github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-07-17 13:47:08 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #4789] deepseek-v2 responding in Chinese by default #3018

New Issue
Closed
opened 2026-04-12 13:25:27 -05:00 by GiteaMirror · 8 comments
GiteaMirror commented 2026-04-12 13:25:27 -05:00
Owner
Copy Link

Originally created by @rb81 on GitHub (Jun 2, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/4789

What is the issue?

deepseek-v2 seems to respond in Chinese only.

OS

Linux

GPU

Other

CPU

Intel

Ollama version

0.1.41

Originally created by @rb81 on GitHub (Jun 2, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/4789 ### What is the issue? deepseek-v2 seems to respond in Chinese only. ### OS Linux ### GPU Other ### CPU Intel ### Ollama version 0.1.41
GiteaMirror added the bug label 2026-04-12 13:25:27 -05:00
GiteaMirror commented 2026-04-12 13:25:28 -05:00
Author
Owner
Copy Link

@jmorganca commented on GitHub (Jun 2, 2024):

Hi @rb81 I believe this is how the lite version of the model was trained. I've gotten it to reply in English with system prompts. Will close this for now however do let me know if this isn't the case for the model elsewhere and I can revisit

<!-- gh-comment-id:2143980291 --> @jmorganca commented on GitHub (Jun 2, 2024): Hi @rb81 I believe this is how the lite version of the model was trained. I've gotten it to reply in English with system prompts. Will close this for now however do let me know if this isn't the case for the model elsewhere and I can revisit
GiteaMirror commented 2026-04-12 13:25:29 -05:00
Author
Owner
Copy Link

@rb81 commented on GitHub (Jun 2, 2024):

@jmorganca - Thanks for the reply. Would you mind sharing how you got it to work in English? I've tried a variety of system prompts but seem to always get Chinese...

<!-- gh-comment-id:2144008456 --> @rb81 commented on GitHub (Jun 2, 2024): @jmorganca - Thanks for the reply. Would you mind sharing how you got it to work in English? I've tried a variety of system prompts but seem to always get Chinese...
GiteaMirror commented 2026-04-12 13:25:30 -05:00
Author
Owner
Copy Link

@randoentity commented on GitHub (Jun 17, 2024):

I also only get Chinese with deepseek-v2:236b-chat-q4_0. I've tried a custom system prompt and the standard "Canada" example messages. @jmorganca help is greatly appreciated :)

Edit: my bad! I was missing a newline between the system prompt and the first message. If anyone wants an example still, here is what worked for me:

SYSTEM You are an English language assistant. You will only reply in EnglishMESSAGE assistant How can I help you today?
MESSAGE user Is Toronto in Canada?
MESSAGE assistant Yes it is.
MESSAGE user Is Sacramento in Canada?
MESSAGE assistant No it is not.
MESSAGE user Is Ontario in Canada?
MESSAGE assistant Yes, Ontario is in Canada too.

I'm not sure if a double post is warranted yet, so I'm editing this message again;
It seems like where the context is loaded makes a difference. With low context (up to 8k) and keeping space free in VRAM for the context by setting OLLAMA_MAX_VRAM (I assume num_gpu would accomplish the same thing) I get English. But when I fill up the VRAM with layers and the context (somehow) gets pushed to RAM, I get Chinese.
It kind of surprises me that context (apparently) goes to RAM, as before with just llama.cpp it would just OOM for me. So: buggy new feature or I'm being dumb.

<!-- gh-comment-id:2172420530 --> @randoentity commented on GitHub (Jun 17, 2024): I also only get Chinese with deepseek-v2:236b-chat-q4_0. I've tried a custom system prompt and the standard "Canada" example messages. @jmorganca help is greatly appreciated :) Edit: my bad! I was missing a newline between the system prompt and the first message. If anyone wants an example still, here is what worked for me: ``` SYSTEM You are an English language assistant. You will only reply in EnglishMESSAGE assistant How can I help you today? MESSAGE user Is Toronto in Canada? MESSAGE assistant Yes it is. MESSAGE user Is Sacramento in Canada? MESSAGE assistant No it is not. MESSAGE user Is Ontario in Canada? MESSAGE assistant Yes, Ontario is in Canada too. ``` I'm not sure if a double post is warranted yet, so I'm editing this message again; It seems like where the context is loaded makes a difference. With low context (up to 8k) and keeping space free in VRAM for the context by setting OLLAMA_MAX_VRAM (I assume num_gpu would accomplish the same thing) I get English. But when I fill up the VRAM with layers and the context (somehow) gets pushed to RAM, I get Chinese. It kind of surprises me that context (apparently) goes to RAM, as before with just llama.cpp it would just OOM for me. So: buggy new feature or I'm being dumb.
GiteaMirror commented 2026-04-12 13:25:31 -05:00
Author
Owner
Copy Link

