Show HN: Skeights – Serialize sklearn models to safetensors and JSON, no pickle (github.com)

A new tool named "skeights" has been introduced to revolutionize the way scikit-learn models are serialized, allowing users to save models as safetensors and JSON files instead of relying on traditional methods like pickle or joblib. This innovation addresses critical issues associated with pickle, including security risks from arbitrary code execution, fragility across different Python and scikit-learn versions, and opacity that prevents easy inspection of model content. With skeights, model hyperparameters and states are stored in human-readable JSON format, while weights are saved in a safe and efficient safetensors format. The significance of skeights lies in its ability to enhance model safety, transparency, and compatibility, which are essential factors for the AI/ML community, especially as models become increasingly complex. By separating model configuration from weight data, users can easily inspect, compare, and manage their machine learning models. Skeights supports various model types, including Ridge regression, decision trees, and gradient boosting, and is compatible with scikit-learn versions 1.5 and above. This provides a forward-compatible solution for machine learning practitioners looking to maintain robust version control and foster collaboration in their AI projects.