diff --git a/torch_xla/csrc/runtime/BUILD b/torch_xla/csrc/runtime/BUILD
index e6f064158ab..4feb4154166 100644
--- a/torch_xla/csrc/runtime/BUILD
+++ b/torch_xla/csrc/runtime/BUILD
@@ -302,6 +302,7 @@ cc_library(
   name = "tensor_source",
   hdrs = ["tensor_source.h"],
   deps = [
+      ":debug_macros",
       "@xla//xla:literal",
       "@xla//xla:shape_util",
       "@torch//:headers",
diff --git a/torch_xla/csrc/runtime/pjrt_computation_client.cc b/torch_xla/csrc/runtime/pjrt_computation_client.cc
index de158582615..cbe1937df00 100644
--- a/torch_xla/csrc/runtime/pjrt_computation_client.cc
+++ b/torch_xla/csrc/runtime/pjrt_computation_client.cc
@@ -271,16 +271,13 @@ std::vector<ComputationClient::DataPtr> PjRtComputationClient::TransferToDevice(
   for (auto& tensor : tensors) {
     xla::PjRtDevice* pjrt_device = StringToPjRtDevice(tensor->device());
 
-    std::vector<int64_t> byte_strides(tensor->shape().dimensions_size());
-    XLA_CHECK_OK(xla::ShapeUtil::ByteStrides(tensor->shape(),
-                                             absl::MakeSpan(byte_strides)));
     total_size += xla::ShapeUtil::ByteSizeOf(tensor->shape());
 
     std::shared_ptr<xla::PjRtBuffer> buffer =
         std::move(client_
                       ->BufferFromHostBuffer(
                           tensor->data(), tensor->shape().element_type(),
-                          tensor->shape().dimensions(), byte_strides,
+                          tensor->shape().dimensions(), tensor->byte_strides(),
                           xla::PjRtClient::HostBufferSemantics::
                               kImmutableUntilTransferCompletes,
                           [tensor]() { /* frees tensor */ }, pjrt_device)
diff --git a/torch_xla/csrc/runtime/tensor_source.h b/torch_xla/csrc/runtime/tensor_source.h
index 14772289d36..d3798231777 100644
--- a/torch_xla/csrc/runtime/tensor_source.h
+++ b/torch_xla/csrc/runtime/tensor_source.h
@@ -1,6 +1,7 @@
 #ifndef XLA_CLIENT_TENSOR_SOURCE_H_
 #define XLA_CLIENT_TENSOR_SOURCE_H_
 
+#include "torch_xla/csrc/runtime/debug_macros.h"
 #include "xla/literal.h"
 #include "xla/shape.h"
 
@@ -18,6 +19,13 @@ class TensorSource {
 
   const std::string& device() const { return device_; }
 
+  std::vector<int64_t> byte_strides() const {
+    std::vector<int64_t> byte_strides(shape().dimensions_size());
+    XLA_CHECK_OK(xla::ShapeUtil::ByteStrides(shape(),
+                                             absl::MakeSpan(byte_strides)));
+    return byte_strides;
+  }
+
  private:
   std::string device_;
 };