a SicI@sdZddlZddlZddlZddlZddlZddlZddlZddlZddl Z ddl m m ZddlmZddlmZddlmZddlmZddlmZddlmZdd lmZdd lmZdd lmZdd lmZdd lmZddlmZddl m!Z!ddl"m#Z#ddl"m$Z$ddl"m%Z%ddl"m&Z&ddl"m'Z'ddl"m(Z(ddl"m)Z)ddl*m+Z+ddl,m-Z-ddl.m/Z/ddl0m1Z1ddl2m3Z3ddl2m4Z4ddl5m6Z6e#7de8dZ9d Z:ej;ejej?j@Ad!d"d#ZBej?j@Ad$d%d#ZCej?j@Ad&d'd#ZDej?j@Ad(d)d*ZEd+aFeGZHejId,d-ZJe4d.Gd/d0d0ejKe)jLZMGd1d2d2eMZNGd3d4d4eMZOGd5d6d6eMZPd7d8ZQd9d:ZRe4d;gd<d=d>ZSe4d?Gd@dAdAeMZTdS)Bz=Contains the base Layer class, from which all layers inherit.N)backend) constraints) initializers) regularizers) lazy_variable)base_layer_utils) input_spec) keras_tensor)node)autocast_variable)loss_scale_optimizer)policy)layer_serialization) generic_utils) layer_utils)object_identity) tf_inspect)tf_utils)traceback_utils) version_utils) to_snake_case)is_tensor_or_tensor_list) json_format) tf_logging)get_canonical_name_for_symbol) keras_export) doc_controls metrics_modz keras.metrics tf_op_layer_z/tensorflow/api/keras/layerszkeras layers usagemethodz/tensorflow/api/keras/modelszkeras model usagez/tensorflow/api/keraszkeras api usagez$/tensorflow/api/keras/premade_modelszpremade keras model usagetypeFccsnttddsdgt_tj|zd?Z1e j$d@dAZ2dBdCZ3dDdEZ4e j,dFdGZ5e j$dHdIZ6e j$dJdKZ7ee j$dLdMZ8ee j$dNdOZ9e j$dPdQZ:e j$dRdSZ;e j$dTdUZedZd[Z?ee j$d\d]Z@d^d_ZAee j$d`daZBedbdcZCedddeZDedfdgZEee j$dhdiZFee j$djdkZGee j,dldmZHee j,dndoZIee j,dpdqZJe j$drdsZKeLeMNdtejOjPZPdZQdudvZRdwdxZSe j dydzZTd{d|ZUd}d~ZVddZWddZXddZYddZZeddZ[e[j"ejjjddZ[eddZ\e\j"ejjjddZ\ddZ]eddZ^dddZ_ddZ`ddZaeddZbebj"ddZbddZcdddZddddZeddZfddZgddZhddZiddZjddZkddZlddZmddZnddZoeddZpejjjddZqfddZrfddZsddZtdddZudddZvddÄZwdddńZxeddDŽZyeddɄZzedd˄Z{e{j"dd˄Z{dd΄Z|ejjjdddЄZ}ddd҄Z~eddԄZeddքZedd؄Zdfddۄ Zedd݄Zdd߄ZddZZS)Layera This is the class from which all layers inherit. A layer is a callable object that takes as input one or more tensors and that outputs one or more tensors. It involves *computation*, defined in the `call()` method, and a *state* (weight variables). State can be created in various places, at the convenience of the subclass implementer: * in `__init__()`; * in the optional `build()` method, which is invoked by the first `__call__()` to the layer, and supplies the shape(s) of the input(s), which may not have been known at initialization time; * in the first invocation of `call()`, with some caveats discussed below. Users will just instantiate a layer and then treat it as a callable. Args: trainable: Boolean, whether the layer's variables should be trainable. name: String name of the layer. dtype: The dtype of the layer's computations and weights. Can also be a `tf.keras.mixed_precision.Policy`, which allows the computation and weight dtype to differ. Default of `None` means to use `tf.keras.mixed_precision.global_policy()`, which is a float32 policy unless set to different value. dynamic: Set this to `True` if your layer should only be run eagerly, and should not be used to generate a static computation graph. This would be the case for a Tree-RNN or a recursive network, for example, or generally for any layer that manipulates tensors using Python control flow. If `False`, we assume that the layer can safely be used to generate a static computation graph. Attributes: name: The name of the layer (string). dtype: The dtype of the layer's weights. variable_dtype: Alias of `dtype`. compute_dtype: The dtype of the layer's computations. Layers automatically cast inputs to this dtype which causes the computations and output to also be in this dtype. When mixed precision is used with a `tf.keras.mixed_precision.Policy`, this will be different than `variable_dtype`. dtype_policy: The layer's dtype policy. See the `tf.keras.mixed_precision.Policy` documentation for details. trainable_weights: List of variables to be included in backprop. non_trainable_weights: List