predictlm 11.1.0

Compact distilled tabular foundation model for in-context regression and classification.

predictlm

A small, calibrated tabular foundation model for in-context regression and classification. Two open-weight checkpoints — Mini (13M) and Base (26M) — both Apache-2.0. Headline: 0.751 classification accuracy / 0.609 regression R² on a locked 25-dataset OpenML benchmark via the published Duo + TTT recipe.

Install

pip install predictlm

Use

from predictlm import PredictLM
import pandas as pd

# One model object — partner checkpoint is auto-downloaded on first .predict()
# and the package runs the published Duo + TTT recipe under the hood.
model = PredictLM.from_pretrained("zerooneresearch/predictlm-mini-13m")

X_train = pd.DataFrame(...)   # labeled context rows
y_train = pd.Series(...)      # labels — float → regression, int/str → classification
X_test  = pd.DataFrame(...)   # rows you want predictions for

preds = model.fit(X_train, y_train).predict(X_test)
probs = model.predict_proba(X_test)

PredictLM auto-detects regression vs classification from the dtype of y_train and routes through the matching head. Same model, single fit/predict API.

Single-model fast path

Skip the Duo + TTT recipe (no partner download, no test-time training):

model = PredictLM.from_pretrained("zerooneresearch/predictlm-mini-13m", auto_duo=False)
preds = model.fit(X_train, y_train).predict(X_test)
# Zero-tuning baseline: ~0.673 cls / 0.536 reg

Model cards

• zerooneresearch/predictlm-mini-13m — 13.5M params, 54 MB. Distilled from Base. Recommended for CPU / edge.
• zerooneresearch/predictlm-base-26m — 26.2M params, 105 MB. Highest-accuracy in the family.

License

Apache-2.0. Commercial use allowed.

Citation

@misc{zerooneresearch_predictlm_2026,
  title  = {PredictLM v1: Open-weight tabular foundation models with calibrated outputs},
  author = {Zero One Research},
  year   = {2026},
  url    = {https://huggingface.co/zerooneresearch}
}