diff --git a/cmake/cpu_extension.cmake b/cmake/cpu_extension.cmake
index 5912c5c02ede7..426189481575b 100644
--- a/cmake/cpu_extension.cmake
+++ b/cmake/cpu_extension.cmake
@@ -16,10 +16,16 @@ include_directories("${CMAKE_SOURCE_DIR}/csrc")
 #
 # Check the compile flags
 #
-list(APPEND CXX_COMPILE_FLAGS
-    "-fopenmp"
-    "-mf16c"
-    "-DVLLM_CPU_EXTENSION")
+if (CMAKE_SYSTEM_PROCESSOR STREQUAL "ppc64le")
+    list(APPEND CXX_COMPILE_FLAGS
+        "-fopenmp"
+        "-DVLLM_CPU_EXTENSION")
+else()
+    list(APPEND CXX_COMPILE_FLAGS
+        "-fopenmp"
+        "-mf16c"
+        "-DVLLM_CPU_EXTENSION")
+endif()
 
 execute_process(COMMAND cat /proc/cpuinfo
                 RESULT_VARIABLE CPUINFO_RET
diff --git a/csrc/cpu/attention.cpp b/csrc/cpu/attention.cpp
index e73eca1b345fd..e6c03dcb034fd 100644
--- a/csrc/cpu/attention.cpp
+++ b/csrc/cpu/attention.cpp
@@ -24,12 +24,20 @@ struct KernelVecType<float> {
 
 template <>
 struct KernelVecType<c10::Half> {
+#ifdef __powerpc64__
+  // Power architecture-specific vector types
+  using q_load_vec_type = vec_op::FP32Vec8;
+  using k_load_vec_type = vec_op::FP32Vec16;
+  using v_load_vec_type = vec_op::FP32Vec16;
+#else
+  // Fallback for other architectures, including x86
   using q_load_vec_type = vec_op::FP16Vec8;
-  using q_vec_type = vec_op::FP32Vec16;
   using k_load_vec_type = vec_op::FP16Vec16;
+  using v_load_vec_type = vec_op::FP16Vec16;
+#endif
+  using q_vec_type = vec_op::FP32Vec16;
   using k_vec_type = vec_op::FP32Vec16;
   using qk_acc_vec_type = vec_op::FP32Vec16;
-  using v_load_vec_type = vec_op::FP16Vec16;
 };
 
 #ifdef __AVX512BF16__
diff --git a/csrc/cpu/quant.cpp b/csrc/cpu/quant.cpp
index f42fa2361a2db..d9aed657a3113 100644
--- a/csrc/cpu/quant.cpp
+++ b/csrc/cpu/quant.cpp
@@ -25,7 +25,13 @@ struct KernelVecType<c10::BFloat16> {
 
 template <>
 struct KernelVecType<c10::Half> {
+#ifdef __powerpc64__
+  // Power architecture-specific vector type
+  using load_vec_type = vec_op::FP32Vec16;
+#else
+  // Fallback for other architectures
   using load_vec_type = vec_op::FP16Vec16;
+#endif
   using azp_adj_load_vec_type = vec_op::INT32Vec16;
   using cvt_vec_type = vec_op::FP32Vec16;
 };