[NFC] polish colossalai/kernel/cuda_native/csrc/kernels/transform_kernels.cu code stype (#629)

colossalai/kernel/cuda_native/csrc/kernels/transform_kernels.cu

@@ -135,10 +135,9 @@ __global__ void bias_add_transform_20314(T *output, const T *input,
                                          const T *bias, int dim_3, int dim_4);
 
 template <>
-__global__ void bias_add_transform_20314<float>(float *output,
-                                                const float *input,
-                                                const float *bias, int dim_3,
-                                                int dim_4) {
+__global__ void
+bias_add_transform_20314<float>(float *output, const float *input,
+                                const float *bias, int dim_3, int dim_4) {
   int id0 = blockIdx.x;
   int id1 = blockIdx.y;
   int id2 = blockIdx.z;
@@ -174,10 +173,9 @@ __global__ void bias_add_transform_20314<float>(float *output,
 }
 
 template <>
-__global__ void bias_add_transform_20314<__half>(__half *output,
-                                                 const __half *input,
-                                                 const __half *bias, int dim_3,
-                                                 int dim_4) {
+__global__ void
+bias_add_transform_20314<__half>(__half *output, const __half *input,
+                                 const __half *bias, int dim_3, int dim_4) {
   int id0 = blockIdx.x;
   int id1 = blockIdx.y;
   int id2 = blockIdx.z;