of variables that should not be included in backprop. weights: The concatenation of the lists trainable_weights and non_trainable_weights (in this order). trainable: Whether the layer should be trained (boolean), i.e. whether its potentially-trainable weights should be returned as part of `layer.trainable_weights`. input_spec: Optional (list of) `InputSpec` object(s) specifying the constraints on inputs that can be accepted by the layer. We recommend that descendants of `Layer` implement the following methods: * `__init__()`: Defines custom layer attributes, and creates layer weights that do not depend on input shapes, using `add_weight()`, or other state. * `build(self, input_shape)`: This method can be used to create weights that depend on the shape(s) of the input(s), using `add_weight()`, or other state. `__call__()` will automatically build the layer (if it has not been built yet) by calling `build()`. * `call(self, inputs, *args, **kwargs)`: Called in `__call__` after making sure `build()` has been called. `call()` performs the logic of applying the layer to the `inputs`. The first invocation may additionally create state that could not be conveniently created in `build()`; see its docstring for details. Two reserved keyword arguments you can optionally use in `call()` are: - `training` (boolean, whether the call is in inference mode or training mode). See more details in [the layer/model subclassing guide]( https://www.tensorflow.org/guide/keras/custom_layers_and_models#privileged_training_argument_in_the_call_method) - `mask` (boolean tensor encoding masked timesteps in the input, used in RNN layers). See more details in [the layer/model subclassing guide]( https://www.tensorflow.org/guide/keras/custom_layers_and_models#privileged_mask_argument_in_the_call_method) A typical signature for this method is `call(self, inputs)`, and user could optionally add `training` and `mask` if the layer need them. `*args` and `**kwargs` is only useful for future extension when more input parameters are planned to be added. * `get_config(self)`: Returns a dictionary containing the configuration used to initialize this layer. If the keys differ from the arguments in `__init__`, then override `from_config(self)` as well. This method is used when saving the layer or a model that contains this layer. Examples: Here's a basic example: a layer with two variables, `w` and `b`, that returns `y = w . x + b`. It shows how to implement `build()` and `call()`. Variables set as attributes of a layer are tracked as weights of the layers (in `layer.weights`). ```python class SimpleDense(Layer): def __init__(self, units=32): super(SimpleDense, self).__init__() self.units = units def build(self, input_shape): # Create the state of the layer (weights) w_init = tf.random_normal_initializer() self.w = tf.Variable( initial_value=w_init(shape=(input_shape[-1], self.units), dtype='float32'), trainable=True) b_init = tf.zeros_initializer() self.b = tf.Variable( initial_value=b_init(shape=(self.units,), dtype='float32'), trainable=True) def call(self, inputs): # Defines the computation from inputs to outputs return tf.matmul(inputs, self.w) + self.b # Instantiates the layer. linear_layer = SimpleDense(4) # This will also call `build(input_shape)` and create the weights. y = linear_layer(tf.ones((2, 2))) assert len(linear_layer.weights) == 2 # These weights are trainable, so they're listed in `trainable_weights`: assert len(linear_layer.trainable_weights) == 2 ``` Note that the method `add_weight()` offers a shortcut to create weights: ```python class SimpleDense(Layer): def __init__(self, units=32): super(SimpleDense, self).