Sebastian Raschka

@BiploveYadav The model (training) should work fine without format reward, it’s just an optional stylistic thing. If the model never gets the answers right then you’d need to start with a simpler dataset.