Connect two nodes. Args: node_in(IrNode): the input node. node_out(IrNode): the output node.

Reduce a tensor to the destination from all others. As shown below, one process is started with a GPU and the data of this process is represente

dummy distributed optimizer Args: optimizer(None): fake optimizer strategy(N

Converts a single numerical type or iterable of numerical types into a numerical type list. Arguments: value: The value to validat

Insert an Operator according to the giving arguments, without sync_with_cpp to make the compilation faster. Args:

Delete all the elements from the queue.

dummy all gather, do nothing Args: obj(any): obj to do all gather

distributed min in fleet Args: input(numpy.array|Variable|string): output of a layer scope(Scope): specific scope Retur

Concat not make rank changed

Returns the holding pointer. NOTE(liujinnan): Make this method as a virtual method. Because we add memory compact func for VirtualMemoryAutoGrowthBest

Add an attribute with the specified name.

\brief Get the tensor data type \param[in] pd_tensor tensor. \return the tensor data type.

Get element tensor of 1D shape tensor

Inverse transformation :math:`x = f^{-1}(y)`. It's useful for "reversing" a transformation to compute one probability in terms of another.