diff --git a/src/core/cuda/sageattention/mma.cuh b/src/core/cuda/sageattention/mma.cuh
index 4de48286..f473fa72 100644
--- a/src/core/cuda/sageattention/mma.cuh
+++ b/src/core/cuda/sageattention/mma.cuh
@@ -22,6 +22,7 @@
 #include <cuda_bf16.h>
 #include <cuda_fp16.h>
 #include <cuda_runtime.h>
+#include <cassert>
 #include <type_traits>
 
 namespace sageattention {
diff --git a/src/core/cuda/sageattention/numeric_conversion.cuh b/src/core/cuda/sageattention/numeric_conversion.cuh
index 066f7fe0..8da67d5f 100644
--- a/src/core/cuda/sageattention/numeric_conversion.cuh
+++ b/src/core/cuda/sageattention/numeric_conversion.cuh
@@ -22,6 +22,7 @@
 #include <cuda_fp16.h>
 #include <cuda_fp8.h>
 #include <cuda_runtime.h>
+#include <cassert>
 
 namespace sageattention {