PriorLabs/TabPFN v8.0.2

on GitHub

Added

• Add flash attention support for MPS to reduce memory usage. Remove attention_backend. (#949)

Changed

• Modernized the SHAP / Shapley Values section in TabPFN_Demo_Local.ipynb to use shapiq (with TabPFN's KV cache enabled), and made small fixes to the feature-selection, time-series, and causal-inference sections. (#960)