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 #4916] Newer models are having problems #28866

New Issue
Closed
opened 2026-04-22 07:25:31 -05:00 by GiteaMirror · 11 comments
GiteaMirror commented 2026-04-22 07:25:31 -05:00
Owner
Copy Link

Originally created by @iplayfast on GitHub (Jun 7, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/4916

What is the issue?

ollama version is 0.1.41

ollama run granite-code
pulling manifest 
pulling 02ab8cd2f514... 100% ▕████████████████████████████████████████████████████████████████████████████████████████████████▏ 2.0 GB                         
pulling 0d7c97d535b6... 100% ▕████████████████████████████████████████████████████████████████████████████████████████████████▏   26 B                         
pulling e50df8490144... 100% ▕████████████████████████████████████████████████████████████████████████████████████████████████▏  123 B                         
pulling 9893bb2c2917... 100% ▕████████████████████████████████████████████████████████████████████████████████████████████████▏  108 B                         
pulling 22b176fd8ef6... 100% ▕████████████████████████████████████████████████████████████████████████████████████████████████▏  485 B                         
verifying sha256 digest 
writing manifest 
removing any unused layers 
success 
>>> hello
Hola, ¿en qué puedo ayudarte?

>>> please respond in english
Hello! ¿What can I help you with in English?

qwen2 was giving garbage and then after doing the above, started behaving.

ollama run deepseek-v2
>>> hello
你好！有什么我可以帮助你的吗？

>>> please respond in english
当然，我可以用英语回答您的问题。如果您有任何问题或需要帮助，请随时告诉我。

llama3 behaves correctly through all this.

OS

Linux

GPU

Nvidia

CPU

Intel

Ollama version

0.1.41

Originally created by @iplayfast on GitHub (Jun 7, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/4916 ### What is the issue? ```ollama --version ollama version is 0.1.41 ``` ``` ollama run granite-code pulling manifest pulling 02ab8cd2f514... 100% ▕████████████████████████████████████████████████████████████████████████████████████████████████▏ 2.0 GB pulling 0d7c97d535b6... 100% ▕████████████████████████████████████████████████████████████████████████████████████████████████▏ 26 B pulling e50df8490144... 100% ▕████████████████████████████████████████████████████████████████████████████████████████████████▏ 123 B pulling 9893bb2c2917... 100% ▕████████████████████████████████████████████████████████████████████████████████████████████████▏ 108 B pulling 22b176fd8ef6... 100% ▕████████████████████████████████████████████████████████████████████████████████████████████████▏ 485 B verifying sha256 digest writing manifest removing any unused layers success >>> hello Hola, ¿en qué puedo ayudarte? >>> please respond in english Hello! ¿What can I help you with in English? ``` qwen2 was giving garbage and then after doing the above, started behaving. ``` ollama run deepseek-v2 >>> hello 你好！有什么我可以帮助你的吗？ >>> please respond in english 当然，我可以用英语回答您的问题。如果您有任何问题或需要帮助，请随时告诉我。 ``` llama3 behaves correctly through all this. ### OS Linux ### GPU Nvidia ### CPU Intel ### Ollama version 0.1.41
GiteaMirror added the bug label 2026-04-22 07:25:31 -05:00
GiteaMirror commented 2026-04-22 07:25:32 -05:00
Author
Owner
Copy Link

@Starlitnightly commented on GitHub (Jun 8, 2024):

I meet the same problem

verifying sha256 digest 
writing manifest 
removing any unused layers 
success 
root@ab-Z370-HD3:/mnt/home/zehuazeng# ollama run qwen2:7b
>>> hello
GGJZ:GDSN

>>> please respond in english
GladGDSGtoNN assistGGGGGGGGGGGGGGGGGGGGGGGGGGGGGGG

>>> hello
GGJGGGGGK,GGGGGGGGGG!GGG
<!-- gh-comment-id:2155860968 --> @Starlitnightly commented on GitHub (Jun 8, 2024): I meet the same problem ```shell verifying sha256 digest writing manifest removing any unused layers success root@ab-Z370-HD3:/mnt/home/zehuazeng# ollama run qwen2:7b >>> hello GGJZ:GDSN >>> please respond in english GladGDSGtoNN assistGGGGGGGGGGGGGGGGGGGGGGGGGGGGGGG >>> hello GGJGGGGGK,GGGGGGGGGG!GGG ```
GiteaMirror commented 2026-04-22 07:25:33 -05:00
Author
Owner
Copy Link

