Fix attention_weights referenced before assigned bug

llama_diffusion_model.py

@@ -51,7 +51,8 @@ class BidirectionalLlamaAttention(LlamaAttention):
         # if attention_mask is not None:
         #     causal_mask = attention_mask[:, :, :, : key_states.shape[-2]]
         #     attn_weights = attn_weights + causal_mask
-
+        
+        attn_weights = torch.matmul(query, key_states.transpose(2, 3)) * scaling
         if attention_mask is not None:
             # Convert bool -> float with -inf where masked
             attn_mask = attention_mask.masked_fill(~attention_mask, float('-inf')).to(query.dtype)