MLLM is a library for simplifying communication with various Large Language Models (LLMs) and multi-modal LLMs such as OpenAI GPT, Anthropic Claude and Google Gemini. It lets you
  • Create a router to communicate with multiple models (LLMs/MLLMs)
  • Configure your preference order for the models
  • Communicate with the models by sending your message thread to the models’ completion endpoints
  • Retry communication upon failure
  • Enforce expected response formats
MLLM uses LiteLLM under the hood.

​
Installation

pip install mllm

​
Basic Example

Provide the API keys for the different models (LLMs / MLLMs) that you would like to use: 
import os

os.environ["OPENAI_API_KEY"] = "..."
os.environ["ANTHROPIC_API_KEY"] = "..."
os.environ["GEMINI_API_KEY"] = "..."
Create a Router object that can access all the desired models: 
from mllm import Router

model_preferences = ["gpt-4-turbo", "anthropic/claude-3-opus-20240229", "gemini/gemini-1.5-pro-latest"]

router = Router(preference=preferences)
Create a message thread using ThreadMem. MLLM also exposes RoleMessage and RoleThread from Threadmem. So, you can also create the thread using MLLM: 
from mllm import RoleThread

thread = RoleThread(owner_id="owner_id")
thread.post(role="user", msg="What do you see on this picture?", images=["https://upload.wikimedia.org/wikipedia/commons/c/c7/Tabby_cat_with_blue_eyes-3336579.jpg"])
Now, use the router to chat with the models using a message thread. 
response = router.chat(thread)
The router will communicate with your most preferred model first. It ensures that less preferred models are used only if the first one fails to respond.

​
Retry communication with model

The router retires communication upon failure, and you can configure the maximum number of retries allowed. 
response = router.chat(thread, retries=3)

​
Enforce response format

To enforce an expected response format, first create a class describing the response format: 
from pydantic import BaseModel

class Animal(BaseModel):
    species: str
    color: str
Now, use the expect parameter in the router.chat() call: 
response = router.chat(thread, expect=Animal)

animal_parsed = response.parsed

assert type(animal_parsed) == Animal
This is handy when you’re working with multiple models and want to ensure consistent outputs.