@Axenide commented on GitHub (Jun 8, 2024):

I meet the same problem

verifying sha256 digest 
writing manifest 
removing any unused layers 
success 
root@ab-Z370-HD3:/mnt/home/zehuazeng# ollama run qwen2:7b
>>> hello
GGJZ:GDSN

>>> please respond in english
GladGDSGtoNN assistGGGGGGGGGGGGGGGGGGGGGGGGGGGGGGG

>>> hello
GGJGGGGGK,GGGGGGGGGG!GGG

I have this exact same output.

<!-- gh-comment-id:2156168363 --> @Axenide commented on GitHub (Jun 8, 2024): > I meet the same problem > > ```shell > verifying sha256 digest > writing manifest > removing any unused layers > success > root@ab-Z370-HD3:/mnt/home/zehuazeng# ollama run qwen2:7b > >>> hello > GGJZ:GDSN > > >>> please respond in english > GladGDSGtoNN assistGGGGGGGGGGGGGGGGGGGGGGGGGGGGGGG > > >>> hello > GGJGGGGGK,GGGGGGGGGG!GGG > ``` I have this exact same output.
GiteaMirror commented 2026-04-22 07:25:33 -05:00
Author
Owner
Copy Link

@igorschlum commented on GitHub (Jun 8, 2024):

I tried with version 0.1.42 of Ollama on MacOS with 32GB Ram.

with granite-code I get answers in english.
But for deepseek-v2, I can reproduce also, it's answering in Chiness even if I ask to answer in English.
For qwen2:7b, the answer is in English

(base) igor@macigor-2 ~ % ollama run granite-code
pulling manifest
pulling 02ab8cd2f514... 100% ▕███████████████████████████████████████████████▏ 2.0 GB
pulling 0d7c97d535b6... 100% ▕███████████████████████████████████████████████▏ 26 B
pulling e50df8490144... 100% ▕███████████████████████████████████████████████▏ 123 B
pulling 9893bb2c2917... 100% ▕███████████████████████████████████████████████▏ 108 B
pulling 22b176fd8ef6... 100% ▕███████████████████████████████████████████████▏ 485 B
verifying sha256 digest
writing manifest
removing any unused layers
success

hello
Hello! How can I assist you today?

(base) igor@macigor-2 ~ % ollama run qwen2:7b
pulling manifest
pulling 43f7a214e532... 100% ▕██████████████████████████████████████████████ 4.4 GB
pulling 62fbfd9ed093... 100% ▕██████████████████████████████████████████████▏ 182 B
pulling c156170b718e... 100% ▕██████████████████████████████████████████████ 11 KB
pulling f02dd72bb242... 100% ▕██████████████████████████████████████████████▏ 59 B
pulling 648f809ced2b... 100% ▕██████████████████████████████████████████████▏ 485 B
verifying sha256 digest
writing manifest
removing any unused layers
success

hello
Hello! How can I assist you today?

(base) igor@macigor-2 ~ % ollama run deepseek-v2
pulling manifest
pulling d8d69f2a1bfa... 100% ▕███████████████████████████████████████████████▏ 8.9 GB
pulling 732caedf08d1... 100% ▕███████████████████████████████████████████████▏ 112 B
pulling 4bb71764481f... 100% ▕███████████████████████████████████████████████▏ 13 KB
pulling 19f2fb9e8bc6... 100% ▕███████████████████████████████████████████████▏ 32 B
pulling efe8e27d9609... 100% ▕███████████████████████████████████████████████▏ 494 B
verifying sha256 digest
writing manifest
removing any unused layers
success

