# Copyright 2022 The TensorFlow Authors. All Rights Reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # ============================================================================== """Python-based idempotent model-saving functionality.""" import datetime import io import json import os import re import tempfile import threading import warnings import zipfile import numpy as np import tensorflow.compat.v2 as tf import keras from keras import losses from keras.engine import base_layer from keras.optimizers import optimizer from keras.saving.serialization_lib import ObjectSharingScope from keras.saving.serialization_lib import deserialize_keras_object from keras.saving.serialization_lib import serialize_keras_object from keras.utils import generic_utils from keras.utils import io_utils try: import h5py except ImportError: h5py = None # isort: off _CONFIG_FILENAME = "config.json" _METADATA_FILENAME = "metadata.json" _VARS_FNAME = "model.weights" # Will become e.g. "model.weights.h5" _ASSETS_DIRNAME = "assets" # A temporary flag to enable the new idempotent saving framework. _SAVING_V3_ENABLED = threading.local() _SAVING_V3_ENABLED.value = False ATTR_SKIPLIST = frozenset( { "_callable_losses", "_captured_weight_regularizer", "_checkpoint_dependencies", "_deferred_dependencies", "_eager_losses", "_inbound_nodes", "_inbound_nodes_value", "_output_layers", "_input_layers", "_keras_api_names", "_keras_api_names_v1", "_name_based_restores", "_non_trainable_weights", "_outbound_nodes", "_outbound_nodes_value", "_saved_model_arg_spec", "_self_name_based_restores", "_self_saveable_object_factories", "_self_tracked_trackables", "_saved_model_inputs_spec", "_self_unconditional_checkpoint_dependencies", "_self_unconditional_deferred_dependencies", "_self_unconditional_dependency_names", "_tf_api_names", "_tf_api_names_v1", "_trainable_weights", "_non_trainable_weights", "_unconditional_checkpoint_dependencies", "_unconditional_dependency_names", "_updates", "_layer_call_argspecs", "inbound_nodes", "outbound_nodes", "input_shape", "output_shape", "submodules", "weights", "non_trainable_weights", "trainable_weights", "variables", "non_trainable_variables", "trainable_variables", "updates", # Would raise a warning if visited. "state_updates", # Would raise a warning if visited. } ) def save_model(model, filepath, weights_format="h5"): """Save a zip-archive representing a Keras model to the given filepath. The zip-based archive contains the following structure: - JSON-based configuration file (config.json): Records of model, layer, and other trackables' configuration. - NPZ-based trackable state files, found in respective directories, such as model/states.npz, model/dense_layer/states.npz, etc. - Metadata file. The states of Keras trackables (layers, optimizers, loss, and metrics) are automatically saved as long as they can be discovered through the attributes returned by `dir(Model)`. Typically, the state includes the variables associated with the trackable, but some specially purposed layers may contain more such as the vocabularies stored in the hashmaps. The trackables define how their states are saved by exposing `save_state()` and `load_state()` APIs. For the case of layer states, the variables will be visited as long as they are either 1) referenced via layer attributes, or 2) referenced via a container (list, tuple, or dict), and the container is referenced via a layer attribute. """ filepath = str(filepath) if not filepath.endswith(".keras"): raise ValueError( "Invalid `filepath` argument: expected a `.keras` extension. " f"Received: filepath={filepath}" ) if weights_format == "h5" and h5py is None: raise ImportError("h5py must be installed in order to save a model.") if not model.built: warnings.warn( "You are saving a model that has not yet been built. " "It might not contain any weights yet. " "Consider building the model first by calling it " "on some data.", stacklevel=2, ) saving_v3_enabled_value = getattr(_SAVING_V3_ENABLED, "value", False) _SAVING_V3_ENABLED.value = True with ObjectSharingScope(): serialized_model_dict = serialize_keras_object(model) config_json = json.dumps(serialized_model_dict) metadata_json = json.dumps( { "keras_version": keras.__version__, "date_saved": datetime.datetime.now().strftime("%Y-%m-%d@%H:%M:%S"), } ) # TODO(rameshsampath): Need a better logic for local vs remote path if re.match(r"^(/cns|/cfs|.*://).