diff --git a/tensorflow/core/common_runtime/direct_session.cc b/tensorflow/core/common_runtime/direct_session.cc
index dc2731c0da0..2cf034061ca 100644
--- a/tensorflow/core/common_runtime/direct_session.cc
+++ b/tensorflow/core/common_runtime/direct_session.cc
@@ -1342,6 +1342,7 @@ Status DirectSession::CreateGraphs(
     // Just return '1'.
     return 1;
   };
+  popts.flib_def = &client_graph->graph.flib_def();
   popts.control_flow_added = false;
 
   std::unordered_map<string, GraphDef> partitions;
diff --git a/tensorflow/core/distributed_runtime/graph_mgr.cc b/tensorflow/core/distributed_runtime/graph_mgr.cc
index 7f77bf8b4ef..dd7af86e647 100644
--- a/tensorflow/core/distributed_runtime/graph_mgr.cc
+++ b/tensorflow/core/distributed_runtime/graph_mgr.cc
@@ -155,6 +155,7 @@ Status GraphMgr::InitItem(const string& session, const GraphDef& gdef,
       return PartitionOptions::kIllegalIncarnation;
     }
   };
+  popts.flib_def = &graph.flib_def();
   popts.control_flow_added = true;
   popts.scheduling_for_recvs = graph_options.enable_recv_scheduling();
   TF_RETURN_IF_ERROR(Partition(popts, &graph, &partitions));
diff --git a/tensorflow/core/distributed_runtime/master_session.cc b/tensorflow/core/distributed_runtime/master_session.cc
index 361e89290d2..920a9c53b28 100644
--- a/tensorflow/core/distributed_runtime/master_session.cc
+++ b/tensorflow/core/distributed_runtime/master_session.cc
@@ -76,7 +76,7 @@ class MasterSession::ReffedClientGraph : public core::RefCounted {
         debug_opts_(bopts.debug_options),
         worker_cache_(worker_cache) {
     VLOG(1) << "Created ReffedClientGraph for node with "
-            << client_graph_->graph.num_node_ids();
+            << client_graph()->graph.num_node_ids();
 
     stats_publisher_ = stats_publisher_factory(handle, bopts, session_opts);
 
@@ -166,8 +166,7 @@ class MasterSession::ReffedClientGraph : public core::RefCounted {
 
   // Partitions the graph into subgraphs and registers them on
   // workers.
-  Status RegisterPartitions(const PartitionOptions& popts,
-                            const FunctionLibraryDefinition& flib_def);
+  Status RegisterPartitions(const PartitionOptions& popts);
 
   // Runs one step of all partitions.
   Status RunPartitions(const MasterEnv* env, int64 step_id,
@@ -263,7 +262,7 @@ class MasterSession::ReffedClientGraph : public core::RefCounted {
       PartitionOptions pots,
       std::unordered_map<string, GraphDef>* out_partitions);
   Status DoRegisterPartitions(
-      const PartitionOptions& popts, const FunctionDefLibrary& func_def_lib,
+      const PartitionOptions& popts,
       std::unordered_map<string, GraphDef> graph_partitions);
 
   // Deregisters the partitions on the workers.  Called in the
@@ -274,7 +273,7 @@ class MasterSession::ReffedClientGraph : public core::RefCounted {
 };
 
 Status MasterSession::ReffedClientGraph::RegisterPartitions(
-    const PartitionOptions& popts, const FunctionLibraryDefinition& flib_def) {
+    const PartitionOptions& popts) {
   {  // Ensure register once.
     mu_.lock();
     if (!init_started_) {
@@ -293,8 +292,7 @@ Status MasterSession::ReffedClientGraph::RegisterPartitions(
           graph_defs_for_publishing.push_back(&name_def.second);
         }
         stats_publisher_->PublishGraphProto(graph_defs_for_publishing);
-        s = DoRegisterPartitions(popts, flib_def.ToProto(),
-                                 std::move(graph_defs));
+        s = DoRegisterPartitions(popts, std::move(graph_defs));
       }
       mu_.lock();
       init_result_ = s;
@@ -374,7 +372,7 @@ Status MasterSession::ReffedClientGraph::DoBuildPartitions(
 }
 
