diff --git a/modelopt/torch/export/plugins/megatron_importer.py b/modelopt/torch/export/plugins/megatron_importer.py
index 696af632..4c805dc0 100644
--- a/modelopt/torch/export/plugins/megatron_importer.py
+++ b/modelopt/torch/export/plugins/megatron_importer.py
@@ -512,7 +512,7 @@ def _import_state_dict(self):
                             self.rules["k_layernorm"](attention.k_layernorm, layer_id)
                         self.rules["linear_qkv"](attention.linear_qkv, layer_id)
                         self.rules["linear_proj"](attention.linear_proj, layer_id)
-                        if hasattr(attention.core_attention, "softmax_offset"):
+                        if getattr(attention.core_attention, "softmax_offset", None) is not None:
                             self.rules["softmax_offset"](
                                 attention.core_attention.softmax_offset, layer_id
                             )