TF version compatibility fixes (#23663)

* New TF version compatibility fixes

* Remove dummy print statement, move expand_1d

* Make a proper framework inference function

* Make a proper framework inference function

* ValueError -> TypeError

src/transformers/modeling_tf_utils.py

@@ -38,7 +38,7 @@ from .activations_tf import get_tf_activation
 from .configuration_utils import PretrainedConfig
 from .dynamic_module_utils import custom_object_save
 from .generation import GenerationConfig, TFGenerationMixin
-from .tf_utils import shape_list
+from .tf_utils import expand_1d, load_attributes_from_hdf5_group, save_attributes_to_hdf5_group, shape_list
 from .utils import (
     DUMMY_INPUTS,
     SAFE_WEIGHTS_INDEX_NAME,
@@ -65,16 +65,15 @@ from .utils import (
 from .utils.hub import convert_file_size_to_int, get_checkpoint_shard_files
 
 
-if parse(tf.__version__) >= parse("2.11.0"):
+if parse(tf.__version__).minor >= 13:
+    from keras import backend as K
+    from keras.__internal__ import KerasTensor
+elif parse(tf.__version__).minor >= 11:
     from keras import backend as K
-    from keras.engine import data_adapter
     from keras.engine.keras_tensor import KerasTensor
-    from keras.saving.legacy import hdf5_format
 else:
     from tensorflow.python.keras import backend as K
-    from tensorflow.python.keras.engine import data_adapter
     from tensorflow.python.keras.engine.keras_tensor import KerasTensor
-    from tensorflow.python.keras.saving import hdf5_format
 
 
 if is_safetensors_available():
@@ -797,9 +796,7 @@ def load_tf_shard(model, model_layer_map, resolved_archive_file, ignore_mismatch
     try:
         with h5py.File(resolved_archive_file, "r") as sharded_checkpoint_file:
             # Retrieve the name of each layer from the H5 file
-            saved_h5_model_layers_name = set(
+            saved_h5_model_layers_name = set(load_attributes_from_hdf5_group(sharded_checkpoint_file, "layer_names"))
-                hdf5_format.load_attributes_from_hdf5_group(sharded_checkpoint_file, "layer_names")
-            )
             weight_value_tuples = []
 
             # Compute missing and unexpected sub layers
@@ -898,9 +895,7 @@ def load_tf_weights_from_h5(model, resolved_archive_file, ignore_mismatched_size
     # Read the H5 file
     with h5py.File(resolved_archive_file, "r") as sharded_checkpoint_file:
         # Retrieve the name of each layer from the H5 file
-        saved_h5_model_layers_name = set(
+        saved_h5_model_layers_name = set(load_attributes_from_hdf5_group(sharded_checkpoint_file, "layer_names"))
-            hdf5_format.load_attributes_from_hdf5_group(sharded_checkpoint_file, "layer_names")
-        )
 
         # Find the missing layers from the high level list of layers
         missing_layers = list({layer.name for layer in model.layers} - saved_h5_model_layers_name)
@@ -924,7 +919,7 @@ def load_tf_weights_from_h5(model, resolved_archive_file, ignore_mismatched_size
 
                 # Create a dict from the H5 saved model that looks like {"weight_name": weight_value}
                 # And a set with only the names
-                for weight_name in hdf5_format.load_attributes_from_hdf5_group(h5_layer_object, "weight_names"):
+                for weight_name in load_attributes_from_hdf5_group(h5_layer_object, "weight_names"):
                     # TF names always start with the model name so we ignore it
                     name = "/".join(weight_name.split("/")[1:])
 
