MCPcopy Create free account
hub / github.com/tensorflow/lingvo / _ZeroOutFirstTimeStep

Method _ZeroOutFirstTimeStep

lingvo/tasks/mt/decoder.py:1433–1455  ·  view source on GitHub ↗

Zeroes out the first time step. Args: token_embs: [batch, time, model_dim] embeding lookups batch: Batch size scalar target_time: Target sequence length scalar. Returns: modified token_embs with the first time step zeroed out.

(self, token_embs, batch, target_time)

Source from the content-addressed store, hash-verified

source not stored for this graph (policy: none)

Callers 1

_FPropMethod · 0.95

Calls

no outgoing calls

Tested by

no test coverage detected