Mirascope is a Python library that lets you access a range of Large Language Models, but in a more straightforward and Pythonic way.

If you've been used to Langchain to call into various model providers such as OpenAI, Anthropic, Mistral etc you'll find that this library does the same but more fluenty and with less overhead. The notion is that working with LLMs should be intuitive and simple just like coding with other Python projects.

Mirascope unites the core modules of accessing each distinct LLM provider under a common API but at the same time gives you full access to every nitty-gritty detail of them when you need to; that is you can use the Mirascope wrapper API but at any point in your workflow you can get access to the original provider classes.

As an API, Mirascope's classes encapsulate everything which can impact the results of using that class, such as
the prompt template, the temperature, the model, and other call parameters, since when extracting structured information, the schema to be extracted should be colocated with the call for extracting it and the prompt template for that extraction.

Let's see an example. You can initialize an OpenAICall instance and invoke its call method to generate an OpenAICallResponse:

You can also Chain Calls, do Retries when LLM providers fail for various reasons, have stateful Chat History, Function calling etc.

The following providers are for the time being supported:

• OpenAI
• Anthropic
• Mistral
• Cohere
• Groq
• Gemini

As said, Mirascope's unified interface makes it fast and simple to swap between various providers. For instance here's how easy is the swapping, first with OpenAI:

and then with Mistral:

So there you have it; access LLMs in a fluent less bloated and Pythonic way.

More Information

Mirascope 

Related Articles

IBM Launches The Granite Code LLM Series

Get Started With Ollama's Python & Javascript Libraries

To be informed about new articles on I Programmer, sign up for our weekly newsletter, subscribe to the RSS feed and follow us on Twitter, Facebook or Linkedin.

Comments

Make a Comment or View Existing Comments Using Disqus

or email your comment to: comments@i-programmer.info