Max Ma comments on Transformer Attention’s High School Math Mistake

[Max Ma]

DeepSeek V3 mitigated this mistake unknowingly. In their MLA, K, V shares the same nn.linear.