“The training results of the DeepSeek-R1-Zero model can be replicated using Low-rank adaptation (LoRA).”