search

NervanaSystems / deepspeech

DeepSpeech neon implementation
Apache License 2.0
222 stars 69 forks source link

Deepbirnn layer split_input parameter #16

Closed ghost closed 7 years ago

ghost commented 7 years ago

neon Deepbirnn layer split_input parameter？ When the second layer rnn time, split_input parameter is set to true ,,, then forward with the former 1152 parameters, backward spread with 1152 parameters ,,, Why is this set?

ghost commented 7 years ago

if i want use tensorlayer to achieve ,what should i do?

tyler-nervana commented 7 years ago

The split_input parameter allows one to stack multiple forward RNNs and multiple backward RNNs before eventually combining their outputs at the top layer. It basically keeps the outputs from each layer separate as they are propagated up through next RNN layer. Could you clarify what you mean by "tensorlayer"?