MinimalLLM 1.0.0

Minimal toolkit for LLM

MinimalLLM

Minimal and handy interfaces of LLMs based on LiteLLM.

Features:

• Minimal and intuitive interfaces. You don't need to learn
• Handy cache system for faster testing
• Support for multiple LLMs by LiteLLM

Usage

Chat

from mllm import Chat
chat = Chat()
chat += "Give me a random number from 1 to 10"
res = chat.complete(cache=True)
print(res)
chat = Chat()
chat += "Give a json dict with keys 'a' and 'b' and values 1 and 2"
res = chat.complete(parse="dict", cache=True)
print(res)

Embedding

from mllm import get_embeddings
embeddings = get_embeddings(["Hello, world!", "Goodbye, world!"])
print(embeddings)
# Embeddings are automatically cached

Visualized Log of chats

from mllm import Chat, display_chats
with display_chats():
    chat = Chat()
    chat += "Who is your developer?"
    res = chat.complete(cache=True)
    # A web page will be opened to show the chat log

Providers

MinimalLLM is based on LiteLLM, which supports multiple providers. You can switch the provider to use different LLMs.

Switch provider

from mllm.provider_switch import set_default_to_anthropic
set_default_to_anthropic()

To setup the API keys for the providers, we recommend to use the following wizard:

from mllm.setup.wizard import setup_wizard
setup_wizard()

You can run in the command line:

python -m mllm.setup.wizard

Installation

pip install -U git+https://github.com/EvoEvolver/MinimalLLM

To add it to your requirements.txt:

MinimalLLM @ git+https://github.com/EvoEvolver/MinimalLLM

Why minimal?

Minimal means stable. We only provide the most basic interfaces and features.