hello
你好！很高兴见到你。有什么我可以帮你的吗？

can you answer in english?
当然可以，有什么问题需要我回答的？

<!-- gh-comment-id:2156216524 --> @igorschlum commented on GitHub (Jun 8, 2024): I tried with version 0.1.42 of Ollama on MacOS with 32GB Ram. with granite-code I get answers in english. But for deepseek-v2, I can reproduce also, it's answering in Chiness even if I ask to answer in English. For qwen2:7b, the answer is in English (base) igor@macigor-2 ~ % ollama run granite-code pulling manifest pulling 02ab8cd2f514... 100% ▕███████████████████████████████████████████████▏ 2.0 GB pulling 0d7c97d535b6... 100% ▕███████████████████████████████████████████████▏ 26 B pulling e50df8490144... 100% ▕███████████████████████████████████████████████▏ 123 B pulling 9893bb2c2917... 100% ▕███████████████████████████████████████████████▏ 108 B pulling 22b176fd8ef6... 100% ▕███████████████████████████████████████████████▏ 485 B verifying sha256 digest writing manifest removing any unused layers success >>> hello Hello! How can I assist you today? (base) igor@macigor-2 ~ % ollama run qwen2:7b pulling manifest pulling 43f7a214e532... 100% ▕██████████████████████████████████████████████ 4.4 GB pulling 62fbfd9ed093... 100% ▕██████████████████████████████████████████████▏ 182 B pulling c156170b718e... 100% ▕██████████████████████████████████████████████ 11 KB pulling f02dd72bb242... 100% ▕██████████████████████████████████████████████▏ 59 B pulling 648f809ced2b... 100% ▕██████████████████████████████████████████████▏ 485 B verifying sha256 digest writing manifest removing any unused layers success >>> hello Hello! How can I assist you today? (base) igor@macigor-2 ~ % ollama run deepseek-v2 pulling manifest pulling d8d69f2a1bfa... 100% ▕███████████████████████████████████████████████▏ 8.9 GB pulling 732caedf08d1... 100% ▕███████████████████████████████████████████████▏ 112 B pulling 4bb71764481f... 100% ▕███████████████████████████████████████████████▏ 13 KB pulling 19f2fb9e8bc6... 100% ▕███████████████████████████████████████████████▏ 32 B pulling efe8e27d9609... 100% ▕███████████████████████████████████████████████▏ 494 B verifying sha256 digest writing manifest removing any unused layers success >>> hello 你好！很高兴见到你。有什么我可以帮你的吗？ >>> can you answer in english? 当然可以，有什么问题需要我回答的？
GiteaMirror commented 2026-04-22 07:25:33 -05:00
Author
Owner
Copy Link

@Axenide commented on GitHub (Jun 8, 2024):

I just tested with 0.1.42 and at least Qwen 2 seems to work fine. Sometimes it mixes languages but that's from the LLM itself.

I tried deepseek-v2 several times but it always gives me this error:
Error: llama runner process has terminated: signal: aborted (core dumped)
I guess this has to do with my system resources, I have only 6 GB of VRAM, but I thought it could offload to CPU and RAM.

<!-- gh-comment-id:2156219764 --> @Axenide commented on GitHub (Jun 8, 2024): I just tested with 0.1.42 and at least Qwen 2 seems to work fine. Sometimes it mixes languages but that's from the LLM itself. I tried deepseek-v2 several times but it always gives me this error: `Error: llama runner process has terminated: signal: aborted (core dumped)` I guess this has to do with my system resources, I have only 6 GB of VRAM, but I thought it could offload to CPU and RAM.
GiteaMirror commented 2026-04-22 07:25:34 -05:00
Author
Owner
Copy Link

@jmorganca commented on GitHub (Jun 9, 2024):

Hi folks, Qwen 2 is fixed now - sorry about that.

@Axenide are you using OLLAMA_FLASH_ATTENTION=1? deepseek-v2 doesn't work with flash attention yet. Let me know if that's not the case and I can re-open this. Sorry you hit an error

<!-- gh-comment-id:2156707546 --> @jmorganca commented on GitHub (Jun 9, 2024): Hi folks, Qwen 2 is fixed now - sorry about that. @Axenide are you using `OLLAMA_FLASH_ATTENTION=1`? `deepseek-v2` doesn't work with flash attention yet. Let me know if that's not the case and I can re-open this. Sorry you hit an error
GiteaMirror commented 2026-04-22 07:25:34 -05:00
Author
Owner
Copy Link

@Axenide commented on GitHub (Jun 9, 2024):

Hi folks, Qwen 2 is fixed now - sorry about that.