@djdookie commented on GitHub (Jun 18, 2024):

For me the model answers english if ask in english if I just do ollama run "ollama run deepseek-coder-v2".
However if I use it on sourcecode in VSCode using the continue extension, then it always answers in chinese.
The deepseek-coder:6.7b-instruct model never did that. I have no idea how to solve that. The ollama template looks fine to me.

grafik

<!-- gh-comment-id:2177197702 --> @djdookie commented on GitHub (Jun 18, 2024): For me the model answers english if ask in english if I just do ollama run "ollama run deepseek-coder-v2". However if I use it on sourcecode in VSCode using the continue extension, then it always answers in chinese. The deepseek-coder:6.7b-instruct model never did that. I have no idea how to solve that. The ollama template looks fine to me. ![grafik](https://github.com/ollama/ollama/assets/17240204/6df74eed-df56-4049-ba7a-60fac88a99dd)
GiteaMirror commented 2026-04-12 13:25:31 -05:00
Author
Owner
Copy Link

@rb81 commented on GitHub (Jun 19, 2024):

@djdookie - I was getting Chinese all the time, then suddenly started getting only English now. I did pull the latest model, so perhaps something was fixed... Oddly enough, I noticed this when I started using Continue in VSCode as well...

<!-- gh-comment-id:2177788401 --> @rb81 commented on GitHub (Jun 19, 2024): @djdookie - I was getting Chinese all the time, then suddenly started getting only English now. I did pull the latest model, so perhaps something was fixed... Oddly enough, I noticed this when I started using Continue in VSCode as well...
GiteaMirror commented 2026-04-12 13:25:32 -05:00
Author
Owner
Copy Link

@DavSimFel commented on GitHub (Jun 19, 2024):

@djdookie - I have the same problem with continue.dev. Even with a system prompt it only works sometimes.

@jmorganca - I guess this is a a valid issue and should not be closed yet.

<!-- gh-comment-id:2178399011 --> @DavSimFel commented on GitHub (Jun 19, 2024): @djdookie - I have the same problem with continue.dev. Even with a system prompt it only works sometimes. @jmorganca - I guess this is a a valid issue and should not be closed yet.
GiteaMirror commented 2026-04-12 13:25:33 -05:00
Author
Owner
Copy Link

@jasalt commented on GitHub (Jun 20, 2024):

Quantization related issue it seems, discussed at https://github.com/deepseek-ai/DeepSeek-Coder-V2/issues/12. All the variants here https://ollama.com/library/deepseek-coder-v2/tags I could run on ~12GB VRAM seem to be affected more or less.

<!-- gh-comment-id:2180233409 --> @jasalt commented on GitHub (Jun 20, 2024): Quantization related issue it seems, discussed at https://github.com/deepseek-ai/DeepSeek-Coder-V2/issues/12. All the variants here https://ollama.com/library/deepseek-coder-v2/tags I could run on ~12GB VRAM seem to be affected more or less.
GiteaMirror commented 2026-04-12 13:25:34 -05:00
Author
Owner
Copy Link

@jmorganca commented on GitHub (Jun 21, 2024):

Hi folks, we had a small correction to the template in https://ollama.com/library/deepseek-coder-v2 and https://ollama.com/library/deepseek-v2 - it should be fixed now and by default will answer in the prompt's language. Sorry about that!

<!-- gh-comment-id:2181782903 --> @jmorganca commented on GitHub (Jun 21, 2024): Hi folks, we had a small correction to the template in https://ollama.com/library/deepseek-coder-v2 and https://ollama.com/library/deepseek-v2 - it should be fixed now and by default will answer in the prompt's language. Sorry about that!
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#3018
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?