PriorLabs/TabPFN v9.2.0

on GitHub

Added

• Add an opt-in enable_torch_compile setting to classifiers and regressors, with selective compilation of v3.5 aggregation regions on PyTorch 2.6 or newer. (#1331)

Fixed

• Fix incorrect predictions and repeated cache dequantization for batched TabPFN-3 and TabPFN-3.5 cached inference with memory saving enabled. (#1365)