tabicl
TabICL v2 — Tabular Classification & Regression
in-context tabular model · classify or predict on your table
Choose this to classify rows or predict a number per row from a labelled table, without training. Inference-only.
Run it
$ pip install pfnstudio
$ pfnstudio run tabicl --input <your file>
# ex: each task on the sample input pfnstudio ships
$ pfnstudio run tabicl --task classify --input sample
$ pfnstudio run tabicl --task regress --input sample
# download it without running · its tasks, inputs, license and digest
$ pfnstudio pull tabicl
$ pfnstudio show tabicl
# models loaded now · delete it from this machine
$ pfnstudio ps
$ pfnstudio rm tabicl
Python 3.10+. The first run downloads the model's adapter, asks before installing PyTorch and the model's own packages (-y installs without asking) and pulls the pretrained weights from Hugging Face. --input sample runs the sample input pfnstudio ships and prints its path: copy it, put your data in, and pass it as <your file>. Later runs reuse the cache.
Serves
| Task | Answers |
|---|---|
| classify | which class is this row? |
| regress | what number for this row? |
Provenance
| Publisher | soda-inria ↗ |
| Weights | pretrained TabICL v2 · classifier + regressor checkpoints from HuggingFace |
| Input | (N, F) table with a target column |
| License | BSD-3-Clause |
| Adapter | basemodels/tabicl/adapters/ · tabicl |
| Source | github.com/soda-inria/tabicl ↗ |
Readme
Predict a column of your table — churn, category, price, risk — from the rows where you already know it, without training a model.
TabICL v2 (soda-inria) is an in-context learning foundation model for tabular data: the labelled rows of your table are its examples and it predicts the target for the rows left empty. It handles classification (with many classes) and regression with quantiles. This listing serves the official v2 checkpoints (HF jingang/TabICL) through the upstream tabicl package. Pre-trained on tables of 300–48,000 rows and 2–100 columns.