Remove in-place operation

mlcommons · Dec 7, 2023 · ece090b · ece090b
1 parent eff374c
commit ece090b
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/algorithmic_efficiency/workloads/wmt/wmt_pytorch/models.py b/algorithmic_efficiency/workloads/wmt/wmt_pytorch/models.py
@@ -1004,7 +1004,7 @@ def forward(self,
     dropout_rate = self.dropout if self.training else 0.0
 
     # Calculate attention.
-    q.mul_(self.attention_temp)
+    q = self.attention_temp * q
     attn_output = torch.nn.functional.scaled_dot_product_attention(
         q, k, v, attn_mask, dropout_rate)
     # Rearrange for output projection.