worker-device.patch

diff --git a/TrainingInterfaces/Text_to_Spectrogram/AutoAligner/autoaligner_train_loop.py b/TrainingInterfaces/Text_to_Spectrogram/AutoAligner/autoaligner_train_loop.py
index 5fe2cd9c..ff5c5bdc 100644
--- a/TrainingInterfaces/Text_to_Spectrogram/AutoAligner/autoaligner_train_loop.py
+++ b/TrainingInterfaces/Text_to_Spectrogram/AutoAligner/autoaligner_train_loop.py
@@ -48,7 +48,7 @@ def train_loop(train_dataset,
     train_loader = DataLoader(batch_size=batch_size,
                               dataset=train_dataset,
                               drop_last=True,
-                              num_workers=8,
+                              num_workers=12 if os.cpu_count() > 12 else max(os.cpu_count() - 2, 1),
                               pin_memory=False,
                               shuffle=True,
                               prefetch_factor=16,
diff --git a/TrainingInterfaces/Text_to_Spectrogram/PortaSpeech/Glow.py b/TrainingInterfaces/Text_to_Spectrogram/PortaSpeech/Glow.py
index 804548a4..4e3cfac6 100644
--- a/TrainingInterfaces/Text_to_Spectrogram/PortaSpeech/Glow.py
+++ b/TrainingInterfaces/Text_to_Spectrogram/PortaSpeech/Glow.py
@@ -119,7 +119,7 @@ class InvConvNear(nn.Module):
             if self.no_jacobian:
                 logdet = 0
 
-        weight = weight.view(self.n_split, self.n_split, 1, 1)
+        weight = weight.view(self.n_split, self.n_split, 1, 1).to(x.device)
         z = F.conv2d(x, weight)
 
         z = z.view(b, self.n_sqz, self.n_split // self.n_sqz, c // self.n_split, t)
diff --git a/TrainingInterfaces/Text_to_Spectrogram/PortaSpeech/portaspeech_train_loop.py b/TrainingInterfaces/Text_to_Spectrogram/PortaSpeech/portaspeech_train_loop.py
index 3b596ddb..ea9c80f9 100644
--- a/TrainingInterfaces/Text_to_Spectrogram/PortaSpeech/portaspeech_train_loop.py
+++ b/TrainingInterfaces/Text_to_Spectrogram/PortaSpeech/portaspeech_train_loop.py
@@ -78,7 +78,7 @@ def train_loop(net,
     train_loader = DataLoader(batch_size=batch_size,
                               dataset=train_dataset,
                               drop_last=True,
-                              num_workers=12,
+                              num_workers=12 if os.cpu_count() > 12 else max(os.cpu_count() - 2, 1),
                               pin_memory=True,
                               shuffle=True,
                               prefetch_factor=8,