Fix load_v1_in_v2 to support sparse tensor outputs.

PiperOrigin-RevId: 268195564

tensorflow/python/saved_model/load_v1_in_v2.py

@@ -26,6 +26,7 @@ from tensorflow.python.eager import wrap_function
 from tensorflow.python.framework import constant_op
 from tensorflow.python.framework import dtypes
 from tensorflow.python.framework import ops
+from tensorflow.python.framework import sparse_tensor
 from tensorflow.python.ops import array_ops
 from tensorflow.python.platform import tf_logging as logging
 from tensorflow.python.saved_model import function_deserialization
@@ -127,12 +128,26 @@ class _EagerSavedModelLoader(loader_impl.SavedModelLoader):
     signature_functions = {}
     for signature_key, signature_def in meta_graph_def.signature_def.items():
       if signature_def.inputs:
-        input_names, input_specs = zip(*signature_def.inputs.items())
+        original_input_names, input_specs = zip(*signature_def.inputs.items())
       else:
-        input_names = []
+        original_input_names = []
         input_specs = []
       # TODO(allenl): Support optional arguments
-      feeds = [wrapped.graph.as_graph_element(inp.name) for inp in input_specs]
+      feeds = [
+          wrap_function._get_element_from_tensor_info(input_spec, wrapped.graph)  # pylint: disable=protected-access
+          for input_spec in input_specs
+      ]
+      input_names = []
+      for original_input_name, feed in zip(original_input_names, feeds):
+        if isinstance(feed, sparse_tensor.SparseTensor):
+          # We have to give explicit name for SparseTensor arguments, because
+          # these are not present in the TensorInfo.
+          indices_name = "%s_indices" % original_input_name
+          values_name = "%s_values" % original_input_name
+          dense_shape_name = "%s_dense_shape" % original_input_name
+          input_names.extend([indices_name, values_name, dense_shape_name])
+        else:
+          input_names.append(original_input_name)
       fetches = {name: out for name, out in signature_def.outputs.items()}
       try:
         signature_fn = wrapped.prune(feeds=feeds, fetches=fetches)

tensorflow/python/saved_model/load_v1_in_v2_test.py

@@ -529,6 +529,38 @@ class LoadTest(test.TestCase):
     forty_two = constant_op.constant([42], dtype=dtypes.int64)
     self.assertEqual([84], imported_fn(forty_two)["output"].values.numpy())
 
+  def _model_with_sparse_input(self):
+    """Generate a graph with a SparseTensor input and serialize in V1 format."""
+    export_graph = ops.Graph()
+    with export_graph.as_default():
+      in_sparse_placeholder = array_ops.sparse_placeholder(
+          dtype=dtypes.int64, shape=[2, 2])
+      out_sparse_tensor = sparse_tensor.SparseTensor(
+          indices=in_sparse_placeholder.indices,
+          values=in_sparse_placeholder.values,
+          dense_shape=in_sparse_placeholder.dense_shape) * 2
+      with session_lib.Session() as session:
+        path = os.path.join(self.get_temp_dir(), "saved_model", str(ops.uid()))
+        simple_save.simple_save(
+            session,
+            path,
+            inputs={"start": in_sparse_placeholder},
+            outputs={"output": out_sparse_tensor})
+    return path
+
+  def test_load_sparse_inputs(self):
+    path = self._model_with_sparse_input()
+    imported = load.load(path)
+    imported_fn = imported.signatures["serving_default"]
+    indices = constant_op.constant([[0, 0], [0, 1], [1, 1]], dtype=dtypes.int64)
+    values = constant_op.constant([42, 43, 44], dtype=dtypes.int64)
+    dense_shape = constant_op.constant([2, 2], dtype=dtypes.int64)
+    result = imported_fn(
+        start_indices=indices,
+        start_values=values,
+        start_dense_shape=dense_shape)
+    self.assertAllEqual([84, 86, 88], result["output"].values.numpy())
+
   def _model_with_defun(self):
     """Generate a graph with a Defun and serialize in V1 format."""
     export_graph = ops.Graph()