Missed TensorFlow Dev Summit? Check out the video playlist. Watch recordings

Module: tfp.experimental.auto_batching.dsl

View source on GitHub

Python-embedded DSL frontend for authoring autobatchable IR programs.

This domain-specific language frontend serves two purposes:

  • Being easier and more pleasant to author by humans than writing IR directly
  • Being close enough to the structure of natural Python programs that the remaining gap can be bridged by a source-to-source transformation

Classes

class ProgramBuilder: An auto-batching DSL context.