constraining the weights to +/- 0.5 so they can be quantized with 8 bits
authorJean-Marc Valin <jmvalin@jmvalin.ca>
Mon, 14 Aug 2017 16:48:27 +0000 (12:48 -0400)
committerJean-Marc Valin <jmvalin@jmvalin.ca>
Mon, 14 Aug 2017 16:48:27 +0000 (12:48 -0400)
commit4d1e630a8e0f9d7dbf8e6ed6e548cc1a95e32e26
tree05bcfbe270c778ec3d6666045da7958e73d6912e
parent54eeea72a8e92818310955ec07f759ee6879e6df
constraining the weights to +/- 0.5 so they can be quantized with 8 bits
src/rnn.h
src/rnn_data.c
training/dump_rnn.py
training/rnn_train.py