Upload README.md

README.md

@@ -12,9 +12,12 @@ tags:
 
 # ArrowFM Base
 
-ArrowFM Base is a pretrained Arrow model for zero-shot causal discovery from observational tabular data.
+Arrow is a zero-shot foundation model for causal discovery from observational tabular data.
 
-Given a numeric dataset, the model predicts edge probabilities and node-order scores, then decodes them into a directed acyclic graph (DAG). Arrow is described in [Arrow: A Foundation Model for Causal Discovery](https://arxiv.org/abs/2605.07204).
+Given a numeric dataset, Arrow predicts edge-existence probabilities and node-order scores, then decodes them into a directed acyclic graph (DAG).
+
+- Paper: [Arrow: A Foundation Model for Causal Discovery](https://arxiv.org/abs/2605.07204)
+- Code: [github.com/ryan-thompson/arrowfm](https://github.com/ryan-thompson/arrowfm)
 
 ## Usage
 
@@ -30,8 +33,7 @@ Run inference:
 import torch
 from arrowfm import ArrowPredictor
 
-predictor = ArrowPredictor.from_pretrained("ryan-thompson/arrowfm-base")
-
+predictor = ArrowPredictor()
 x = torch.randn(100, 10)
 adj = predictor.predict_adjacency(x)
 ```
@@ -44,12 +46,6 @@ To also return edge probabilities:
 adj, p_hat = predictor.predict_adjacency(x, return_p_hat = True)
 ```
 
-## Notes
-
-- Inputs are standardized internally before inference.
-- The model is intended for research and exploratory causal discovery.
-- Predicted graphs should be treated as hypotheses, not definitive causal claims.
-
 ## Citation
 
 ```bibtex