Update ip_adapter/attention_processor_faceid.py

ip_adapter/attention_processor_faceid.py

@@ -280,7 +280,7 @@ class LoRAAttnProcessor2_0(nn.Module):
         head_dim = inner_dim // attn.heads
 
         query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-
+        query = query.to(dtype=key.dtype)
         key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
         value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
 
@@ -389,10 +389,9 @@ class LoRAIPAttnProcessor2_0(nn.Module):
         head_dim = inner_dim // attn.heads
 
         query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-
         key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
         value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
-
+        
         # the output of sdp = (batch, num_heads, seq_len, head_dim)
         # TODO: add support for attn.scale when we move to Torch 2.1
         hidden_states = F.scaled_dot_product_attention(
@@ -416,7 +415,6 @@ class LoRAIPAttnProcessor2_0(nn.Module):
             query, ip_key, ip_value, attn_mask=None, dropout_p=0.0, is_causal=False
         )
         
-
         ip_hidden_states = ip_hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
         ip_hidden_states = ip_hidden_states.to(dtype=ip_key.dtype)