Replace dm-tree with optree (#19306)

* Refactor `keras.utils.tree`

* Fix tests

* Replace `dm-tree` with `optree`

* Eliminate `tf.nest`

* Resolve comments

* Fix merge conflicts

* Update exporting path

Author: james77777778

keras/backend/tensorflow/core.py

@@ -9,6 +9,7 @@
 from keras.backend.common.name_scope import name_scope as base_name_scope
 from keras.backend.common.stateless_scope import StatelessScope
 from keras.backend.common.stateless_scope import in_stateless_scope
+from keras.utils import tree
 from keras.utils.naming import auto_name
 
 SUPPORTS_SPARSE_TENSORS = True
@@ -189,7 +190,7 @@ def convert_keras_tensor_to_tf(x):
                         )
                 return x
 
-            args, kwargs = tf.nest.map_structure(
+            args, kwargs = tree.map_structure(
                 convert_keras_tensor_to_tf, (args, kwargs)
             )
             tf_out = fn(*args, **kwargs)
@@ -201,9 +202,7 @@ def convert_tf_to_keras_tensor(x):
                     )
                 return x
 
-            output_spec = tf.nest.map_structure(
-                convert_tf_to_keras_tensor, tf_out
-            )
+            output_spec = tree.map_structure(convert_tf_to_keras_tensor, tf_out)
     return output_spec
 
 

keras/backend/tensorflow/layer.py

@@ -3,6 +3,7 @@
 from keras.backend.tensorflow.trackable import KerasAutoTrackable
 from keras.utils import tf_utils
 from keras.utils import tracking
+from keras.utils import tree
 
 
 class TFLayer(KerasAutoTrackable):
@@ -27,16 +28,16 @@ def _set_save_spec(self, inputs, args=None, kwargs=None):
         if self._saved_model_inputs_spec is not None:
             return  # Already set.
 
-        inputs_spec = tf.nest.map_structure(tf_utils.get_tensor_spec, inputs)
-        args_spec = tf.nest.map_structure(tf_utils.get_tensor_spec, args or [])
+        inputs_spec = tree.map_structure(tf_utils.get_tensor_spec, inputs)
+        args_spec = tree.map_structure(tf_utils.get_tensor_spec, args or [])
         kwargs_spec = {}
         # Filter out non-tensor arguments from kwargs.
         for key, kwarg in kwargs.items():
-            flat_kwarg = tf.nest.flatten(kwarg)
+            flat_kwarg = tree.flatten(kwarg)
             flat_specs = [tf_utils.get_tensor_spec(x) for x in flat_kwarg]
             if any(s is None for s in flat_specs):
                 continue
-            kwargs_spec[key] = tf.nest.pack_sequence_as(kwarg, flat_specs)
+            kwargs_spec[key] = tree.pack_sequence_as(kwarg, flat_specs)
 
         self._saved_model_inputs_spec = inputs_spec
         self._saved_model_arg_spec = (
@@ -94,7 +95,7 @@ def _default_save_signature(self):
 
         if inputs is not None:
             input_signature = [
-                tf.nest.map_structure(
+                tree.map_structure(
                     lambda x: tf.TensorSpec(x.shape, self.compute_dtype),
                     inputs,
                 )
@@ -108,7 +109,7 @@ def _default_save_signature(self):
                 ]
             else:
                 input_signature = [
-                    tf.nest.map_structure(
+                    tree.map_structure(
                         lambda x: tf.TensorSpec(x.shape, self.compute_dtype),
                         shapes_dict,
                     )

keras/backend/tensorflow/numpy.py

@@ -17,6 +17,7 @@
 from keras.backend.common.backend_utils import to_tuple_or_list
 from keras.backend.tensorflow import sparse
 from keras.backend.tensorflow.core import convert_to_tensor
+from keras.utils import tree
 