*$", filepath): # Remote path. Zip to local drive and copy to remote is_remote_path = True zip_filepath = os.path.join(_get_temp_dir(), "tmp_model.keras") else: is_remote_path = False zip_filepath = filepath try: with zipfile.ZipFile(zip_filepath, "w") as zf: with zf.open(_METADATA_FILENAME, "w") as f: f.write(metadata_json.encode()) with zf.open(_CONFIG_FILENAME, "w") as f: f.write(config_json.encode()) if weights_format == "h5": weights_store = H5IOStore( _VARS_FNAME + ".h5", archive=zf, mode="w" ) elif weights_format == "npz": weights_store = NpzIOStore( _VARS_FNAME + ".npz", archive=zf, mode="w" ) else: raise ValueError( "Unknown `weights_format` argument. " "Expected 'h5' or 'npz'. " f"Received: weights_format={weights_format}" ) asset_store = DiskIOStore(_ASSETS_DIRNAME, archive=zf, mode="w") _save_state( model, weights_store=weights_store, assets_store=asset_store, inner_path="", visited_trackables=set(), ) weights_store.close() asset_store.close() if is_remote_path: # Using tf.io.gfile context manager doesn't close zip file when # writing to GCS. Hence writing to local and copying to filepath. tf.io.gfile.copy(zip_filepath, filepath, overwrite=True) os.remove(zip_filepath) except Exception as e: raise e finally: _SAVING_V3_ENABLED.value = saving_v3_enabled_value def load_model(filepath, custom_objects=None, compile=True, safe_mode=True): """Load a zip archive representing a Keras model.""" filepath = str(filepath) if not filepath.endswith(".keras"): raise ValueError( "Invalid filename: expected a `.keras` extension. " f"Received: filepath={filepath}" ) saving_v3_enabled_value = getattr(_SAVING_V3_ENABLED, "value", False) _SAVING_V3_ENABLED.value = True try: with tf.io.gfile.GFile( filepath, mode="r+b" ) as gfile_handle, zipfile.ZipFile(gfile_handle, "r") as zf: with zf.open(_CONFIG_FILENAME, "r") as f: config_json = f.read() # Note: we should NOT use a custom JSON decoder. Anything that # needs custom decoding must be handled in deserialize_keras_object. config_dict = json.loads(config_json) if not compile: # Disable compilation config_dict["compile_config"] = None # Construct the model from the configuration file in the archive. with ObjectSharingScope(): model = deserialize_keras_object( config_dict, custom_objects, safe_mode=safe_mode ) all_filenames = zf.namelist() if _VARS_FNAME + ".h5" in all_filenames: weights_store = H5IOStore( _VARS_FNAME + ".h5", archive=zf, mode="r" ) elif _VARS_FNAME + ".npz" in all_filenames: weights_store = NpzIOStore( _VARS_FNAME + ".npz", archive=zf, mode="r" ) else: raise ValueError( f"Expected a {_VARS_FNAME}.h5 or {_VARS_FNAME}.npz file." ) if len(all_filenames) > 3: asset_store = DiskIOStore(_ASSETS_DIRNAME, archive=zf, mode="r") else: asset_store = None _load_state( model, weights_store=weights_store, assets_store=asset_store, inner_path="", visited_trackables=set(), ) weights_store.close() if asset_store: asset_store.close() except Exception as e: raise e else: return model finally: _SAVING_V3_ENABLED.value = saving_v3_enabled_value def save_weights_only(model, filepath): """Save only the weights of a model to a target filepath (.weights.h5). Note: only supports h5 for now. """ # TODO: if h5 filepath is remote, create the file in a temporary directory # then upload it filepath = str(filepath) if not filepath.endswith(".weights.h5"): raise ValueError( "Invalid `filepath` argument: expected a `.weights.h5` extension. " f"Received: filepath={filepath}" ) weights_store = H5IOStore(filepath, mode="w") _save_state( model, weights_store=weights_store, assets_store=None, inner_path="", visited_trackables=set(), ) weights_store.close() def load_weights_only(model, filepath, skip_mismatch=False): """Load the weights of a model from a filepath (.keras or .weights.h5). Note: only supports h5 for now. """ temp_dir = None archive = None filepath = str(filepath) if filepath.endswith(".weights.h5"): # TODO: download file if h5 filepath is remote weights_store = H5IOStore(filepath, mode="r") elif filepath.endswith(".keras"): archive = zipfile.ZipFile(filepath, "r") weights_store = H5IOStore( _VARS_FNAME + ".h5", archive=archive, mode="r" ) _load_state( model, weights_store=weights_store, assets_store=None, inner_path="", skip_mismatch=skip_mismatch, visited_trackables=set(), ) weights_store.close() if temp_dir and tf.io.gfile.exists(temp_dir): tf.io.gfile.rmtree(temp_dir) if archive: archive.close() def _write_to_zip_recursively(zipfile_to_save, system_path, zip_path): if not tf.io.gfile.isdir(system_path): zipfile_to_save.write(system_path, zip_path) else: for file_name in tf.io.gfile.listdir(system_path): system_file_path = tf.io.gfile.join(system_path, file_name) zip_file_path = tf.io.gfile.join(zip_path, file_name) _write_to_zip_recursively( zipfile_to_save, system_file_path, zip_file_path ) def _walk_trackable(trackable): for child_attr in dir(trackable): if child_attr.startswith("__") or child_attr in ATTR_SKIPLIST: continue try: child_obj = getattr(trackable, child_attr) except Exception: # Avoid raising the exception when visiting the attributes. continue yield child_attr, child_obj def _save_state( trackable, weights_store, assets_store, inner_path, visited_trackables ): # If the trackable has already been saved, skip it. if id(trackable) in visited_trackables: return # TODO(fchollet): better name? if hasattr(trackable, "_save_own_variables") and weights_store: trackable._save_own_variables(weights_store.make(inner_path)) if hasattr(trackable, "_save_assets") and assets_store: trackable._save_assets(assets_store.make(inner_path)) visited_trackables.add(id(trackable)) # Recursively save state of children trackables (layers, optimizers, etc.) for child_attr, child_obj in _walk_trackable(trackable): if _is_keras_trackable(child_obj): _save_state( child_obj, weights_store, assets_store, inner_path=tf.io.gfile.join(inner_path, child_attr), visited_trackables=visited_trackables, ) elif isinstance(child_obj, (list, dict, tuple, set)): _save_container_state( child_obj, weights_store, assets_store, inner_path=tf.io.gfile.join(inner_path, child_attr), visited_trackables=visited_trackables, ) def _load_state( trackable, weights_store, assets_store, inner_path, skip_mismatch=False, visited_trackables=None, ): if visited_trackables and id(trackable) in visited_trackables: return if hasattr(trackable, "_load_own_variables") and weights_store: if skip_mismatch: try: trackable._load_own_variables(weights_store.get(inner_path)) except Exception as e: warnings.warn( f"Could not load weights in object {trackable}. " "Skipping object. " f"Exception encountered: {e}", stacklevel=2, ) else: trackable._load_own_variables(weights_store.get(inner_path)) if hasattr(trackable, "_load_assets") and assets_store: if skip_mismatch: try: trackable._load_assets(assets_store.get(inner_path)) except Exception as e: warnings.warn( f"Could not load assets in object {trackable}. " "Skipping object. " f"Exception encountered: {e}", stacklevel=2, ) else: trackable._load_assets(assets_store.get(inner_path)) if visited_trackables is not None: visited_trackables.add(id(trackable)) # Recursively load states for Keras trackables such as layers/optimizers. for child_attr, child_obj in _walk_trackable(trackable): if _is_keras_trackable(child_obj): _load_state( child_obj, weights_store, assets_store, inner_path=tf.io.gfile.join(inner_path, child_attr), skip_mismatch=skip_mismatch, visited_trackables=visited_trackables, ) elif isinstance(child_obj, (list, dict, tuple, set)): _load_container_state( child_obj, weights_store, assets_store, inner_path=tf.io.gfile.join(inner_path, child_attr), skip_mismatch=skip_mismatch, visited_trackables=visited_trackables, ) def _save_container_state( container, weights_store, assets_store, inner_path, visited_trackables ): used_names = {} if isinstance(container, dict): container = list(container.values()) for trackable in container: if _is_keras_trackable(trackable): # Do NOT address the trackable via `trackable.name`, since # names are usually autogenerated and thus not reproducible # (i.e. they may vary across two instances of the same model). name = generic_utils.to_snake_case(trackable.__class__.__name__) if name in used_names: used_names[name] += 1 name = f"{name}_{used_names[name]}" else: used_names[name] = 0 _save_state( trackable, weights_store, assets_store, inner_path=tf.io.gfile.join(inner_path, name), visited_trackables=visited_trackables, ) def _load_container_state( container, weights_store, assets_store, inner_path, skip_mismatch, visited_trackables, ): used_names = {} if isinstance(container, dict): container = list(container.values()) for trackable in container: if _is_keras_trackable(trackable): name = generic_utils.to_snake_case(trackable.__class__.