diff --git a/onnxruntime/contrib_ops/cpu/transformers/beam_search_impl_gpt.h b/onnxruntime/contrib_ops/cpu/transformers/beam_search_impl_gpt.h
index fd02825e75..97d62a81e7 100644
--- a/onnxruntime/contrib_ops/cpu/transformers/beam_search_impl_gpt.h
+++ b/onnxruntime/contrib_ops/cpu/transformers/beam_search_impl_gpt.h
@@ -55,11 +55,11 @@ class BeamSearchGpt : public BeamSearchBase<T> {
     cuda_device_prop_ = cuda_device_prop;
     cuda_device_arch_ = cuda_device_arch;
     if (gpt_subgraph_.has_decoder_masked_attention_) {
-      ORT_RETURN_IF(cuda_device_arch_ >= 530,
-                    "Decoder masked self attention can only be used on "
-                    "GPU cards of compute capability 5.3 or higher. "
-                    "This card has compute capability ",
-                    cuda_device_arch_);
+      ORT_RETURN_IF_NOT(cuda_device_arch_ >= 530,
+                        "Decoder masked self attention can only be used on "
+                        "GPU cards of compute capability 5.3 or higher. "
+                        "This card has compute capability ",
+                        cuda_device_arch_);
     }
     return Status::OK();
   }
diff --git a/onnxruntime/contrib_ops/cpu/transformers/beam_search_impl_t5.h b/onnxruntime/contrib_ops/cpu/transformers/beam_search_impl_t5.h
index ea15656e7e..72db3b1d0b 100644
--- a/onnxruntime/contrib_ops/cpu/transformers/beam_search_impl_t5.h
+++ b/onnxruntime/contrib_ops/cpu/transformers/beam_search_impl_t5.h
@@ -64,11 +64,11 @@ class BeamSearchT5 : public BeamSearchBase<T> {
     cuda_device_prop_ = cuda_device_prop;
     cuda_device_arch_ = cuda_device_arch;
     if (decoder_subgraph_.has_decoder_masked_attention_) {
-      ORT_RETURN_IF(cuda_device_arch_ >= 530,
-                    "Decoder masked multihead attention can only be used on "
-                    "GPU cards of compute capability 5.3 or higher. "
-                    "This card has compute capability ",
-                    cuda_device_arch_);
+      ORT_RETURN_IF_NOT(cuda_device_arch_ >= 530,
+                        "Decoder masked multihead attention can only be used on "
+                        "GPU cards of compute capability 5.3 or higher. "
+                        "This card has compute capability ",
+                        cuda_device_arch_);
     }
     return Status::OK();
   }
diff --git a/onnxruntime/contrib_ops/cpu/transformers/beam_search_impl_whisper.h b/onnxruntime/contrib_ops/cpu/transformers/beam_search_impl_whisper.h
index 9bfdc71802..e5bc01ef1f 100644
--- a/onnxruntime/contrib_ops/cpu/transformers/beam_search_impl_whisper.h
+++ b/onnxruntime/contrib_ops/cpu/transformers/beam_search_impl_whisper.h
@@ -62,11 +62,11 @@ class BeamSearchWhisper : public BeamSearchBase<T> {
     cuda_device_prop_ = cuda_device_prop;
     cuda_device_arch_ = cuda_device_arch;
     if (decoder_subgraph_.has_decoder_masked_attention_) {
-      ORT_RETURN_IF(cuda_device_arch_ >= 530,
-                    "Decoder masked multihead attention can only be used on "
-                    "GPU cards of compute capability 5.3 or higher. "
-                    "This card has compute capability ",
-                    cuda_device_arch_);
+      ORT_RETURN_IF_NOT(cuda_device_arch_ >= 530,
+                        "Decoder masked multihead attention can only be used on "
+                        "GPU cards of compute capability 5.3 or higher. "
+                        "This card has compute capability ",
+                        cuda_device_arch_);
     }
     return Status::OK();
   }
diff --git a/onnxruntime/contrib_ops/cpu/transformers/greedy_search_impl_gpt.h b/onnxruntime/contrib_ops/cpu/transformers/greedy_search_impl_gpt.h
index 0d10e0691d..f426b88ec6 100644
--- a/onnxruntime/contrib_ops/cpu/transformers/greedy_search_impl_gpt.h
+++ b/onnxruntime/contrib_ops/cpu/transformers/greedy_search_impl_gpt.h
@@ -69,11 +69,11 @@ class GreedySearchGpt : public GreedySearchBase<T, ParametersT> {
     cuda_device_prop_ = cuda_device_prop;
     cuda_device_arch_ = cuda_device_arch;
     if (gpt_subgraph_.has_decoder_masked_attention_) {
-      ORT_RETURN_IF(cuda_device_arch_ >= 530,
-                    "Decoder masked self attention can only be used on "
-                    "GPU cards of compute capability 5.3 or higher. "
-                    "This card has compute capability ",
-                    cuda_device_arch_);
+      ORT_RETURN_IF_NOT(cuda_device_arch_ >= 530,
+                        "Decoder masked self attention can only be used on "
+                        "GPU cards of compute capability 5.3 or higher. "
+                        "This card has compute capability ",
+                        cuda_device_arch_);
     }
     return Status::OK();
   }