← Back to feed

TabPFN

GitHub Repo Pretty sure · synthetic training, real results
https://github.com/PriorLabs/TabPFN

Neural network that memorizes synthetic tabular data, ships as a sklearn-compatible API, actually outperforms tree ensembles on real benchmarks. The skepticism tax is lower than usual.

15%
30%
55%
Slop 15%Signal 30%Science 55%

TabPFN is a prior function network (transformer) trained on synthetic tabular data distributions. The science: genuine contribution to meta-learning for tabular problems, published research, claims grounded in benchmarks against XGBoost/RF. The signal: sklearn interface works, inference is fast with GPU, ecosystem around it (extensions) suggests actual use. The slop: marketing leans heavy on the 'foundation model' framing (it's not), the README is a wall of badges and flowcharts that could co...

6403 stars Python 2026-05-05 1404 days old

Become a MFer to rate — log in