DeepSeek V4.1-Flash’s use of head-wise Muon has put the optimizer back in the spotlight. Building on our March study of LoRA with AdamW, we asked a practical question: what changes when you switch to Muon? The best tested learning rate generally moves upward, while the