__init__() self.units = units def build(self, input_shape): self.w = self.add_weight(shape=(input_shape[-1], self.units), initializer='random_normal', trainable=True) self.b = self.add_weight(shape=(self.units,), initializer='random_normal', trainable=True) def call(self, inputs): return tf.matmul(inputs, self.w) + self.b ``` Besides trainable weights, updated via backpropagation during training, layers can also have non-trainable weights. These weights are meant to be updated manually during `call()`. Here's a example layer that computes the running sum of its inputs: ```python class ComputeSum(Layer): def __init__(self, input_dim): super(ComputeSum, self).__init__() # Create a non-trainable weight. self.total = tf.Variable(initial_value=tf.zeros((input_dim,)), trainable=False) def call(self, inputs): self.total.assign_add(tf.reduce_sum(inputs, axis=0)) return self.total my_sum = ComputeSum(2) x = tf.ones((2, 2)) y = my_sum(x) print(y.numpy()) # [2. 2.] y = my_sum(x) print(y.numpy()) # [4. 4.] assert my_sum.weights == [my_sum.total] assert my_sum.non_trainable_weights == [my_sum.total] assert my_sum.trainable_weights == [] ``` For more information about creating layers, see the guide [Making new Layers and Models via subclassing]( https://www.tensorflow.org/guide/keras/custom_layers_and_models) TNFc Ks|hd}t||t|tsRt|tjtjfrD|jtjusRt d|||_ d|_ d|_ d|_ d|_d|_d|_t|j |_||t|dd|_|dg|dgg|_t|_g|_g|_g|_ t!|_"|#||dt$%|_&|dgg|_'g|_(|)t|tsDt d |||_*d |vrld |vrl|d f|d <d |vsd |vrd |vrt+|d }n4d |vrd |vr|d }nd}|ft+|d }||_,|dd|_-d|_.d|_/t0|_1g|_2dS)N> input_dimweightsautocastimplementation input_shapebatch_input_shapeactivity_regularizer batch_sizez8Expected `trainable` argument to be a boolean, but got: Fr6_trainable_weights_non_trainable_weightsr2_self_tracked_trackablesz6Expected `dynamic` argument to be a boolean, but got: r0r4r5r7r1T)3_instrument_layer_creationrvalidate_kwargs isinstancebooltfTensorVariabledtype TypeError _trainable _statefulbuilt _input_spec_build_input_shape_saved_model_inputs_spec_saved_model_arg_spec is_default compute_mask_supports_masking_init_set_namergetr*_activity_regularizer_maybe_create_attribute_updates threadinglocal _thread_local_callable_losses_losses_metricsLock _metrics_lock_set_dtype_policyrv2_dtype_behavior_enabled _autocast_inbound_nodes_value_outbound_nodes_value_init_call_fn_args_dynamictuple_batch_input_shape_initial_weights_auto_track_sub_layers#_preserve_input_structure_in_configget_current_name_scope_name_scope_on_declaration_captured_weight_regularizer) self trainablenamerBdynamickwargsallowed_kwargsr5r7r,r,r-__init__:s                zLayer.__init__cCs||_d|_dS)aCreates the variables of the layer (optional, for subclass implementers). This is a method that implementers of subclasses of `Layer` or `Model` can override if they need a state-creation step in-between layer instantiation and layer call. It is invoked automatically before the first execution of `call()`. This is typically used to create the weights of `Layer` subclasses (at the discretion of the subclass implementer). Args: input_shape: Instance of `TensorShape`, or list of instances of `TensorShape` if the layer expects a list of inputs (one instance per input). TN)rHrFrjr4r,r,r-buildsz Layer.buildcOs|S)a This is where the layer's logic lives. The `call()` method may not create state (except in its first invocation, wrapping the creation of variables or other resources in `tf.init_scope()`). It is recommended to create state in `__init__()`, or the `build()` method that is called automatically before `call()` executes the first time. Args: inputs: Input tensor, or dict/list/tuple of input tensors. The first positional `inputs` argument is subject to special rules: - `inputs` must be explicitly passed. A layer cannot have zero arguments, and `inputs` cannot be provided via the default value of a keyword argument. - NumPy array or Python scalar values in `inputs` get cast as tensors. - Keras