@@ -1528,8 +1523,8 @@ class TFPreTrainedModel(tf.keras.Model, TFModelUtilsMixin, TFGenerationMixin, Pu
         output_to_label = {val: key for key, val in label_to_output.items()}
         if not self._using_dummy_loss and parse(tf.__version__) < parse("2.11.0"):
             # Newer TF train steps leave this out
-            data = data_adapter.expand_1d(data)
+            data = expand_1d(data)
-        x, y, sample_weight = data_adapter.unpack_x_y_sample_weight(data)
+        x, y, sample_weight = tf.keras.utils.unpack_x_y_sample_weight(data)
         # If the inputs are mutable dictionaries, make a shallow copy of them because we will modify
         # them during input/label pre-processing. This avoids surprising the user by wrecking their data.
         # In addition, modifying mutable Python inputs makes XLA compilation impossible.
@@ -1635,8 +1630,8 @@ class TFPreTrainedModel(tf.keras.Model, TFModelUtilsMixin, TFGenerationMixin, Pu
         output_to_label = {val: key for key, val in label_to_output.items()}
         if not self._using_dummy_loss and parse(tf.__version__) < parse("2.11.0"):
             # Newer versions leave this out
-            data = data_adapter.expand_1d(data)
+            data = expand_1d(data)
-        x, y, sample_weight = data_adapter.unpack_x_y_sample_weight(data)
+        x, y, sample_weight = tf.keras.utils.unpack_x_y_sample_weight(data)
         # If the inputs are mutable dictionaries, make a shallow copy of them because we will modify
         # them during input/label pre-processing. This avoids surprising the user by wrecking their data.
         # In addition, modifying mutable Python inputs makes XLA compilation impossible.
@@ -2402,7 +2397,7 @@ class TFPreTrainedModel(tf.keras.Model, TFModelUtilsMixin, TFGenerationMixin, Pu
                         )
                         param_dset[:] = layer.numpy()
                         layers.append(layer_name.encode("utf8"))
-                    hdf5_format.save_attributes_to_hdf5_group(shard_file, "layer_names", layers)
+                    save_attributes_to_hdf5_group(shard_file, "layer_names", layers)
 
         if push_to_hub:
             self._upload_modified_files(

src/transformers/pipelines/base.py

@@ -15,7 +15,6 @@
 import collections
 import csv
 import importlib
-import inspect
 import json
 import os
 import pickle
@@ -36,7 +35,7 @@ from ..image_processing_utils import BaseImageProcessor
 from ..modelcard import ModelCard
 from ..models.auto.configuration_auto import AutoConfig
 from ..tokenization_utils import PreTrainedTokenizer
-from ..utils import ModelOutput, add_end_docstrings, is_tf_available, is_torch_available, logging
+from ..utils import ModelOutput, add_end_docstrings, infer_framework, is_tf_available, is_torch_available, logging
 
 
 GenericTensor = Union[List["GenericTensor"], "torch.Tensor", "tf.Tensor"]
@@ -278,7 +277,7 @@ def infer_framework_load_model(
         if isinstance(model, str):
             raise ValueError(f"Could not load model {model} with any of the following classes: {class_tuple}.")
 
-    framework = "tf" if "keras.engine.training.Model" in str(inspect.getmro(model.__class__)) else "pt"
+    framework = infer_framework(model.__class__)
     return framework, model
 
 
@@ -351,7 +350,7 @@ def get_framework(model, revision: Optional[str] = None):
             except OSError:
                 model = TFAutoModel.from_pretrained(model, revision=revision)
 
-    framework = "tf" if "keras.engine.training.Model" in str(inspect.getmro(model.__class__)) else "pt"
+    framework = infer_framework(model.__class__)
     return framework
 
 