__name__) if name in used_names: used_names[name] += 1 name = f"{name}_{used_names[name]}" else: used_names[name] = 0 _load_state( trackable, weights_store, assets_store, inner_path=tf.io.gfile.join(inner_path, name), skip_mismatch=skip_mismatch, visited_trackables=visited_trackables, ) class DiskIOStore: """Asset store backed by disk storage. If `archive` is specified, then `root_path` refers to the filename inside the archive. If `archive` is not specified, then `root_path` refers to the full path of the target directory. """ def __init__(self, root_path, archive=None, mode=None): self.mode = mode self.root_path = root_path self.archive = archive self.tmp_dir = None if self.archive: self.tmp_dir = _get_temp_dir() if self.mode == "r": self.archive.extractall(path=self.tmp_dir) self.working_dir = tf.io.gfile.join(self.tmp_dir, self.root_path) if self.mode == "w": tf.io.gfile.makedirs(self.working_dir) else: if mode == "r": self.working_dir = root_path else: self.tmp_dir = _get_temp_dir() self.working_dir = tf.io.gfile.join( self.tmp_dir, self.root_path ) tf.io.gfile.makedirs(self.working_dir) def make(self, path): if not path: return self.working_dir path = tf.io.gfile.join(self.working_dir, path) if not tf.io.gfile.exists(path): tf.io.gfile.makedirs(path) return path def get(self, path): if not path: return self.working_dir path = tf.io.gfile.join(self.working_dir, path) if tf.io.gfile.exists(path): return path return None def close(self): if self.mode == "w" and self.archive: _write_to_zip_recursively( self.archive, self.working_dir, self.root_path ) if self.tmp_dir and tf.io.gfile.exists(self.tmp_dir): tf.io.gfile.rmtree(self.tmp_dir) class H5IOStore: def __init__(self, root_path, archive=None, mode="r"): """Numerical variable store backed by HDF5. If `archive` is specified, then `root_path` refers to the filename inside the archive. If `archive` is not specified, then `root_path` refers to the path of the h5 file on disk. """ self.root_path = root_path self.mode = mode self.archive = archive self.io_file = None if self.archive: if self.mode == "w": self.io_file = io.BytesIO() else: self.io_file = self.archive.open(self.root_path, "r") self.h5_file = h5py.File(self.io_file, mode=self.mode) else: self.h5_file = h5py.File(root_path, mode=self.mode) def make(self, path): if not path: return self.h5_file.create_group("vars") return self.h5_file.create_group(path).create_group("vars") def get(self, path): if not path: return self.h5_file["vars"] if path in self.h5_file and "vars" in self.h5_file[path]: return self.h5_file[path]["vars"] return {} def close(self): self.h5_file.close() if self.mode == "w" and self.archive: self.archive.writestr(self.root_path, self.io_file.getvalue()) if self.io_file: self.io_file.close() class NpzIOStore: def __init__(self, root_path, archive=None, mode="r"): """Numerical variable store backed by NumPy.savez/load. If `archive` is specified, then `root_path` refers to the filename inside the archive. If `archive` is not specified, then `root_path` refers to the path of the npz file on disk. """ self.root_path = root_path self.mode = mode self.archive = archive if mode == "w": self.contents = {} else: if self.archive: self.f = archive.open(root_path, mode="r") else: self.f = open(root_path, mode="rb") self.contents = np.load(self.f, allow_pickle=True) def make(self, path): if not path: self.contents["__root__"] = {} return self.contents["__root__"] self.contents[path] = {} return self.contents[path] def get(self, path): if not path: if "__root__" in self.contents: return dict(self.contents["__root__"]) return {} if path in self.contents: return self.contents[path].tolist() return {} def close(self): if self.mode == "w": if self.archive: self.f = self.archive.open( self.root_path, mode="w", force_zip64=True ) else: self.f = open(self.root_path, mode="wb") np.savez(self.f, **self.contents) self.f.close() def _get_temp_dir(): temp_dir = tempfile.mkdtemp() testfile = tempfile.TemporaryFile(dir=temp_dir) testfile.close() return temp_dir def _is_keras_trackable(obj): from keras.metrics import base_metric # To avoid circular import return isinstance( obj, ( base_layer.Layer, optimizer.Optimizer, base_metric.Metric, losses.Loss, ), ) def saving_v3_enabled(): return getattr(_SAVING_V3_ENABLED, "value", False) # Some debugging utilities. def _print_h5_file(h5_file, prefix="", action=None): if not prefix: print(f"Keras weights file ({h5_file}) {action}:") if not hasattr(h5_file, "keys"): return for key in h5_file.keys(): print(f"...{prefix}{key}") _print_h5_file(h5_file[key], prefix=prefix + "...") def _print_zip_file(zipfile, action): # TODO(fchollet): move to debugging logs. io_utils.print_msg(f"Keras model archive {action}:") # Same as `ZipFile.printdir()` except for using Keras' printing utility. io_utils.print_msg( "%-46s %19s %12s" % ("File Name", "Modified ", "Size") ) for zinfo in zipfile.filelist: date = "%d-%02d-%02d %02d:%02d:%02d" % zinfo.date_time[:6] io_utils.print_msg( "%-46s %s %12d" % (zinfo.filename, date, zinfo.file_size) )