Fix

Author: co63oc

paddle/phi/kernels/funcs/math.h

@@ -16,6 +16,7 @@
 
 #include "math.h"  // NOLINT
 #include "paddle/common/hostdevice.h"
+#include "paddle/phi/common/bfloat16.h"
 #include "paddle/phi/common/float16.h"
 
 namespace phi {

paddle/phi/kernels/gpu/distributed_fused_lamb_kernel.cu

@@ -1739,8 +1739,6 @@ void DistributedFusedLambKernel(
   int64_t global_rank = 0, local_rank = 0;
   ncclComm_t global_comm = nullptr, local_comm = nullptr,
              external_comm = nullptr;
-  paddle::platform::NCCLComm *nccl_comm_handle = nullptr,
-                             *local_nccl_comm_handle = nullptr;
   distributed::NCCLCommContext *comm_ctx = nullptr, *local_comm_ctx = nullptr,
                                *external_comm_ctx = nullptr;