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

train.py

@@ -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)