[tune] upgrade gpytorch, bump default pytorch to 1.7.0 (#12776)

* upgrade gpytorch

Signed-off-by: Richard Liaw <rliaw@berkeley.edu>

* pin

Signed-off-by: Richard Liaw <rliaw@berkeley.edu>

* version-torch

Signed-off-by: Richard Liaw <rliaw@berkeley.edu>

* fix-build

Signed-off-by: Richard Liaw <rliaw@berkeley.edu>

python/ray/util/sgd/torch/examples/dcgan.py

@@ -182,8 +182,8 @@ class GANOperator(TrainingOperator):
     @override(TrainingOperator)
     def train_batch(self, batch, batch_info):
         """Trains on one batch of data from the data creator."""
-        real_label = 1
-        fake_label = 0
+        real_label = 1.0
+        fake_label = 0.
         discriminator, generator = self.models
         optimD, optimG = self.optimizers
 

python/requirements_tune.txt

@@ -1,5 +1,4 @@
 ax-platform
-gpytorch==1.2
 bayesian-optimization
 ConfigSpace==0.4.10
 dragonfly-opt