diff --git a/PixArt/models/PixArt_blocks.py b/PixArt/models/PixArt_blocks.py
index a7b604b..2fbc176 100644
--- a/PixArt/models/PixArt_blocks.py
+++ b/PixArt/models/PixArt_blocks.py
@@ -189,6 +189,7 @@ class AttentionKVCompress(Attention_):
             q, k, v = map(lambda t: t.transpose(1, 2),(q, k, v),)
             x = torch.nn.functional.scaled_dot_product_attention(
                 q, k, v,
+                dropout_p=self.attn_drop.p,
                 attn_mask=attn_bias
             ).transpose(1, 2).contiguous()
         x = x.view(B, N, C)
@@ -467,4 +468,4 @@ class CaptionEmbedderDoubleBr(nn.Module):
         if (train and use_dropout) or (force_drop_ids is not None):
             global_caption, caption = self.token_drop(global_caption, caption, force_drop_ids)
         y_embed = self.proj(global_caption)
-        return y_embed, caption
\ No newline at end of file
+        return y_embed, caption