A collection of preference model pretraining checkpoints trained on general preference datasets intended as precursors for code reward models.