Use eval to reduce time.

app.py

@@ -105,10 +105,11 @@ class Model(nn.Module):
             loss = F.cross_entropy(x.view(-1, x.shape[-1]), targ.view(-1))
         return x, loss
 
-m = Model()
-m.to(device)
+#m = Model()
+#m.to(device)
 
 m=torch.load("model_v4t.pkl",map_location=torch.device('cpu'))
+m.eval()
 
 top_k = 20
 def generate(s, num = 60):