Google's TabFM skips per-dataset training and still predicts on unseen tables, matching tuned baselines and cutting pipeline ...
The new method fills the gap between GET and POST, combining the body of POST with the safety and idempotence of GET.