fix(transformerblock): conditionally initialize cross attention components

Initialize cross attention layers only when with_cross_attention is True to avoid unnecessary computation and memory usage

Author: yunhaoli24

monai/networks/blocks/transformerblock.py

@@ -79,16 +79,16 @@ def __init__(
         )
         self.norm2 = nn.LayerNorm(hidden_size)
         self.with_cross_attention = with_cross_attention
-
-        self.norm_cross_attn = nn.LayerNorm(hidden_size)
-        self.cross_attn = CrossAttentionBlock(
-            hidden_size=hidden_size,
-            num_heads=num_heads,
-            dropout_rate=dropout_rate,
-            qkv_bias=qkv_bias,
-            causal=False,
-            use_flash_attention=use_flash_attention,
-        )
+        if with_cross_attention:
+            self.norm_cross_attn = nn.LayerNorm(hidden_size)
+            self.cross_attn = CrossAttentionBlock(
+                hidden_size=hidden_size,
+                num_heads=num_heads,
+                dropout_rate=dropout_rate,
+                qkv_bias=qkv_bias,
+                causal=False,
+                use_flash_attention=use_flash_attention,
+            )
 
     def forward(
         self, x: torch.Tensor, context: Optional[torch.Tensor] = None, attn_mask: Optional[torch.Tensor] = None