Universal LLM Interface

This module provides wrapper classes for interacting with various Large Language Models (LLMs) in a unified way.

Installation

You can install this package directly from PyPI:

pip install unillm

To install from source:

git clone https://github.com/BrachioLab/universal_llm.git
cd universal_llm
pip install -e .

Classes

OurLLM

A wrapper for local LLM models, particularly those from Hugging Face like Llama and Qwen.

Features:

• Supports Llama-3.2, Llama-3.3, and Qwen models.
• Handles both text and image inputs.
• Provides consistent interface for model inference.

Usage:

from src.llm_models import OurLLM

# Initialize the model
model = OurLLM(model_name="meta-llama/Llama-3.2-90B-Vision-Instruct")

# Use the model
response = model.chat(prompt, sampling_params, use_tqdm=False)

APIModel

A wrapper for API-based LLM services like Claude, Gemini, and GPT.

Features:

• Supports Claude, Gemini, and GPT models.
• Handles API authentication and retries.
• Provides consistent interface for model inference.

Usage:

from src.llm_models import APIModel

# Initialize the model
model = APIModel(model_name="claude-3-opus-20240229")

# Use the model
response = model.chat(prompt, sampling_params, use_tqdm=False)

Common Interface

Both classes implement a common interface with the following methods:

• __init__(model_name): Initialize the model with the specified model name
• chat(prompt, sampling_params, use_tqdm): Generate a response from the model

Dependencies

• transformers
• torch
• openai
• boto3
• anthropic
• python-dotenv
• google-genai
• pillow

Building and Distribution

To build the package:

pip install build
python -m build

This will create both source distribution (.tar.gz) and wheel (.whl) packages in the dist/ directory.

To upload to PyPI (requires PyPI credentials):

pip install twine
twine upload dist/*

Notes

• API keys should be configured securely, not hardcoded in the source code. Please create a .env directory which contains your API keys.
• The module is designed to be used with the vLLM framework for local models.