src/transformers/tf_utils.py

@@ -166,3 +166,90 @@ def check_embeddings_within_bounds(tensor: tf.Tensor, embed_dim: int, tensor_nam
             f"layer's input dimension ({embed_dim}). The likely cause is some problem at tokenization time."
         ),
     )
+
+
+def save_attributes_to_hdf5_group(group, name, data):
+    """Saves attributes (data) of the specified name into the HDF5 group.
+
+    This method deals with an inherent problem of HDF5 file which is not able to store data larger than
+    HDF5_OBJECT_HEADER_LIMIT bytes.
+
+    Args:
+        group: A pointer to a HDF5 group.
+        name: A name of the attributes to save.
+        data: Attributes data to store.
+
+    Raises:
+      RuntimeError: If any single attribute is too large to be saved.
+
+    Copied from Keras to Transformers to avoid versioning issues.
+    """
+    HDF5_OBJECT_HEADER_LIMIT = 64512
+    # Check that no item in `data` is larger than `HDF5_OBJECT_HEADER_LIMIT`
+    # because in that case even chunking the array would not make the saving
+    # possible.
+    bad_attributes = [x for x in data if len(x) > HDF5_OBJECT_HEADER_LIMIT]
+
+    # Expecting this to never be true.
+    if bad_attributes:
+        raise RuntimeError(
+            "The following attributes cannot be saved to HDF5 file because "
+            f"they are larger than {HDF5_OBJECT_HEADER_LIMIT} "
+            f"bytes: {bad_attributes}"
+        )
+
+    data_npy = np.asarray(data)
+
+    num_chunks = 1
+    chunked_data = np.array_split(data_npy, num_chunks)
+
+    # This will never loop forever thanks to the test above.
+    while any(x.nbytes > HDF5_OBJECT_HEADER_LIMIT for x in chunked_data):
+        num_chunks += 1
+        chunked_data = np.array_split(data_npy, num_chunks)
+
+    if num_chunks > 1:
+        for chunk_id, chunk_data in enumerate(chunked_data):
+            group.attrs["%s%d" % (name, chunk_id)] = chunk_data
+    else:
+        group.attrs[name] = data
+
+
+def load_attributes_from_hdf5_group(group, name):
+    """Loads attributes of the specified name from the HDF5 group.
+
+    This method deals with an inherent problem of HDF5 file which is not able to store data larger than
+    HDF5_OBJECT_HEADER_LIMIT bytes.
+
+    Args:
+        group: A pointer to a HDF5 group.
+        name: A name of the attributes to load.
+
+    Returns:
+        data: Attributes data.
+
+    Copied from Keras to Transformers to avoid versioning issues.
+    """
+    if name in group.attrs:
+        data = [n.decode("utf8") if hasattr(n, "decode") else n for n in group.attrs[name]]
+    else:
+        data = []
+        chunk_id = 0
+        while "%s%d" % (name, chunk_id) in group.attrs:
+            data.extend(
+                [n.decode("utf8") if hasattr(n, "decode") else n for n in group.attrs["%s%d" % (name, chunk_id)]]
+            )
+            chunk_id += 1
+    return data
+
+
+def expand_1d(data):
+    """Expands 1-dimensional `Tensor`s into 2-dimensional `Tensor`s.
+    Copied from Keras to here to avoid versioning issues."""
+
+    def _expand_single_1d_tensor(t):
+        if isinstance(t, tf.Tensor) and t.shape.rank == 1:
+            return tf.expand_dims(t, axis=-1)
+        return t
+
+    return tf.nest.map_structure(_expand_single_1d_tensor, data)

src/transformers/utils/__init__.py

@@ -39,6 +39,7 @@ from .generic import (
     expand_dims,
     find_labels,
     flatten_dict,
+    infer_framework,
     is_jax_tensor,
     is_numpy_array,
     is_tensor,

src/transformers/utils/generic.py

@@ -398,11 +398,10 @@ def can_return_loss(model_class):
     Args:
         model_class (`type`): The class of the model.
     """
-    base_classes = str(inspect.getmro(model_class))
+    framework = infer_framework(model_class)
-
+    if framework == "tf":
-    if "keras.engine.training.Model" in base_classes:
         signature = inspect.signature(model_class.call)  # TensorFlow models
-    elif "torch.nn.modules.module.Module" in base_classes:
+    elif framework == "pt":
         signature = inspect.signature(model_class.forward)  # PyTorch models
     else:
         signature = inspect.signature(model_class.__call__)  # Flax models
@@ -422,11 +421,10 @@ def find_labels(model_class):
         model_class (`type`): The class of the model.
     """
     model_name = model_class.__name__
-    base_classes = str(inspect.getmro(model_class))
+    framework = infer_framework(model_class)
-
+    if framework == "tf":
-    if "keras.engine.training.Model" in base_classes:
         signature = inspect.signature(model_class.call)  # TensorFlow models
-    elif "torch.nn.modules.module.Module" in base_classes:
+    elif framework == "pt":
         signature = inspect.signature(model_class.forward)  # PyTorch models
     else:
         signature = inspect.signature(model_class.__call__)  # Flax models
@@ -565,3 +563,21 @@ def add_model_info_to_auto_map(auto_map, repo_id):
             auto_map[key] = f"{repo_id}--{value}"
 
     return auto_map
+
+
+def infer_framework(model_class):
+    """
+    Infers the framework of a given model without using isinstance(), because we cannot guarantee that the relevant
+    classes are imported or available.
+    """
+    for base_class in inspect.getmro(model_class):
+        module = base_class.__module__
+        name = base_class.__name__
+        if module.startswith("tensorflow") or module.startswith("keras") or name == "TFPreTrainedModel":
+            return "tf"
+        elif module.startswith("torch") or name == "PreTrainedModel":
+            return "pt"
+        elif module.startswith("flax") or module.startswith("jax") or name == "FlaxPreTrainedModel":
+            return "flax"
+    else:
+        raise TypeError(f"Could not infer framework from class {model_class}.")