feat: introduce reward ablation configurations for enhanced training flexibility, implement YAML loading with extends support, and add reward variant tracking in training scripts

f7b8ac6

Humanlearning commited on 12 days ago