 
 @sparse.elementwise_binary_union(tf.sparse.add)
@@ -95,7 +96,7 @@ def _normalize_einsum_subscripts(subscripts):
 
 
 def einsum(subscripts, *operands, **kwargs):
-    operands = tf.nest.map_structure(convert_to_tensor, operands)
+    operands = tree.map_structure(convert_to_tensor, operands)
     subscripts = _normalize_einsum_subscripts(subscripts)
 
     def is_valid_for_custom_ops(subscripts, *operands):
@@ -240,15 +241,15 @@ def use_custom_ops(subscripts, *operands, output_type):
         # output_type="int32"
         if "int" in compute_dtype and output_type is None:
             compute_dtype = config.floatx()
-        operands = tf.nest.map_structure(
+        operands = tree.map_structure(
             lambda x: tf.cast(x, compute_dtype), operands
         )
         result = use_custom_ops(subscripts, *operands, output_type=output_type)
     else:
         # TODO: tf.einsum doesn't support integer dtype with gpu
         if "int" in compute_dtype:
             compute_dtype = config.floatx()
-        operands = tf.nest.map_structure(
+        operands = tree.map_structure(
             lambda x: tf.cast(x, compute_dtype), operands
         )
         result = tf.einsum(subscripts, *operands, **kwargs)
@@ -763,11 +764,11 @@ def concatenate(xs, axis=0):
                 )
                 for x in xs
             ]
-    xs = tf.nest.map_structure(convert_to_tensor, xs)
+    xs = tree.map_structure(convert_to_tensor, xs)
     dtype_set = set([x.dtype for x in xs])
     if len(dtype_set) > 1:
         dtype = dtypes.result_type(*dtype_set)
-        xs = tf.nest.map_structure(lambda x: tf.cast(x, dtype), xs)
+        xs = tree.map_structure(lambda x: tf.cast(x, dtype), xs)
     return tf.concat(xs, axis=axis)
 
 
@@ -872,7 +873,7 @@ def digitize(x, bins):
     bins = list(bins)
 
     # bins must be float type
-    bins = tf.nest.map_structure(lambda x: float(x), bins)
+    bins = tree.map_structure(lambda x: float(x), bins)
 
     # TODO: tf.raw_ops.Bucketize doesn't support bool, bfloat16, float16, int8
     # int16, uint8, uint16, uint32
@@ -1023,7 +1024,7 @@ def hstack(xs):
     dtype_set = set([getattr(x, "dtype", type(x)) for x in xs])
     if len(dtype_set) > 1:
         dtype = dtypes.result_type(*dtype_set)
-        xs = tf.nest.map_structure(lambda x: convert_to_tensor(x, dtype), xs)
+        xs = tree.map_structure(lambda x: convert_to_tensor(x, dtype), xs)
     rank = tf.rank(xs[0])
     return tf.cond(
         tf.equal(rank, 1),
@@ -1328,9 +1329,7 @@ def ndim(x):
 def nonzero(x):
     x = convert_to_tensor(x)
     result = tf.unstack(tf.where(tf.cast(x, "bool")), x.shape.rank, axis=1)
-    return tf.nest.map_structure(
-        lambda indices: tf.cast(indices, "int32"), result
-    )
+    return tree.map_structure(lambda indices: tf.cast(indices, "int32"), result)
 
 
 def not_equal(x1, x2):
@@ -1620,7 +1619,7 @@ def stack(x, axis=0):
     dtype_set = set([getattr(a, "dtype", type(a)) for a in x])
     if len(dtype_set) > 1:
         dtype = dtypes.result_type(*dtype_set)
-        x = tf.nest.map_structure(lambda a: convert_to_tensor(a, dtype), x)
+        x = tree.map_structure(lambda a: convert_to_tensor(a, dtype), x)
     return tf.stack(x, axis=axis)
 
 
@@ -1807,7 +1806,7 @@ def vstack(xs):
     dtype_set = set([getattr(x, "dtype", type(x)) for x in xs])
     if len(dtype_set) > 1:
         dtype = dtypes.result_type(*dtype_set)
-        xs = tf.nest.map_structure(lambda x: convert_to_tensor(x, dtype), xs)
+        xs = tree.map_structure(lambda x: convert_to_tensor(x, dtype), xs)
     return tf.concat(xs, axis=0)
 
