malarinv
/
tacotron2
mirror of https://github.com/malarinv/tacotron2
1
0
Fork 0
Code Issues Releases Wiki Activity

train.py: patching score_mask_value formerly inf, not concrete value, for compatibility with pytorch

load_mel_from_disk
Rafael Valle 2018-05-15 09:50:56 -07:00
parent cd851585cb
commit 1071023017
1 changed files with 4 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
train.py
View File

@ -2,6 +2,7 @@ import os
import time
import argparse
import math
from numpy import finfo
import torch
from distributed import DistributedDataParallel
@ -77,7 +78,9 @@ def prepare_directories_and_logger(output_directory, log_directory, rank):
def load_model(hparams):
model = Tacotron2(hparams).cuda()
model = batchnorm_to_float(model.half()) if hparams.fp16_run else model
if hparams.fp16_run:
model = batchnorm_to_float(model.half())
model.decoder.attention_layer.score_mask_value = float(finfo('float16').min)
if hparams.distributed_run:
model = DistributedDataParallel(model)