Show HN: TabPFN Scaling Mode – Tabular Foundation Model on millions of rows (priorlabs.ai)

TabPFN has introduced Scaling Mode, a groundbreaking update that allows its foundation model for tabular data to handle datasets of up to 10 million rows. This development builds on the progress made since the initial launch of TabPFN, which began with a focus on small datasets. Initially limited to 1,000 rows, the newer iterations, including TabPFN-2.5, successfully processed up to 100,000 rows. The latest Scaling Mode effectively removes fixed upper limits on dataset sizes, enabling scalability only constrained by computational resources. The significance of Scaling Mode lies in its potential to revolutionize tabular machine learning by significantly enhancing data processing capabilities while maintaining high performance. The model supports 2,000 features, ensuring a seamless transition for users dealing with large datasets, as it allows for strong predictions in a single forward pass without the need for extensive hyperparameter tuning or task-specific training. Internal benchmarks indicate that Scaling Mode enhances performance across various applications and continues to improve as dataset sizes increase, distinguishing itself from traditional gradient boosting libraries. This advancement not only broadens the applicability of TabPFN but also solidifies its role in tackling real-world ML challenges in diverse industries.