def nonstatic(features, labels, mode, params):
with tf.device('/cpu:0'):
word_vecs = embed.nonstatic(features)
with tf.device('/gpu:0'):
normed_word_vecs = normalize(word_vecs, mode)
rnn = convolution(normed_word_vecs, features['INPUTLEN'], mode)
fully_connected = dense(rnn, mode)
outputs = output.output(fully_connected)
return output.train_or_predict(features, labels, mode, params, outputs)
def nonstatic(features, labels, mode, params):
with tf.device('/gpu:{}'.format(hypers.get_param('sg'))):
word_vecs = embed.nonstatic(features)
with tf.device('/gpu:{}'.format(hypers.get_param('sg'))):
normed_word_vecs = normalize(word_vecs, mode)
rnn = recurrent(normed_word_vecs, features['INPUTLEN'], mode)
fully_connected = dense(rnn, mode)
outputs = output.output(fully_connected)
return output.train_or_predict(features, labels, mode, params, outputs)