github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 08:02:14 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #10001] Improve compatibility with OpenAI structured outputs json_schema response format #6556

New Issue
Closed
opened 2026-04-12 18:10:59 -05:00 by GiteaMirror · 2 comments
GiteaMirror commented 2026-04-12 18:10:59 -05:00
Owner
Copy Link

Originally created by @SuperPat45 on GitHub (Mar 26, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/10001

What is the issue?

Can you improve the /chat/completions compatibility with OpenAI about the json_schema response_format?

I tested with the gemma3 model, and this OpenAI syntax is ignored:

    "response_format": {,
        "type": "json_schema",
        "json_schema": {
            "name": "result",
            "schema": {
                "additionalProperties": false,,
                "type": "object",
                "required": [
                    "parm1"
                ],
                "properties": {
                    "parm1": {
....

Whereas Ollama only need a simpler format parameter:

            "format": {
                "type": "object",
                "required": [
                    "parm1"
                ],
                "properties": {
                    "parm1": {
....

It should be easy to modify the request coming from /chat/completions before relaying to the /api/chat.

Relevant log output

OS

Linux

GPU

Nvidia

CPU

AMD

Ollama version

0.6.2

Originally created by @SuperPat45 on GitHub (Mar 26, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/10001 ### What is the issue? Can you improve the /chat/completions compatibility with OpenAI about the json_schema response_format? I tested with the gemma3 model, and this OpenAI syntax is ignored: ``` "response_format": {, "type": "json_schema", "json_schema": { "name": "result", "schema": { "additionalProperties": false,, "type": "object", "required": [ "parm1" ], "properties": { "parm1": { .... ``` Whereas Ollama only need a simpler format parameter: ``` "format": { "type": "object", "required": [ "parm1" ], "properties": { "parm1": { .... ``` It should be easy to modify the request coming from /chat/completions before relaying to the /api/chat. ### Relevant log output ```shell ``` ### OS Linux ### GPU Nvidia ### CPU AMD ### Ollama version 0.6.2
GiteaMirror added the bug label 2026-04-12 18:10:59 -05:00
GiteaMirror commented 2026-04-12 18:11:01 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Mar 27, 2025):

Can you provide more information about how the schemas are created and passed to ollama? I created a little script to test schema passing and the wire traffic looks like the fragments you have posted. Both schemas return results.

#!/usr/bin/env python3

from pydantic import BaseModel
from openai import OpenAI
from ollama import Client

ollama_url = "http://localhost:11434"
model = "gemma3:4b"

class result(BaseModel):
    parm1: str

messages=[
    {"role": "system", "content": "Extract the information."},
    {"role": "user", "content": "parmeter 1 is 'hello'."},
]

def openai():
  client = OpenAI(api_key="key", base_url=f"{ollama_url}/v1")
  completion = client.beta.chat.completions.parse(
      model=model,
      messages=messages,
      response_format=result,
  )

  r = completion.choices[0].message.parsed
  print(r)

def ollama():
  ollama_client = Client(host=ollama_url)
  response = ollama_client.chat(
      model=model,
      messages=messages,
      format=result.model_json_schema())
  r = result.model_validate_json(response.message.content)
  print(r)

openai()
ollama()

$ strace -e sendto -s 1024 ./10001.py  2>&1 1>/dev/tty | grep -v POST | sed -ne 's/^[^"]*"\(.*\)"[^"]*$/\1/p' | tr \\n \\0 | xargs -0i@ printf @'\n' | jq '{"schema_format": (.response_format // .format)}' 
parm1='hello'
parm1='hello'
{
  "schema_format": {
    "type": "json_schema",
    "json_schema": {
      "schema": {
        "properties": {
          "parm1": {
            "title": "Parm1",
            "type": "string"
          }
        },
        "required": [
          "parm1"
        ],
        "title": "result",
        "type": "object",
        "additionalProperties": false
      },
      "name": "result",
      "strict": true
    }
  }
}
{
  "schema_format": {
    "properties": {
      "parm1": {
        "title": "Parm1",
        "type": "string"
      }
    },
    "required": [
      "parm1"
    ],
    "title": "result",
    "type": "object"
  }
}
<!-- gh-comment-id:2756267970 --> @rick-github commented on GitHub (Mar 27, 2025): Can you provide more information about how the schemas are created and passed to ollama? I created a little script to test schema passing and the wire traffic looks like the fragments you have posted. Both schemas return results. ```python #!/usr/bin/env python3 from pydantic import BaseModel from openai import OpenAI from ollama import Client ollama_url = "http://localhost:11434" model = "gemma3:4b" class result(BaseModel): parm1: str messages=[ {"role": "system", "content": "Extract the information."}, {"role": "user", "content": "parmeter 1 is 'hello'."}, ] def openai(): client = OpenAI(api_key="key", base_url=f"{ollama_url}/v1") completion = client.beta.chat.completions.parse( model=model, messages=messages, response_format=result, ) r = completion.choices[0].message.parsed print(r) def ollama(): ollama_client = Client(host=ollama_url) response = ollama_client.chat( model=model, messages=messages, format=result.model_json_schema()) r = result.model_validate_json(response.message.content) print(r) openai() ollama() ``` ```console $ strace -e sendto -s 1024 ./10001.py 2>&1 1>/dev/tty | grep -v POST | sed -ne 's/^[^"]*"\(.*\)"[^"]*$/\1/p' | tr \\n \\0 | xargs -0i@ printf @'\n' | jq '{"schema_format": (.response_format // .format)}' parm1='hello' parm1='hello' { "schema_format": { "type": "json_schema", "json_schema": { "schema": { "properties": { "parm1": { "title": "Parm1", "type": "string" } }, "required": [ "parm1" ], "title": "result", "type": "object", "additionalProperties": false }, "name": "result", "strict": true } } } { "schema_format": { "properties": { "parm1": { "title": "Parm1", "type": "string" } }, "required": [ "parm1" ], "title": "result", "type": "object" } } ```
GiteaMirror commented 2026-04-12 18:11:01 -05:00
Author
Owner
Copy Link

@SuperPat45 commented on GitHub (Mar 27, 2025):

Sorry, The problem seems to be with Open WebUI

<!-- gh-comment-id:2757968273 --> @SuperPat45 commented on GitHub (Mar 27, 2025): Sorry, The problem seems to be with Open WebUI
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#6556
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?