nvidia/Kumo-Tabular

NVIDIA developed a pretrained machine learning model designed for table-form datasets. The system handles both classification and regression tasks across structured information. It operates using context-based prediction methods rather than traditional manual tuning. The release includes software weights that run on specific graphics hardware. Users install a designated Python package to enable local inference capabilities. The process involves loading labeled examples to establish a context window for analysis. New data entries are then compared against this context to generate class probability scores. The code example in the source demonstrates using a breast cancer dataset for this workflow. The provided documentation lacks specific performance metrics for real-world accuracy assessment. Independent validation results are not included in the model card text. The license identifies the weights as released under OpenMDW 1.1 terms. Current statistics show zero downloads and sixty-nine likes at the time of discovery.