---
license: cc-by-4.0
---

# steer-rt-checkpoints

RT-12 leave-one-database-out checkpoints (one directory per held-out RelBench database), trained
by the STEER authors with the [Relational Transformer](https://github.com/stanford-star/relational-transformer)
training pipeline on the eight RelBench databases used in the paper. Cite RelBench and the Relational Transformer.

Part of [STEER](https://github.com/lids-lab/steer): *STEER: Reducing Inference Cost in Relational
Foundation Models through Semantically Informed Sampling* (Mohamed and Aboulnaga, arXiv:2610.00907).