@Axenide are you using OLLAMA_FLASH_ATTENTION=1? deepseek-v2 doesn't work with flash attention yet. Let me know if that's not the case and I can re-open this. Sorry you hit an error

Hi, no worries. No, I'm not using flash attention. Just the defaults.

<!-- gh-comment-id:2156717411 --> @Axenide commented on GitHub (Jun 9, 2024): > Hi folks, Qwen 2 is fixed now - sorry about that. > > @Axenide are you using `OLLAMA_FLASH_ATTENTION=1`? `deepseek-v2` doesn't work with flash attention yet. Let me know if that's not the case and I can re-open this. Sorry you hit an error Hi, no worries. No, I'm not using flash attention. Just the defaults.
GiteaMirror commented 2026-04-22 07:25:34 -05:00
Author
Owner
Copy Link

@igorschlum commented on GitHub (Jun 9, 2024):

@jmorganca Qwen2 is fixed but deepseek-v2 is not. It's always answering in chines.

<!-- gh-comment-id:2156722218 --> @igorschlum commented on GitHub (Jun 9, 2024): @jmorganca Qwen2 is fixed but deepseek-v2 is not. It's always answering in chines.
GiteaMirror commented 2026-04-22 07:25:35 -05:00
Author
Owner
Copy Link

@jmorganca commented on GitHub (Jun 9, 2024):

@igorschlum I think may be how the model was trained!

@Axenide Shoot, sorry then – may I ask what hardware you are running it on?

<!-- gh-comment-id:2156723160 --> @jmorganca commented on GitHub (Jun 9, 2024): @igorschlum I think may be how the model was trained! @Axenide Shoot, sorry then – may I ask what hardware you are running it on?
GiteaMirror commented 2026-04-22 07:25:35 -05:00
Author
Owner
Copy Link

@Axenide commented on GitHub (Jun 9, 2024):

@Axenide Shoot, sorry then – may I ask what hardware you are running it on?

Sure, here are my specs:

CPU: AMD Ryzen 5 2600 (12) @ 3,40 GHz
GPU: NVIDIA GeForce RTX 2060 (6 GB VRAM)
RAM: 16 GB

I'm using Arch Linux with kernel 6.9.3-arch1-1

<!-- gh-comment-id:2156735760 --> @Axenide commented on GitHub (Jun 9, 2024): > @Axenide Shoot, sorry then – may I ask what hardware you are running it on? Sure, here are my specs: CPU: AMD Ryzen 5 2600 (12) @ 3,40 GHz GPU: NVIDIA GeForce RTX 2060 (6 GB VRAM) RAM: 16 GB I'm using Arch Linux with kernel 6.9.3-arch1-1
GiteaMirror commented 2026-04-22 07:25:35 -05:00
Author
Owner
Copy Link

@igorschlum commented on GitHub (Jun 10, 2024):

@jmorganca In Ollama Deepseek answers in Chiness :
Deepseekv2 in Ollama
In chat.deepseek.com it answers in English :
Deepseekv2 in chat deepseek com

<!-- gh-comment-id:2157447496 --> @igorschlum commented on GitHub (Jun 10, 2024): @jmorganca In Ollama Deepseek answers in Chiness : <img width="511" alt="Deepseekv2 in Ollama" src="https://github.com/ollama/ollama/assets/2884312/464cd03c-a353-45ec-9d52-0e2c054170bd"> In [chat.deepseek.com](https://chat.deepseek.com) it answers in English : <img width="632" alt="Deepseekv2 in chat deepseek com" src="https://github.com/ollama/ollama/assets/2884312/9ae5e5fb-015a-4a5f-80e2-16749d03b0c8">
GiteaMirror commented 2026-04-22 07:25:35 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jun 15, 2024):

@Axenide Your problem may be that ollama is trying to load too many layers for deepseek-v2, see https://github.com/ollama/ollama/issues/4799#issuecomment-2148573008

<!-- gh-comment-id:2170944038 --> @rick-github commented on GitHub (Jun 15, 2024): @Axenide Your problem may be that ollama is trying to load too many layers for deepseek-v2, see https://github.com/ollama/ollama/issues/4799#issuecomment-2148573008
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#28866
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?