Should axes
be required parameter for layerNormalization build method?
#487
Labels
axes
be required parameter for layerNormalization build method?
#487
(This was raised by @wacky6 at Chromium CL-5068275 review)
In proposal Add support for operations needed for well-known transformers , the
axes
oflayerNormalization
operator is defined as an optional member ofMLLayerNormalizationOptions
dictionary with default value [1, 2, 3].@wacky6 mentioned TensorFlow's layerNormalization defaults to the last dimension (axis=-1) while PyTorch's LayerNorm requires the
normalized_shape
to be passed./cc @wchao1115 @fdwr
The text was updated successfully, but these errors were encountered: