The `tensorflow.mlir.experimental.convert_graph_def()` API gains an extra argument to define the pass_pipeline to run on the imported graph. Providing an empty string will not trigger any MLIR passes to be ran, the default is to run a standard set of target-independent optimizations. The syntax for this textual string representation of the pass pipeline is the MLIR standard notation, refer to the MLIR documentation for more details. PiperOrigin-RevId: 269438392
8 lines
226 B
Plaintext
8 lines
226 B
Plaintext
path: "tensorflow.mlir.experimental"
|
|
tf_module {
|
|
member_method {
|
|
name: "convert_graph_def"
|
|
argspec: "args=[\'graph_def\', \'pass_pipeline\'], varargs=None, keywords=None, defaults=[\'tf-standard-pipeline\'], "
|
|
}
|
|
}
|