Update modeling_openelm.py

modeling_openelm.py

@@ -779,7 +779,9 @@ class OpenELMModel(OpenELMPreTrainedModel):
                 :, None, None, :
             ].eq(0.0)
             causal_mask = causal_mask.clone()
-            causal_mask[..., :mask_length] = causal_mask[..., :mask_length].masked_fill(...)
+            causal_mask[..., :mask_length] = causal_mask[..., :mask_length].masked_fill(
+                padding_mask, min_dtype
+            )
             #causal_mask[..., :mask_length] = causal_mask[..., :mask_length].masked_fill(
             #    padding_mask, min_dtype
             #)