diff --git a/python/mxnet/gluon/contrib/estimator/batch_processor.py b/python/mxnet/gluon/contrib/estimator/batch_processor.py
index aa5adbfdea5f..5545a7678bcf 100644
--- a/python/mxnet/gluon/contrib/estimator/batch_processor.py
+++ b/python/mxnet/gluon/contrib/estimator/batch_processor.py
@@ -21,6 +21,7 @@
 
 from ...utils import split_and_load
 from .... import autograd
+from .... import npx
 
 __all__ = ['BatchProcessor']
 
@@ -102,4 +103,6 @@ def fit_batch(self, estimator,
         for l in loss:
             l.backward()
 
+        npx.waitall()
+
         return data, label, pred, loss
diff --git a/tests/nightly/estimator/test_sentiment_rnn.py b/tests/nightly/estimator/test_sentiment_rnn.py
index 8d1a44d71baf..311e4277179e 100644
--- a/tests/nightly/estimator/test_sentiment_rnn.py
+++ b/tests/nightly/estimator/test_sentiment_rnn.py
@@ -264,6 +264,7 @@ def test_estimator_gpu():
     num_hiddens, num_layers = 100, 2
     net = BiRNN(vocab, embed_size, num_hiddens, num_layers)
     net.initialize(mx.init.Xavier(), ctx=ctx)
+    net.hybridize()
 
     glove_embedding = text.embedding.create(
         'glove', pretrained_file_name='glove.6B.100d.txt', vocabulary=vocab)
diff --git a/tests/nightly/test_kvstore.py b/tests/nightly/test_kvstore.py
index 54801c2fbdee..aaed7055cf91 100755
--- a/tests/nightly/test_kvstore.py
+++ b/tests/nightly/test_kvstore.py
@@ -100,7 +100,7 @@ def as_float32(s):
 def test_kvstore(kv_type, stype):
     print(kv_type)
     kv = mx.kv.create(kv_type)
-    kv.set_optimizer(mx.optimizer.create('test', rescale_grad=lr))
+    kv.set_optimizer(mx.optimizer.create('test', learning_rate=-lr))
     for k, s in zip(keys, shapes):
         kv.init(k, mx.nd.zeros(s))
 
@@ -130,7 +130,7 @@ def test_compress_kvstore(kv_type, compression='2bit', threshold=0.5):
         raise RuntimeError("Unknown gradient compression type!")
     kv = mx.kv.create(kv_type)
     kv.set_gradient_compression({'type':compression, 'threshold':threshold})
-    kv.set_optimizer(mx.optimizer.create('test', rescale_grad=rate))
+    kv.set_optimizer(mx.optimizer.create('test', learning_rate=-rate))
     for k, s in zip(keys, shapes):
         kv.init(k, mx.nd.zeros(s))
     # init one key with 1s so we can check if it was compressed during init
@@ -297,7 +297,7 @@ def check_compr_random(kv, threshold):
 def test_group_kvstore(kv_type, stype):
     print(kv_type)
     kv = mx.kv.create(kv_type)
-    kv.set_optimizer(mx.optimizer.create('test', rescale_grad=lr))
+    kv.set_optimizer(mx.optimizer.create('test', learning_rate=-lr))
     kv.init(keys, [mx.nd.zeros(s) for s in shapes])
     res = [np.zeros(s) for s in shapes]
     out = [[mx.nd.zeros(s, mx.gpu(g)) for g in range(nworker)] for s in shapes]
diff --git a/tests/nightly/test_np_large_array.py b/tests/nightly/test_np_large_array.py
index fa103fdce699..c5f2a8fe4e6e 100644
--- a/tests/nightly/test_np_large_array.py
+++ b/tests/nightly/test_np_large_array.py
@@ -642,6 +642,7 @@ def test_shares_memory():
     assert out == False and out2 == True
 
 @use_np
+@pytest.mark.skip(reason='times out (20 mins)')
 def test_where():
     inp1 = np.zeros((2, INT_OVERFLOW))
     inp1[-1, -1] = 1