You Could Have Come Up with Kimi Delta Attention

# Summary The article provides a detailed mathematical explanation of how linear attention mechanisms—specifically Kimi Delta Attention (KDA)—can be derived from first principles by starting with standard softmax attention and progressively introducing constraints on the hidden state. Rather than presenting KDA as a complex black box, the author shows how one could naturally arrive at its equations through a logical progression: softmax attention → linear attention → DeltaNet → Gated DeltaNet → KDA, ultimately demystifying the mechanics used in recent Qwen and Kimi language models.