How to use the byteps.mxnet.ops.byteps_push_pull function in byteps

def _do_push_pull_param(self, index, delta_weight):
        if isinstance(index, (tuple, list)):
            for i in range(len(index)):
                byteps_declare_tensor("weight_" + str(index[i]))
                byteps_push_pull(delta_weight[i], version=0, priority=-index[i],
                                 name="weight_" + str(index[i]), is_average=False)
        else:
            byteps_declare_tensor("weight_" + str(index))
            byteps_push_pull(delta_weight, version=0, priority=-index,
                             name="weight_" + str(index), is_average=False)

def _do_push_pull(self, index, grad):
        if isinstance(index, (tuple, list)):
            for i in range(len(index)):
                byteps_declare_tensor("gradient_" + str(index[i]))
                byteps_push_pull(grad[i], version=0, priority=-index[i],
                                 name="gradient_" + str(index[i]), is_average=True)
        else:
            byteps_declare_tensor("gradient_" + str(index))
            byteps_push_pull(grad, version=0, priority=-index,
                             name="gradient_" + str(index), is_average=True)

def _init_params(self):
        tensors = []
        for param in self._params_to_init:
            if param._deferred_init:
                tensors.append(param)
            else:
                param_arrays = param._check_and_get(param._data, list)
                idx = self._param2idx[param.name]

                if rank() != self.root_rank:
                    param_arrays[0].__imul__(0)
                byteps_push_pull(param_arrays[0], version=0, priority=0,
                                 name="parameter_" + str(idx), is_average=False)
                param_arrays[0].wait_to_read()

        self._params_to_init = tensors

def _do_push_pull(self, index, grad):
        if isinstance(index, (tuple, list)):
            for i in range(len(index)):
                byteps_declare_tensor("gradient_" + str(index[i]))
                byteps_push_pull(grad[i], version=0, priority=-index[i],
                                 name="gradient_" + str(index[i]), is_average=True)
        else:
            byteps_declare_tensor("gradient_" + str(index))
            byteps_push_pull(grad, version=0, priority=-index,
                             name="gradient_" + str(index), is_average=True)

def _do_push_pull_param(self, index, delta_weight):
        if isinstance(index, (tuple, list)):
            for i in range(len(index)):
                byteps_declare_tensor("weight_" + str(index[i]))
                byteps_push_pull(delta_weight[i], version=0, priority=-index[i],
                                 name="weight_" + str(index[i]), is_average=False)
        else:
            byteps_declare_tensor("weight_" + str(index))
            byteps_push_pull(delta_weight, version=0, priority=-index,
                             name="weight_" + str(index), is_average=False)

def _allreduce_grads(self):
        for i, param in enumerate(self._params):
            if param.grad_req != 'null':
                byteps_push_pull(param.list_grad()[0], is_average=False,
                                 name="gradient_" + str(i), priority=-i)