experiments/STT
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Reload weights after plateau.

Browse Source
This commit is contained in:
Daniel 2020-08-12 16:47:02 +02:00 committed by Reuben Morais
parent eb23728538
commit 09e1422278
2 changed files with 4 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
.gitignore vendored
View File

@ -37,3 +37,4 @@ Dockerfile.train
doc/xml-c doc/xml-c
doc/xml-java doc/xml-java
doc/xml-dotnet doc/xml-dotnet
convert_graphdef_memmapped_format

3
training/mozilla_voice_stt_training/train.py
View File

@ -645,6 +645,9 @@ def train():
log_info('Encountered a plateau, reducing learning rate to {}'.format( log_info('Encountered a plateau, reducing learning rate to {}'.format(
current_learning_rate)) current_learning_rate))
# Reload checkpoint that we use the best_dev weights again
load_or_init_graph_for_training(session)
if FLAGS.metrics_files: if FLAGS.metrics_files:
# Read only metrics, not affecting best validation loss tracking # Read only metrics, not affecting best validation loss tracking
for source, init_op in zip(metrics_sources, metrics_init_ops): for source, init_op in zip(metrics_sources, metrics_init_ops):