Edge TPU CompilerがUpdate

@Vengineerの戯言 : Twitter
SystemVerilogの世界へようこそ、すべては、SystemC v0.9公開から始まった

6月3日のブログ、Google Edge TPUの offline compilerから約2か月

July 2019 Updates

　　
　　　・New on-device transfer learning APIs
　　　・post-training quant support
　　　・new TF Lite delegate

この中で、new TF Lite delegate が気になったので。。。

TensorFlow Liteの例題、load_image.pyの

    interpreter = Interpreter(model_path=args.model_file)

を

    interpreter = Interpreter(model_path=args.model_file,
    experimental_delegates=[load_delegate('libedgetpu.so.1.0')])

に変更するだけ、experimental_delegatesに、load_delegateにてライブラリを呼び出すだけ。。。

    from tensorflow.lite.python.interpreter import load_delegate

で、load_delegate を取り込めばいい。

おっと、なんだ、experimental_delegates なんか、無かったんだけどね。

interpreter.pyを見たら、追加されていたわ。。。

  def __init__(self,
               model_path=None,
               model_content=None,
               experimental_delegates=None):
    """Constructor.
    Args:
      model_path: Path to TF-Lite Flatbuffer file.
      model_content: Content of model.
      experimental_delegates: Experimental. Subject to change. List of
        [TfLiteDelegate](https://www.tensorflow.org/lite/performance/delegates)
        objects returned by lite.load_delegate().

lite.load_delegate()

@_tf_export('lite.experimental.load_delegate')
def load_delegate(library, options=None):
  """Returns loaded Delegate object.
  Args:
    library: Name of shared library containing the
      [TfLiteDelegate](https://www.tensorflow.org/lite/performance/delegates).
    options: Dictionary of options that are required to load the delegate. All
      keys and values in the dictionary should be convertible to str. Consult
      the documentation of the specific delegate for required and legal options.
      (default None)
  Returns:
    Delegate object.
  """

  # TODO(b/137299813): Fix darwin support for delegates.
  if sys.platform == 'darwin':
    raise RuntimeError('Dynamic loading of delegates on Darwin not supported.')

  try:
    delegate = Delegate(library, options)
  except ValueError as e:
    raise ValueError('Failed to load delegate from {}\n{}'.format(
        library, str(e)))
  return delegate

まだ、darwin ではサポートされていないけど、Delegateクラスを生成するだけ。。

Delegateクラスは、

  def __init__(self, library, options=None):
    """Loads delegate from the shared library.
    Args:
      library: Shared library name.
      options: Dictionary of options that are required to load the delegate. All
        keys and values in the dictionary should be serializable. Consult the
        documentation of the specific delegate for required and legal options.
        (default None)
    Raises:
      RuntimeError: 

のように定義されていて、library が 共有ライブラリ名、optionsはオプション。デフォルト値は、None。

    self._library = ctypes.pydll.LoadLibrary(library)

のように、共有ライブラリをロードして、optionsパラメータを使って、
共有ライブラリ内のtflite_plugin_create_delegateを呼びだせばいいんだね。

    self._delegate_ptr = self._library.tflite_plugin_create_delegate(
        options_keys, options_values, len(options), error_capturer_cb)

}}}
にあって、解放は

  def __del__(self):
    # __del__ can be called multiple times, so if the delegate is destroyed.
    # don't try to destroy it twice.
    if self._library is not None:
      self._library.tflite_plugin_destroy_delegate.argtypes = [ctypes.c_void_p]
      self._library.tflite_plugin_destroy_delegate(self._delegate_ptr)
      self._library = None

にあるように、tflite_plugin_destroy_delegateを呼び出せばいいと。。

つまり、

　・tflite_plugin_create_delegate
　・tflite_plugin_destroy_delegate

があれば、Delegate作れるんだね。