 

keras/backend/tensorflow/trainer.py

@@ -225,7 +225,7 @@ def multi_step_on_data(data):
             outputs = one_step_on_data_distributed(data[:1])
             for single_step_data in data[1:]:
                 step_outputs = one_step_on_data_distributed([single_step_data])
-                outputs = tf.nest.map_structure(
+                outputs = tree.map_structure(
                     lambda t1, t2: concat([t1, t2]), outputs, step_outputs
                 )
             return outputs
@@ -473,7 +473,7 @@ def predict(
 
         def append_to_outputs(batch_outputs, outputs):
             if outputs is None:
-                outputs = tf.nest.map_structure(
+                outputs = tree.map_structure(
                     lambda batch_output: [batch_output],
                     batch_outputs,
                 )
@@ -521,7 +521,7 @@ def get_data(iterator):
         outputs = tree.map_structure_up_to(
             batch_outputs, potentially_ragged_concat, outputs
         )
-        return tf.nest.map_structure(convert_to_np_if_not_ragged, outputs)
+        return tree.map_structure(convert_to_np_if_not_ragged, outputs)
 
     def train_on_batch(
         self,
@@ -549,7 +549,7 @@ def data():
             yield (x, y, sample_weight)
 
         logs = self.train_function(data())
-        logs = tf.nest.map_structure(lambda x: np.array(x), logs)
+        logs = tree.map_structure(lambda x: np.array(x), logs)
         if return_dict:
             return logs
         return self._flatten_metrics_in_order(logs)
@@ -568,15 +568,15 @@ def data():
             yield (x, y, sample_weight)
 
         logs = self.test_function(data())
-        logs = tf.nest.map_structure(lambda x: np.array(x), logs)
+        logs = tree.map_structure(lambda x: np.array(x), logs)
         if return_dict:
             return logs
         return self._flatten_metrics_in_order(logs)
 
     def predict_on_batch(self, x):
         self.make_predict_function()
         batch_outputs = self.predict_function([(x,)])
-        batch_outputs = tf.nest.map_structure(
+        batch_outputs = tree.map_structure(
             convert_to_np_if_not_ragged, batch_outputs
         )
         return batch_outputs
@@ -771,7 +771,7 @@ def _reduce(v):
                 f"Received: reduction={reduction}."
             )
 
-    return tf.nest.map_structure(_reduce, values)
+    return tree.map_structure(_reduce, values)
 
 
 def _multi_worker_concat(v, strategy):

keras/export/export_lib.py

@@ -8,6 +8,7 @@
 from keras.models import Functional
 from keras.models import Sequential
 from keras.utils import io_utils
+from keras.utils import tree
 from keras.utils.module_utils import tensorflow as tf
 
 
@@ -143,16 +144,16 @@ def track(self, resource):
             # Variables in the lists below are actually part of the trackables
             # that get saved, because the lists are created in __init__.
             if backend.backend() == "jax":
-                self._tf_trackable.variables += tf.nest.flatten(
-                    tf.nest.map_structure(tf.Variable, resource.variables)
+                self._tf_trackable.variables += tree.flatten(
+                    tree.map_structure(tf.Variable, resource.variables)
                 )
-                self._tf_trackable.trainable_variables += tf.nest.flatten(
-                    tf.nest.map_structure(
+                self._tf_trackable.trainable_variables += tree.flatten(
+                    tree.map_structure(
                         tf.Variable, resource.trainable_variables
                     )
                 )
-                self._tf_trackable.non_trainable_variables += tf.nest.flatten(
-                    tf.nest.map_structure(
+                self._tf_trackable.non_trainable_variables += tree.flatten(
+                    tree.map_structure(
                         tf.Variable, resource.non_trainable_variables
                     )
                 )
@@ -362,9 +363,7 @@ def add_variable_collection(self, name, variables):
                 f"{list(set(type(v) for v in variables))}"
             )
         if backend.backend() == "jax":
-            variables = tf.nest.flatten(
-                tf.nest.map_structure(tf.Variable, variables)
-            )
+            variables = tree.flatten(tree.map_structure(tf.Variable, variables))
         setattr(self._tf_trackable, name, list(variables))
 
     def write_out(self, filepath, options=None):
@@ -470,7 +469,7 @@ def _convert_jax2tf_function(self, fn, input_signature):
 