 Status MasterSession::ReffedClientGraph::DoRegisterPartitions(
-    const PartitionOptions& popts, const FunctionDefLibrary& func_def_lib,
+    const PartitionOptions& popts,
     std::unordered_map<string, GraphDef> graph_partitions) {
   partitions_.reserve(graph_partitions.size());
   Status s;
@@ -408,8 +406,6 @@ Status MasterSession::ReffedClientGraph::DoRegisterPartitions(
     Call* c = &calls[i];
     c->req.set_session_handle(session_handle_);
     c->req.mutable_graph_def()->Swap(&graph_partitions[part.name]);
-    // For simplicity, we ship the library completely to every worker.
-    *c->req.mutable_graph_def()->mutable_library() = func_def_lib;
     *c->req.mutable_graph_options() = session_opts_.config.graph_options();
     *c->req.mutable_debug_options() = debug_opts_;
     VLOG(2) << "Register " << c->req.graph_def().DebugString();
@@ -1305,6 +1301,7 @@ Status MasterSession::BuildAndRegisterPartitions(ReffedClientGraph* rcg) {
     mutex_lock l(mu_);
     return strings::StrCat(prefix, "_S", next_node_id_++);
   };
+  popts.flib_def = rcg->client_graph()->flib_def.get();
   popts.get_incarnation = [this](const string& name) -> int64 {
     Device* d = devices_->FindDeviceByName(name);
     if (d == nullptr) {
@@ -1332,8 +1329,7 @@ Status MasterSession::BuildAndRegisterPartitions(ReffedClientGraph* rcg) {
     popts.need_to_record_start_times = true;
   }
 
-  TF_RETURN_IF_ERROR(
-      rcg->RegisterPartitions(popts, *rcg->client_graph()->flib_def));
+  TF_RETURN_IF_ERROR(rcg->RegisterPartitions(popts));
 
   return Status::OK();
 }
diff --git a/tensorflow/core/graph/graph_partition.cc b/tensorflow/core/graph/graph_partition.cc
index bf8dcb2fcf2..71d8cdd6ab5 100644
--- a/tensorflow/core/graph/graph_partition.cc
+++ b/tensorflow/core/graph/graph_partition.cc
@@ -1165,11 +1165,16 @@ Status Partition(const PartitionOptions& opts, Graph* g,
     }
   }
 
+  const FunctionLibraryDefinition* flib_def = opts.flib_def;
+  if (flib_def == nullptr) {
+    flib_def = &g->flib_def();
+  }
+
   // Set versions, function library and send/recv incarnation.
   for (auto& it : *partitions) {
     GraphDef* gdef = &it.second;
     *gdef->mutable_versions() = g->versions();
-    *gdef->mutable_library() = g->flib_def().ToProto();
+    *gdef->mutable_library() = flib_def->ToProto();
 
     // Traverse the graph to fill every send/recv op's incarnation
     // information.
diff --git a/tensorflow/core/graph/graph_partition.h b/tensorflow/core/graph/graph_partition.h
index 8820b8821f2..67fafddd519 100644
--- a/tensorflow/core/graph/graph_partition.h
+++ b/tensorflow/core/graph/graph_partition.h
@@ -21,6 +21,7 @@ limitations under the License.
 #include <unordered_map>
 #include <vector>
 
+#include "tensorflow/core/framework/function.h"
 #include "tensorflow/core/framework/graph.pb.h"
 #include "tensorflow/core/graph/costmodel.h"
 #include "tensorflow/core/graph/graph.h"
@@ -45,6 +46,10 @@ struct PartitionOptions {
   typedef std::function<uint64(const string&)> GetIncarnationFunc;
   GetIncarnationFunc get_incarnation = nullptr;
 
+  // If specified, flib_def defines a function library that should be
+  // partitioned and replicated into each resulting partition graphs.
+  const FunctionLibraryDefinition* flib_def = nullptr;
+
   // True if all the control flow "code" has already been added. The
   // control flow code needs to be added when we still have the entire
   // graph before any partitioning. So this flag should be false for