github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 08:30:05 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #1948] Understanding Response Data Structure #1122

New Issue
Closed
opened 2026-04-12 10:52:04 -05:00 by GiteaMirror · 3 comments
GiteaMirror commented 2026-04-12 10:52:04 -05:00
Owner
Copy Link

Originally created by @tmattoneill on GitHub (Jan 12, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/1948

I'm really confused by Ollama's response from the API. Most other LLM's I've used return a consistent model / JSON object that can serve as the 'assistant' response. However, Ollama returns a different, seemingly random JSON / object every time. This makes it nearly impossible to extract the reply from any prompt. See below:

generate_response("Hello world")
Hello world
{'dialogue': {'bot': 'Hello! How can I help you today?', 'user': 'Hello world'}}
generate_response("Hello world")
Hello world
{'outputText': 'Hello, World!\n'}
generate_response("Hello world")
Hello world
{'message': 'Hello! How can I assist you today?'}

The code generating this is:

    HOST = "localhost"
    PORT = "11434"
    api_request = {
        "model": "mistral",
        "stream": False,
        "raw": True,
        "format": "json",
        "prompt": f"[INST]{prompt}[/INST]"
    }

    try:
        response = requests.post(f"http://{HOST}:{PORT}/api/generate", json=api_request)
        response.raise_for_status()
        message = json.loads(response.text)['response']
        response = json.loads(message)
    except requests.exceptions.RequestException as e:
        raise ValueError("Error making API request") from e
    except json.JSONDecodeError as e:
        raise ValueError("Error parsing API response") from e

Can someone explain this to me? I've been through the docs extensively and can not for the life of me figure out how to do this pretty straightforward task.

Originally created by @tmattoneill on GitHub (Jan 12, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/1948 I'm really confused by Ollama's response from the API. Most other LLM's I've used return a consistent model / JSON object that can serve as the 'assistant' response. However, Ollama returns a different, seemingly random JSON / object every time. This makes it nearly impossible to extract the reply from any prompt. See below: ``` generate_response("Hello world") Hello world {'dialogue': {'bot': 'Hello! How can I help you today?', 'user': 'Hello world'}} generate_response("Hello world") Hello world {'outputText': 'Hello, World!\n'} generate_response("Hello world") Hello world {'message': 'Hello! How can I assist you today?'} ``` The code generating this is: ``` HOST = "localhost" PORT = "11434" api_request = { "model": "mistral", "stream": False, "raw": True, "format": "json", "prompt": f"[INST]{prompt}[/INST]" } try: response = requests.post(f"http://{HOST}:{PORT}/api/generate", json=api_request) response.raise_for_status() message = json.loads(response.text)['response'] response = json.loads(message) except requests.exceptions.RequestException as e: raise ValueError("Error making API request") from e except json.JSONDecodeError as e: raise ValueError("Error parsing API response") from e ``` Can someone explain this to me? I've been through the docs extensively and can not for the life of me figure out how to do this pretty straightforward task.
GiteaMirror commented 2026-04-12 10:52:05 -05:00
Author
Owner
Copy Link

@mxyng commented on GitHub (Jan 13, 2024):

I think there's some confusion here. What you're experiencing is the LLM responding in JSON, as requested by your Python script "format": "json". It looks like you've already figured out the structure of the response json.loads(response.text)['response']. The message you're returning (json.loads(message)) is the output from the LLM.

If you unset format you will notice the response cannot be JSON deserialized. That's because the response from the LLM is no longer valid JSON but rather plain text

<!-- gh-comment-id:1890227009 --> @mxyng commented on GitHub (Jan 13, 2024): I think there's some confusion here. What you're experiencing is the LLM responding in JSON, as requested by your Python script `"format": "json"`. It looks like you've already figured out the structure of the response `json.loads(response.text)['response']`. The message you're returning (`json.loads(message)`) is the output from the LLM. If you unset `format` you will notice the response cannot be JSON deserialized. That's because the response from the LLM is no longer valid JSON but rather plain text
GiteaMirror commented 2026-04-12 10:52:05 -05:00
Author
Owner
Copy Link

@tmattoneill commented on GitHub (Jan 13, 2024):

Thank you @mxyng I appreciate it. Yes, I can get the payload of the 'response' but the issue is that the contents of the response are different every time so I can't reliably extract the contents of that response. As you can see in my examples at the top, each has a different structure.

Is there a best-practice to get these results? If, say, I was building a chat-bot how could I use that response?

<!-- gh-comment-id:1890374598 --> @tmattoneill commented on GitHub (Jan 13, 2024): Thank you @mxyng I appreciate it. Yes, I can get the payload of the 'response' but the issue is that the contents of the response are different every time so I can't reliably extract the contents of that response. As you can see in my examples at the top, each has a different structure. Is there a best-practice to get these results? If, say, I was building a chat-bot how could I use that response?
GiteaMirror commented 2026-04-12 10:52:06 -05:00
Author
Owner
Copy Link

@MarsThunder commented on GitHub (Oct 1, 2025):

This is over a year later, so useful to others. I would imagine what you want is to inject a 'seed' number into your query. That would lock the response into the same response each time (at least it should).
response_chat = ollama.chat(model, messages=[], options={'seed': 120})

<!-- gh-comment-id:3356711195 --> @MarsThunder commented on GitHub (Oct 1, 2025): This is over a year later, so useful to others. I would imagine what you want is to inject a 'seed' number into your query. That would lock the response into the same response each time (at least it should). `response_chat = ollama.chat(model, messages=[], options={'seed': 120})`
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#1122
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?