on GitHub (Jan 1, 2025): The root cause is the imprecise prompting. When you see no response, the model is not hanging, it is generating tokens. If you limit the token generation in any of the depicted ways, the generation will stop and you get a response.
GiteaMirror commented 2026-04-22 11:10:53 -05:00
Author
Owner
Copy Link

@ttww commented on GitHub (Jan 1, 2025):

The root cause is the imprecise prompting. When you see no response, the model is not hanging, it is generating tokens. If you limit the token generation in any of the depicted ways, the generation will stop and you get a response.

OK, thanks for the explanation; that makes sense.
If you know more about how the models work, it’s actually obvious.
From my perspective, we can close this issue.

A timeout parameter in the API might be useful, but that’s another topic.

<!-- gh-comment-id:2567078924 --> @ttww commented on GitHub (Jan 1, 2025): > The root cause is the imprecise prompting. When you see no response, the model is not hanging, it is generating tokens. If you limit the token generation in any of the depicted ways, the generation will stop and you get a response. OK, thanks for the explanation; that makes sense. If you know more about how the models work, it’s actually obvious. From my perspective, we can close this issue. A timeout parameter in the API might be useful, but that’s another topic.
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#31052
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?