diff --git a/Dockerfile.rocm b/Dockerfile.rocm
index 3c76305303037..f49b321372ed0 100644
--- a/Dockerfile.rocm
+++ b/Dockerfile.rocm
@@ -76,6 +76,7 @@ RUN cd /app \
     && cd vllm \
     && pip install -U -r requirements-rocm.txt \
     && bash patch_xformers.rocm.sh \
+    && patch /opt/rocm/include/hip/amd_detail/amd_hip_bf16.h /app/vllm/rocm_patch/rocm_bf16.patch \
     && python3 setup.py install \
     && cd ..
 
diff --git a/rocm_patch/rocm_bf16.patch b/rocm_patch/rocm_bf16.patch
new file mode 100644
index 0000000000000..a0f07da2a3e2b
--- /dev/null
+++ b/rocm_patch/rocm_bf16.patch
@@ -0,0 +1,15 @@
+--- amd_hip_bf16.h	2024-02-06 18:28:58.268699142 +0000
++++ amd_hip_bf16.h.new	2024-02-06 18:28:31.988647133 +0000
+@@ -90,10 +90,10 @@
+ #include "math_fwd.h"              // ocml device functions
+ 
+ #if defined(__HIPCC_RTC__)
+-#define __HOST_DEVICE__ __device__
++#define __HOST_DEVICE__ __device__ static
+ #else
+ #include <climits>
+-#define __HOST_DEVICE__ __host__ __device__
++#define __HOST_DEVICE__ __host__ __device__ static inline
+ #endif
+ 
+ // Since we are using unsigned short to represent data in bfloat16, it can be of different sizes on