ollama-instructor 1.1.0

Validate structured outputs from LLMs with Ollama and automatic retries.

ollama-instructor

ollama-instructor is a lightweight Python library that provides a convenient wrapper around the Ollama Client, extending it with validation features for obtaining valid JSON responses from Large Language Models (LLMs). Utilizing Pydantic, ollama-instructor ensures that responses from LLMs adhere to defined schemas.

Note: This library depends on having Ollama installed and running. For more information, please refer to the official website of Ollama.

Breaking Changes in Version 1.0.0

Version 1.0.0 introduces significant changes from version 0.5.2:

• Complete refactoring to directly inherit from Ollama's official Client classes
• Simplified API that aligns more closely with Ollama's native interface
• Improved logging system using Python's built-in logging module
• Streamlined validation process using Pydantic
• Removal of partial validation features to focus on core functionality
• New method names: chat_completion and chat_stream (previously chat_completion_with_stream)

Features

• Direct Integration: Inherits directly from Ollama's official client for seamless integration
• Schema Validation: Uses Pydantic BaseModel to ensure valid JSON responses
• Retry Mechanism: Automatically retries failed validations with configurable attempts
• Logging: Comprehensive logging system with configurable levels
• Async Support: Full async/await support through OllamaInstructorAsync

Installation

pip install ollama-instructor

Quick Start

For streaming examples click here

Synchronous Usage:

from pydantic import BaseModel
from ollama_instructor import OllamaInstructor

class FriendInfo(BaseModel):
    name: str
    age: int
    is_available: bool

class FriendList(BaseModel):
    friends: list[FriendInfo]

# Create client with logging enabled
client = OllamaInstructor(enable_logging=True, log_level='DEBUG')

# Chat completion
response = client.chat_completion(
    format=FriendList,
    model='llama2:latest',
    messages=[
        {
            'role': 'user',
            'content': 'I have two friends: John (25, available) and Mary (30, busy)'
        }
    ]
)

Asynchronous Usage:

import asyncio
from pydantic import BaseModel
from ollama_instructor import OllamaInstructorAsync

class FriendInfo(BaseModel):
    name: str
    age: int
    is_available: bool

async def main():
    client = OllamaInstructorAsync(enable_logging=True)

    response = await client.chat_completion(
        format=FriendInfo,
        model='llama2:latest',
        messages=[
            {
                'role': 'user',
                'content': 'John is 25 years old and available to hang out'
            }
        ]
    )

if __name__ == "__main__":
    asyncio.run(main())

Logging

The library includes comprehensive logging capabilities. You can enable and configure logging when initializing the client:

client = OllamaInstructor(
    enable_logging=True,
    log_level="DEBUG",  # Options: DEBUG, INFO, WARNING, ERROR, CRITICAL
    log_format="%(asctime)s - %(name)s - %(levelname)s - %(message)s"
)

Support and Community

If you need help or want to discuss ollama-instructor, feel free to:

• Open an issue on GitHub
• Start a discussion in the GitHub repository
• Contact via email: [e-mail]

Contributions and feedback are always welcome! 😊

License

ollama-instructor is released under the MIT License. See the LICENSE file for more details.