Google's TabFM: 400M-param tabular foundation model beats tuned AutoML zero-shot
google · hf · 2026-09-30
Google presents TabFM, a 400M-parameter foundation model for tabular data that frames supervised tabular prediction as in-context learning, producing calibrated zero-shot predictions in a single forward pass.
- Trained entirely on synthetic tables from structural causal models, learning general representations that transfer zero-shot to real-world tasks
- Ranks first among default tabular foundation models across all 51 TabArena benchmarks (38 classification, 13 regression), outperforming tuned AutoML pipelines
- Two extensions on the same frozen weights go further: TabFM+ adds multi-view feature expansion with ensembling and post-hoc calibration; TabFM-Auto adds LLM-guided data processing and feature engineering
More from Models
- mradermacher quants get Gemma 26B to 75 tok/s on 2x RTX 4060 8GB — Spiritual_Impress_30 · 2026-09-30
- DeepSeek is giving users 6 yuan in free API credits via its harness — teortaxesTex · 2026-09-30
- User claims OpenAI bots autonomously scan your Gmail after connecting and keep the data — alexcovo_eth · 2026-09-30
- Rumor: DeepSeek's rumored single-GPU model may have been trained on Ascend — teortaxesTex · 2026-09-30
- DeepSeek opens community feedback channel: flip a toggle in the harness to help improve its models — teortaxesTex · 2026-09-30
- You can drop the vision encoder once pretraining compute exceeds 1e22 FLOPs — heghbalz · 2026-09-30