diff --git a/tensorflow/core/kernels/data/model_dataset_op.cc b/tensorflow/core/kernels/data/model_dataset_op.cc
index f790b4bf07fe3870ab946d49c9586092f4f24939..fdf8aebc3abcc56cc58865699e0977c8850f8229 100644
--- a/tensorflow/core/kernels/data/model_dataset_op.cc
+++ b/tensorflow/core/kernels/data/model_dataset_op.cc
@@ -140,7 +140,7 @@ class ModelDatasetOp::Dataset : public DatasetBase {
       IteratorContext::Params params(ctx);
       {
         mutex_lock l(mu_);
-        TF_RETURN_IF_ERROR(EnsureOptimizeThreadStarted(ctx));
+        TF_RETURN_IF_ERROR(EnsureModelThreadStarted(ctx));
         params.model = model_;
         int64 now_nanos = EnvTime::NowNanos();
         RecordInput(now_nanos);
@@ -175,18 +175,16 @@ class ModelDatasetOp::Dataset : public DatasetBase {
     }
 
    private:
-    Status EnsureOptimizeThreadStarted(IteratorContext* ctx)
+    Status EnsureModelThreadStarted(IteratorContext* ctx)
         TF_EXCLUSIVE_LOCKS_REQUIRED(mu_) {
       if (!model_thread_) {
-        std::shared_ptr<IteratorContext> new_ctx =
-            std::make_shared<IteratorContext>(*ctx);
-        model_thread_ = ctx->StartThread(
-            "tf_data_model", [this, new_ctx]() { ModelThread(new_ctx); });
+        model_thread_ =
+            ctx->StartThread("tf_data_model", [this]() { ModelThread(); });
       }
       return Status::OK();
     }
 
-    void ModelThread(const std::shared_ptr<IteratorContext>& ctx) {
+    void ModelThread() {
       int64 last_optimization_ms = 0;
       int64 optimization_period_ms = 10;
       int64 current_time_ms = EnvTime::NowMicros() / EnvTime::kMillisToMicros;
diff --git a/tensorflow/python/data/experimental/ops/distribute.py b/tensorflow/python/data/experimental/ops/distribute.py
index 568c01646de4f2867e2b3896d5f828842886dd1e..a65a9d79340d626e56aff1e3b61e30732afdce98 100644
--- a/tensorflow/python/data/experimental/ops/distribute.py
+++ b/tensorflow/python/data/experimental/ops/distribute.py
@@ -330,6 +330,14 @@ def replicate(dataset, devices):
     return datasets
 
   with ops.colocate_with(dataset._variant_tensor):
+    # We apply options before replicating the dataset because options are
+    # currently not automatically preserved through dataset serialization and
+    # thus an explicit application of options here is needed to avoid losing
+    # `dataset` options.
+    #
+    # TODO(b/147325552): Propagating options to C++ upon their setting would
+    # allow us to preserve the options across both variant and GraphDef based
+    # serialization, avoiding the need to explicitly apply options here.
     dataset = dataset._apply_options()
     policy = dataset.options().experimental_external_state_policy
     if policy is None: