|View source on GitHub|
Base class for polymorphic graph functions.
Graph functions are Python callable objects that dispatch calls to a TensorFlow graph. Polymorphic graph functions can be backed by multiple TF graphs, and automatically select the appropriate specialization based on the type of input they were called with. They may also create specializations on the fly if necessary, for example by tracing.
experimental_get_compiler_ir( *args, **kwargs )
Returns compiler IR for the compiled function.
This API is intended only for debugging as there are no guarantees on
backwards compatibility of returned IR or the allowed values of
||Arguments used for compilation; same arguments as used for calling the function. Need to be eager tensors.|
||Keyword arguments used for compilation.|
Function callable with the following kwargs:
For example, for
the output is:
If an invalid
||When called with input in graph mode.|
get_concrete_function( *args, **kwargs ) ->
ConcreteFunction specialized to input types.
The arguments specified by
kwargs follow normal function call
rules. The returned
ConcreteFunction has the same set of positional and
keyword arguments as
self, but their types are refined to the types