hub / github.com/tensorflow/lingvo / FProp

Method FProp

lingvo/tasks/mt/encoder.py:144–198 · view source on GitHub ↗

Encodes source as represented by `inputs` and `paddings`. Args: theta: A `.NestedMap` object containing weights' values of this layer and its children layers. input_batch: A `.NestedMap` with fields: - ids: The inputs tensor. It is expected to be of shape [batch, tim

(self, theta, input_batch)

Source from the content-addressed store, hash-verified

source not stored for this graph (policy: none)

Callers 15

_ComputeXentLossMethod · 0.45

ComputePredictionsMethod · 0.45

_DecodeStepMethod · 0.45

_FPropMethod · 0.45

ExtendStepMethod · 0.45

ComputePredictionsMethod · 0.45

_FPropMethod · 0.45

ExtendStepMethod · 0.45

_FPropMethod · 0.45

_FPropSoftmaxMethod · 0.45

testDecoderFPropWithPackingMethod · 0.45

testDecoderFPropWithContextMethod · 0.45

Calls 2

ApplyClippingMethod · 0.95

EmbLookupMethod · 0.45

Tested by 6

testDecoderFPropWithPackingMethod · 0.36

testDecoderFPropWithContextMethod · 0.36

_TransformerStackFPropMethod · 0.36

testContextualTransformerStackFPropMethod · 0.36

testMaskedTransformerStackFPropMethod · 0.36

testForwardPassWithDoubleBatchMethod · 0.36