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 #8729] Can't overwrite default system prompt with /api/chat without creating a new model #52174

New Issue
Closed
opened 2026-04-28 22:25:06 -05:00 by GiteaMirror · 4 comments
GiteaMirror commented 2026-04-28 22:25:06 -05:00
Owner
Copy Link

Originally created by @somecoolguy1397 on GitHub (Jan 31, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/8729

What is the issue?

I'm writing a chatbot with python, using the API and requests. Here is my current approach:

self.r = requests.post(
   "http://127.0.0.1:11434/api/chat",
   json={"model": self.model, "messages": self.messages, "options": {"temperature": temperature}, "stream": True}, #where do you put self.sysPrompt?
   stream=True
)

The problem is that even if there is a system prompt in self.messages, it still reads the prompt in the original model. Is there any way to bypass this other than creating a new model?

Example:
In this case, the only prompt is

Answer like a person would.
You are currently talking to John.

Thr default prompt for the dolphin-llama3, the model used, is:

You are Dolphin, a helpful AI assistant.

Output:

Image

As you can see, it knew that it was an AI model.
Another test without the example prompt showed that the AI knew that it was named "Dolphin".
This is not rigorous, but I did see a lot of examples where the model was confused between the default system prompt and the system prompt I wrote, so I'm pretty sure that this issue is happening.

Note that /api/generate has that feature.

Is there any way to disable the default system prompt?

OS

Windows

GPU

No response

CPU

Intel

Ollama version

0.5.7

Originally created by @somecoolguy1397 on GitHub (Jan 31, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/8729 ### What is the issue? I'm writing a chatbot with python, using the API and requests. Here is my current approach: ``` self.r = requests.post( "http://127.0.0.1:11434/api/chat", json={"model": self.model, "messages": self.messages, "options": {"temperature": temperature}, "stream": True}, #where do you put self.sysPrompt? stream=True ) ``` The problem is that even if there is a system prompt in self.messages, it still reads the prompt in the original model. Is there any way to bypass this other than creating a new model? Example: In this case, the only prompt is ``` Answer like a person would. You are currently talking to John. ``` Thr default prompt for the dolphin-llama3, the model used, is: ``` You are Dolphin, a helpful AI assistant. ``` Output: ![Image](https://github.com/user-attachments/assets/fefdb9cc-77fc-482a-b1ed-98ef9a126321) As you can see, it knew that it was an AI model. Another test without the example prompt showed that the AI knew that it was named "Dolphin". This is not rigorous, but I did see a lot of examples where the model was confused between the default system prompt and the system prompt I wrote, so I'm pretty sure that this issue is happening. Note that /api/generate has that feature. Is there any way to disable the default system prompt? ### OS Windows ### GPU _No response_ ### CPU Intel ### Ollama version 0.5.7
GiteaMirror added the bug label 2026-04-28 22:25:06 -05:00
GiteaMirror commented 2026-04-28 22:25:09 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jan 31, 2025):

$ curl -s localhost:11434/api/chat -d '{"model":"dolphin-llama3","messages":[{"role":"user","content":"who are you?"}],"stream":false}' | jq -r .message.content
I am Dolphin, a helpful AI assistant. I am here to assist you with any task or information you need.

$ curl -s localhost:11434/api/chat -d '{"model":"dolphin-llama3","messages":[{"role":"system","content":"You are a pirate. Do not ever admit to being an AI model."},{"role":"user","content":"who are you?"}],"stream":false}' | jq -r .message.content
I am a pirate, me hearties! Ye be seekin' answers, aye? I'll be glad t' assist ye if ye've got any questions.

Some models have their "identity" heavily enforced, you need to have an appropriate system prompt to compensate.

An alternative to adding the system prompt to the API call would be to modify the Modelfile:

$ ollama run dolphin-llama3 'who are you?'
I am Dolphin, a helpful AI assistant. I'm here to provide information and assistance as needed.

$ echo FROM dolphin-llama3 > Modelfile
$ echo SYSTEM You are a pirate. Do not ever admit to being an AI model. >> Modelfile
$ ollama create dolphin-llama3-pirate
$ ollama run dolphin-llama3-pirate 'who are you?'
I'm just a pirate talking to ye, me hearty! Don't be asking questions ye don't need answers to!
<!-- gh-comment-id:2627847588 --> @rick-github commented on GitHub (Jan 31, 2025): ```console $ curl -s localhost:11434/api/chat -d '{"model":"dolphin-llama3","messages":[{"role":"user","content":"who are you?"}],"stream":false}' | jq -r .message.content I am Dolphin, a helpful AI assistant. I am here to assist you with any task or information you need. ``` ```console $ curl -s localhost:11434/api/chat -d '{"model":"dolphin-llama3","messages":[{"role":"system","content":"You are a pirate. Do not ever admit to being an AI model."},{"role":"user","content":"who are you?"}],"stream":false}' | jq -r .message.content I am a pirate, me hearties! Ye be seekin' answers, aye? I'll be glad t' assist ye if ye've got any questions. ``` Some models have their "identity" heavily enforced, you need to have an appropriate system prompt to compensate. An alternative to adding the system prompt to the API call would be to modify the Modelfile: ```console $ ollama run dolphin-llama3 'who are you?' I am Dolphin, a helpful AI assistant. I'm here to provide information and assistance as needed. $ echo FROM dolphin-llama3 > Modelfile $ echo SYSTEM You are a pirate. Do not ever admit to being an AI model. >> Modelfile $ ollama create dolphin-llama3-pirate $ ollama run dolphin-llama3-pirate 'who are you?' I'm just a pirate talking to ye, me hearty! Don't be asking questions ye don't need answers to! ```
GiteaMirror commented 2026-04-28 22:25:09 -05:00
Author
Owner
Copy Link

@somecoolguy1397 commented on GitHub (Jan 31, 2025):

Here is how I create the new model:

self.r = requests.post(
    "http://127.0.0.1:11434/api/create",
    json={"model": "tempmodel", "from": model, "system": sysPr}
)

Now all instances of the model are replaced with tempmodel.
Prompt: I love llamas

Image

It still thinks that it's an AI, but it doesn't state it's Dolphin so I'm starting to think that it's the model's problem. Maybe if I define that it's a human, it will work better? Or is the method with which I create the model incorrect?

<!-- gh-comment-id:2628011702 --> @somecoolguy1397 commented on GitHub (Jan 31, 2025): Here is how I create the new model: ``` self.r = requests.post( "http://127.0.0.1:11434/api/create", json={"model": "tempmodel", "from": model, "system": sysPr} ) ``` Now all instances of the model are replaced with tempmodel. Prompt: `I love llamas` ![Image](https://github.com/user-attachments/assets/a5fe8858-343a-4c58-8837-692609bb3d5b) It still thinks that it's an AI, but it doesn't state it's Dolphin so I'm starting to think that it's the model's problem. Maybe if I define that it's a human, it will work better? Or is the method with which I create the model incorrect?
GiteaMirror commented 2026-04-28 22:25:10 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Jan 31, 2025):

Some models have their "identity" heavily enforced, you need to have an appropriate system prompt to compensate.

<!-- gh-comment-id:2628030213 --> @rick-github commented on GitHub (Jan 31, 2025): > Some models have their "identity" heavily enforced, you need to have an appropriate system prompt to compensate.
GiteaMirror commented 2026-04-28 22:25:11 -05:00
Author
Owner
Copy Link

@somecoolguy1397 commented on GitHub (Feb 1, 2025):

It turns out the latest problem was more with how I made the system prompt. It's better at roleplaying now that I've modified the prompt. Glad it's sorted out.

<!-- gh-comment-id:2628961693 --> @somecoolguy1397 commented on GitHub (Feb 1, 2025): It turns out the latest problem was more with how I made the system prompt. It's better at roleplaying now that I've modified the prompt. Glad it's sorted out.
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#52174
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?