feat: Add support for ollama structured outputs

[vitreuz]

Title

Add support for ollama structured outputs

Relevant issues

Fixes #7131

Type

🆕 New Feature

Changes

• Added process_response_format function to convert OpenAI-style response format to Ollama API format.
• Updated OllamaConfig and OllamaChatConfig to use the new process_response_format function.
• Added unit test test_ollama_structured_format to validate the structured JSON schema format.

[REQUIRED] Testing - Attach a screenshot of any new tests passing locall

If UI changes, send a screenshot/GIF of working UI fixes

from pydantic import BaseModel

import litellm


class Country(BaseModel):
    name: str
    capital: str
    languages: list[str]


messages = [{"role": "user", "content": "Tell me about Canada."}]
print("Sending the following messages to the model:", messages)

response = litellm.completion(
    "ollama_chat/mistral-nemo", messages, response_format=Country
)

print(response.choices[0].message)

output:

❯ python test.py
Sending the following messages to the model: [{'role': 'user', 'content': 'Tell me about Canada.'}]
Message(content='{ "name": "Canada", "capital": "Ottawa", "languages": ["English", "French", "Aboriginal languages"] }', role='assistant', tool_calls=None, function_call=None)

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
litellm	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Dec 21, 2024 7:07am

[vitreuz]

Added an updated to revert to old behavior when type is neither json_schema or json