Update modeling_quiet.py

modeling_quiet.py

@@ -719,6 +719,12 @@ class QuietSdpaAttention(QuietAttention):
             key_states = key_states.contiguous()
             value_states = value_states.contiguous()
 
+
+        # Cast query_states, key_states, and value_states to the same data type as attention_mask
+        query_states = query_states.to(attention_mask.dtype)
+        key_states = key_states.to(attention_mask.dtype)
+        value_states = value_states.to(attention_mask.dtype)
+
         attn_output = torch.nn.functional.scaled_dot_product_attention(
             query_states,
             key_states,