4 ms·
No hard numbers to present, but it would be beneficial in long-sequence LSTM networks, because TensorFlow has to do time-major <=> batch-major transposition bet
by pmalynin 10y ago
No hard numbers to present, but it would be beneficial in long-sequence LSTM networks, because TensorFlow has to do time-major <=> batch-major transposition between steps.