LoRA（低秩适应，Low-Rank Adaptation）的 alpha 参数解析（54）

如果 聚焦于层本身 ，情况就会像下面这样简单直接

regular_output = base_layer(batch)
out_A = layer_A(batch)
additional_output = layer_B(out_A)
output = regular_output + additional_output
regular_output, additional_output, output

输出