diff --git a/ci/jenkins_tests/run_multi_node_tests.sh b/ci/jenkins_tests/run_multi_node_tests.sh
index e405f1bee..62db30625 100755
--- a/ci/jenkins_tests/run_multi_node_tests.sh
+++ b/ci/jenkins_tests/run_multi_node_tests.sh
@@ -435,7 +435,7 @@ python3 $ROOT_DIR/multi_node_docker_test.py \
     --docker-image=$DOCKER_SHA \
     --num-nodes=5 \
     --num-redis-shards=10 \
-    --test-script=/ray/test/jenkins_tests/multi_node_tests/test_0.py
+    --test-script=/ray/ci/jenkins_tests/multi_node_tests/test_0.py
 
 python3 $ROOT_DIR/multi_node_docker_test.py \
     --docker-image=$DOCKER_SHA \
@@ -444,7 +444,7 @@ python3 $ROOT_DIR/multi_node_docker_test.py \
     --num-gpus=0,1,2,3,4 \
     --num-drivers=7 \
     --driver-locations=0,1,0,1,2,3,4 \
-    --test-script=/ray/test/jenkins_tests/multi_node_tests/remove_driver_test.py
+    --test-script=/ray/ci/jenkins_tests/multi_node_tests/remove_driver_test.py
 
 python3 $ROOT_DIR/multi_node_docker_test.py \
     --docker-image=$DOCKER_SHA \
@@ -452,14 +452,14 @@ python3 $ROOT_DIR/multi_node_docker_test.py \
     --num-redis-shards=2 \
     --num-gpus=0,0,5,6,50 \
     --num-drivers=100 \
-    --test-script=/ray/test/jenkins_tests/multi_node_tests/many_drivers_test.py
+    --test-script=/ray/ci/jenkins_tests/multi_node_tests/many_drivers_test.py
 
 python3 $ROOT_DIR/multi_node_docker_test.py \
     --docker-image=$DOCKER_SHA \
     --num-nodes=1 \
     --mem-size=60G \
     --shm-size=60G \
-    --test-script=/ray/test/jenkins_tests/multi_node_tests/large_memory_test.py
+    --test-script=/ray/ci/jenkins_tests/multi_node_tests/large_memory_test.py
 
 
 docker run --rm --shm-size=${SHM_SIZE} --memory=${MEMORY_SIZE} $DOCKER_SHA \
diff --git a/python/ray/experimental/sgd/mnist_example.py b/python/ray/experimental/sgd/mnist_example.py
index 836126a9c..69abffa3c 100755
--- a/python/ray/experimental/sgd/mnist_example.py
+++ b/python/ray/experimental/sgd/mnist_example.py
@@ -24,7 +24,7 @@ from ray.tune import run_experiments
 from ray.tune.examples.tune_mnist_ray import deepnn
 from ray.experimental.sgd.model import Model
 from ray.experimental.sgd.sgd import DistributedSGD
-import ray.experimental.tf_utils
+import ray.experimental.tf_utils as ray_tf_utils
 
 parser = argparse.ArgumentParser()
 parser.add_argument("--redis-address", default=None, type=str)
@@ -67,7 +67,7 @@ class MNISTModel(Model):
             tf.nn.softmax_cross_entropy_with_logits(
                 labels=self.y_, logits=y_conv))
         self.optimizer = tf.train.AdamOptimizer(1e-4)
-        self.variables = ray.experimental.tfutils.TensorFlowVariables(
+        self.variables = ray_tf_utils.TensorFlowVariables(
             self.loss, tf.get_default_session())
 
         # For evaluating test accuracy
diff --git a/python/ray/experimental/sgd/modified_allreduce.py b/python/ray/experimental/sgd/modified_allreduce.py
index 7c446aa97..adf79a060 100644
--- a/python/ray/experimental/sgd/modified_allreduce.py
+++ b/python/ray/experimental/sgd/modified_allreduce.py
@@ -27,7 +27,6 @@ import re
 from six.moves import xrange  # pylint: disable=redefined-builtin
 import tensorflow as tf
 
-from tensorflow.contrib import nccl
 from tensorflow.contrib.all_reduce.python import all_reduce
 
 logger = logging.getLogger(__name__)
@@ -309,7 +308,8 @@ def sum_grad_and_var_all_reduce(grad_and_vars,
         #   ((grad0_gpu0, var0_gpu0), ... , (grad0_gpuN, var0_gpuN))
         scaled_grads = [g for g, _ in grad_and_vars]
         if alg == 'nccl':
-            summed_grads = nccl.all_sum(scaled_grads)
+            from tensorflow.python.ops import nccl_ops
+            summed_grads = nccl_ops.all_sum(scaled_grads)
         elif alg == 'simple':
             summed_grads = build_reduce_sum(scaled_grads)
         elif alg == 'trivial':
diff --git a/python/ray/experimental/sgd/tfbench/test_model.py b/python/ray/experimental/sgd/tfbench/test_model.py
index 0fe081607..ab625143c 100644
--- a/python/ray/experimental/sgd/tfbench/test_model.py
+++ b/python/ray/experimental/sgd/tfbench/test_model.py
@@ -6,7 +6,7 @@ import tensorflow as tf
 
 from tfbench import model_config
 from ray.experimental.sgd.model import Model
-import ray.experimental.tf_utils
+import ray.experimental.tf_utils as ray_tf_utils
 
 
 class MockDataset():
@@ -47,7 +47,7 @@ class TFBenchModel(Model):
         self.loss = tf.reduce_mean(loss, name='xentropy-loss')
         self.optimizer = tf.train.GradientDescentOptimizer(1e-6)
 
-        self.variables = ray.experimental.tf_utils.TensorFlowVariables(
+        self.variables = ray_tf_utils.TensorFlowVariables(
             self.loss, tf.get_default_session())
 
     def get_loss(self):