NPU Adaption for Sanna

Author: J石页

src/diffusers/models/attention_processor.py

@@ -3158,7 +3158,7 @@ def __call__(
             attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
             # scaled_dot_product_attention expects attention_mask shape to be
             # (batch, heads, source_length, target_length)
-            attention_mask = attention_mask.view(batch_size, 1, -1, attention_mask.shape[-1])
+            attention_mask = attention_mask.view(batch_size, attn.heads, -1, attention_mask.shape[-1])
             attention_mask = attention_mask.repeat_interleave(hidden_states.shape[1], dim=2)
             if attention_mask.dtype == torch.bool:
                 attention_mask = torch.logical_not(attention_mask.bool())