fix error of parameter initialization for LoRa embedding

loralib/layers.py

@@ -56,8 +56,8 @@ class Embedding(nn.Embedding, LoRALayer):
         nn.Embedding.reset_parameters(self)
         if hasattr(self, 'lora_A'):
             # initialize A the same way as the default for nn.Linear and B to zero
-            nn.init.zeros_(self.lora_A)
-            nn.init.normal_(self.lora_B)
+            nn.init.normal_(self.lora_A)
+            nn.init.zeros_(self.lora_B)
 
     def train(self, mode: bool = True):
         nn.Embedding.train(self, mode)