     def _spec_to_poly_shape(self, spec):
         if isinstance(spec, (dict, list)):
-            return tf.nest.map_structure(self._spec_to_poly_shape, spec)
+            return tree.map_structure(self._spec_to_poly_shape, spec)
         spec_shape = spec.shape
         spec_shape = str(spec_shape).replace("None", "b")
         return spec_shape
@@ -500,7 +499,7 @@ def export_model(model, filepath):
     export_archive = ExportArchive()
     export_archive.track(model)
     if isinstance(model, (Functional, Sequential)):
-        input_signature = tf.nest.map_structure(_make_tensor_spec, model.inputs)
+        input_signature = tree.map_structure(_make_tensor_spec, model.inputs)
         if isinstance(input_signature, list) and len(input_signature) > 1:
             input_signature = [input_signature]
         export_archive.add_endpoint("serve", model.__call__, input_signature)

keras/models/cloning.py

@@ -261,7 +261,7 @@ def _clone_layer(layer):
             )
         try:
             tree.assert_same_structure(input_tensors, model.input)
-        except TypeError as e:
+        except (ValueError, TypeError) as e:
             raise ValueError(
                 "`input_tensors` must have the same structure as model.input"
                 f"\nReference structure: {model.input}"

keras/ops/core_test.py

@@ -789,9 +789,8 @@ def test_cond_check_output_spec_list_tuple(self):
 
     def test_cond_check_output_spec_other_types(self):
         cond_op = core.Cond()
-        # Create mock objects with dtype and shape attributes
-        mock_spec1 = Mock(dtype="float32", shape=(2, 2))
-        mock_spec2 = Mock(dtype="float32", shape=(2, 2))
+        mock_spec1 = KerasTensor(shape=(2, 2), dtype="float32")
+        mock_spec2 = KerasTensor(shape=(2, 2), dtype="float32")
         self.assertTrue(cond_op._check_output_spec(mock_spec1, mock_spec2))
 
     def test_cond_check_output_spec_none(self):

keras/utils/tracking.py

@@ -1,5 +1,8 @@
 from functools import wraps
 
+import optree
+import optree.utils
+
 from keras.backend.common.global_state import get_global_attribute
 from keras.backend.common.global_state import set_global_attribute
 from keras.utils import python_utils
@@ -110,6 +113,7 @@ def add_to_store(self, store_name, value):
         self.stored_ids[store_name].add(id(value))
 
 
+@optree.register_pytree_node_class(namespace="keras")
 class TrackedList(list):
     def __init__(self, values=None, tracker=None):
         self.tracker = tracker
@@ -160,7 +164,17 @@ def __delitem__(self, index):
         if self.tracker:
             self.tracker.untrack(value)
 
+    def tree_flatten(self):
+        # For optree
+        return (self, None)
+
+    @classmethod
+    def tree_unflatten(cls, metadata, children):
+        # For optree
+        return cls(children)
 
+
+@optree.register_pytree_node_class(namespace="keras")
 class TrackedDict(dict):
     def __init__(self, values=None, tracker=None):
         self.tracker = tracker
@@ -199,7 +213,20 @@ def clear(self):
                 self.tracker.untrack(value)
         super().clear()
 
+    def tree_flatten(self):
+        # For optree
+        keys, values = optree.utils.unzip2(
+            optree.utils.total_order_sorted(self.items(), key=lambda kv: kv[0])
+        )
+        return values, list(keys), keys
+
+    @classmethod
+    def tree_unflatten(cls, keys, values):
+        # For optree
+        return cls(optree.utils.safe_zip(keys, values))
+
 
+@optree.register_pytree_node_class(namespace="keras")
 class TrackedSet(set):
     def __init__(self, values=None, tracker=None):
         self.tracker = tracker
@@ -233,3 +260,12 @@ def clear(self):
             for value in self:
                 self.tracker.untrack(value)
         super().clear()
+
+    def tree_flatten(self):
+        # For optree
+        return (self, None)
+
+    @classmethod
+    def tree_unflatten(cls, metadata, children):
+        # For optree
+        return cls(children)