cdfm

Run it

$ pip install pfnstudio
$ pfnstudio run cdfm --input <your file>

# ex: each task on the sample input pfnstudio ships
$ pfnstudio run cdfm --task discover --input sample
$ pfnstudio run cdfm --task impute --input sample

# download it without running · its tasks, inputs, license and digest
$ pfnstudio pull cdfm
$ pfnstudio show cdfm

# models loaded now · delete it from this machine
$ pfnstudio ps
$ pfnstudio rm cdfm

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

Headline numbers

Provenance

Readme

A fast first causal map of any dataset where the rows are independent samples (not a time series): which columns drive which others — before you invest in a deeper analysis.

CDFM (Qiao, Cai, Li et al., arXiv:2607.11508) is a 9.66M-parameter transformer pretrained on a large, diverse space of synthetic structural causal models spanning 15 mechanism families. Given an observational table X (N samples × D variables) it predicts the D×D causal graph in one forward pass, for any N and D. It also carries a quantile-trained imputation head for missing values. This listing installs the real pretrained weights as an inference-only base model: the checkpoint is pulled from HuggingFace (DMIRLAB/CDFM) and served through the upstream cdfm-base package. Continued pretraining is not offered — the upstream package ships no training prior.