mllm-shap 0.1.9

Shap implementation for Multi Modal Large Language Models with audio and text on input.

Welcome to MLLM-SHAP

MLLM-SHAP is a Python package designed to interpret the predictions of large language models (LLMs) using SHAP (SHapley Additive exPlanations) values. It helps you understand the contribution of input features to model outputs, enabling transparent and explainable AI workflows. This work also has companion GUI visualization tools for easier interpretation of results, which is available at the official shap-mllm-explainer repository.

---

✨ Key Features

• Integration with audio and text models, supporting multi-modal inputs and outputs.
• Flexible aggregation strategies: mean, sum, max, min, etc.
• Multiple similarity metrics (cosine, euclidean, etc.) for embedding analysis.
• Customizable SHAP calculation algorithms: exact, Monte Carlo approximations, and more.
• Examples showcasing common explainability pipelines in examples/ on the official GitHub repository.

---

📊 Visualization & Examples

If you’re interested in GUI visualization of SHAP values, check out the section Extension - GUI Visualization in the docs.

For more advanced CLI usages, refer to:

• The official GitHub repository examples
• Or explore more advanced pipelines from exemplary research projects

---

🤖 Supported LLM Integrations

• Liquid-Audio