A zero-shot causal-discovery foundation model that estimates direct causes and effects of a target, with open-source code and model weights.
Nori-Causal analyzes an observational table and estimates the probability of direct causal edges between columns, including whether each feature is more likely to cause or follow a chosen target. It is a roughly 30-million-parameter model trained on synthetic structural causal models and makes its estimates in a forward pass without fitting to each dataset. The model is available through Synthefy’s open-source synthefy-nori library and downloadable weights. Its outputs are probabilities rather than a definitive causal graph; hidden confounders and other data properties can mislead the estimates, so they should be treated as evidence to investigate rather than causal conclusions.
Be the first Cosign
Add it to your Products I use list. You can also share why you use it.
