{"categories":[{"label":"Scientific/Engineering","url":"https://skillfed.io/packages/category/scientific-engineering/6"}],"enrichment":{"capability":"TabICL is a scikit-learn-compatible tabular foundation model that performs classification and regression on structured data via in-context learning, combining fit and predict in a single forward pass through a pre-trained transformer.","skillfed_tags":["tabular-learning","foundation-model","in-context-learning"],"use_cases":["Quick baseline predictions on tabular datasets without spending time tuning hyperparameters","Classification or regression on datasets with hundreds to tens of thousands of samples where GPU inference is available","Fine-tuning a pre-trained model on a single important dataset using FinetunedTabICLClassifier or FinetunedTabICLRegressor","Repeated inference on the same training data with KV caching enabled to reduce latency","Handling datasets with many features (up to 2,000) where traditional tabular models may struggle"],"what_it_does":"TabICL is a pre-trained transformer model designed to handle tabular classification and regression tasks without hyperparameter tuning. It works by performing in-context learning\u2014learning from training data within a single forward pass\u2014and can be used like a scikit-learn estimator with fit() and predict() methods. The model comes with pre-trained checkpoints and downloads them automatically on first use.\n\nThe package supports datasets ranging from hundreds to hundreds of thousands of samples with up to 2,000 features. It includes optional fine-tuning for single-dataset adaptation, KV caching to speed up repeated inference on the same training data, and CPU/disk offloading for larger datasets. Core dependencies are torch, numpy, scipy, scikit-learn, einops, huggingface-hub, psutil, and tqdm.","worth_installing":"Yes, if you work with tabular data and want a zero-tuning baseline or a scikit-learn-compatible alternative to gradient boosting. The permissive license, low install friction, active maintenance, and no known vulnerabilities make it safe to try. Caveat: torch is a heavy dependency, and GPU access is recommended for the speed claims; CPU-only inference may be slower than claimed."},"id":"tabicl","links":{"html":"https://skillfed.io/packages/tabicl","md":"https://skillfed.io/packages/tabicl.md","pypi":"https://pypi.org/project/tabicl/"},"maintenance":{"status":"active"},"meta":{"latest_release":"2026-04-29","license_spdx":null,"license_treatment":"permissive","name":"tabicl","python_support":"supports_current","summary":"TabICL: A state-of-the-art tabular foundation model"},"popularity":{"monthly_downloads":142950,"position":11193,"tier":"top_15000"},"security":{"n_vulnerabilities":0},"version":"2.1.1"}