mask metadata is only collected from `inputs`. - Layers are built (`build(input_shape)` method) using shape info from `inputs` only. - `input_spec` compatibility is only checked against `inputs`. - Mixed precision input casting is only applied to `inputs`. If a layer has tensor arguments in `*args` or `**kwargs`, their casting behavior in mixed precision should be handled manually. - The SavedModel input specification is generated using `inputs` only. - Integration with various ecosystem packages like TFMOT, TFLite, TF.js, etc is only supported for `inputs` and not for tensors in positional and keyword arguments. *args: Additional positional arguments. May contain tensors, although this is not recommended, for the reasons above. **kwargs: Additional keyword arguments. May contain tensors, although this is not recommended, for the reasons above. The following optional keyword arguments are reserved: - `training`: Boolean scalar tensor of Python boolean indicating whether the `call` is meant for training or inference. - `mask`: Boolean input mask. If the layer's `call()` method takes a `mask` argument, its default value will be set to the mask generated for `inputs` by the previous layer (if `input` did come from a layer that generated a corresponding mask, i.e. if it came from a Keras layer with masking support). Returns: A tensor or list/tuple of tensors. r,)rjinputsargsrnr,r,r-calls.z Layer.callc  s|dur d}| dd| D]} | dvrtd| q| dd} | dd}| d d}| d d}|s||r|t||d d}|dur|jpt}t|}|jj dur| t |j jt|}t|}t|}| tjjkr|rtd nd }n|durd}|durz|jr&td}nT|js>|js>|jrJtd}n0d| vrztd|d|j d|jd|d | dtj}|r|jj|jj kr|jr|fdd}|durtdd}|rtj ||d}|j!|||d|||||| | | |d }|dur6|jd|j"d}|#|||t$|rz|D]0}t%||rh|j&'|n |j('|qFn*t%||r|j&'|n |j('||S)aAdds a new variable to the layer. Args: name: Variable name. shape: Variable shape. Defaults to scalar if unspecified. dtype: The type of the variable. Defaults to `self.dtype`. initializer: Initializer instance (callable). regularizer: Regularizer instance (callable). trainable: Boolean, whether the variable should be part of the layer's "trainable_variables" (e.g. variables, biases) or "non_trainable_variables" (e.g. BatchNorm mean and variance). Note that `trainable` cannot be `True` if `synchronization` is set to `ON_READ`. constraint: Constraint instance (callable). use_resource: Whether to use a `ResourceVariable` or not. See [this guide]( https://www.tensorflow.org/guide/migrate/tf1_vs_tf2#resourcevariables_instead_of_referencevariables) for more information. synchronization: Indicates when a distributed a variable will be aggregated. Accepted values are constants defined in the class `tf.VariableSynchronization`. By default the synchronization is set to `AUTO` and the current `DistributionStrategy` chooses when to synchronize. If `synchronization` is set to `ON_READ`, `trainable` must not be set to `True`. aggregation: Indicates how a distributed variable will be aggregated. Accepted values are constants defined in the class `tf.VariableAggregation`. **kwargs: Additional keyword arguments. Accepted values are `getter`, `collections`, `experimental_autocast` and `caching_device`. Returns: The variable created. Raises: ValueError: When giving unsupported dtype and no initializer or when trainable has been set to True with synchronization set as `ON_READ`. Nr, partitioner) collectionsexperimental_autocastcaching_devicegetterlayoutzUnknown keyword argument:rwrxTryr{_layoutzSynchronization value can be set to VariableSynchronization.ON_READ only for non-trainable variables. You have specified trainable=True and synchronization=VariableSynchronization.ON_READ.Fglorot_uniformzerosrzzAn initializer for variable z of type z is required for layer z . Received: .cs|i|}t|SN)r create_autocast_variable)rtrnvariable old_getterr,r-rzsz Layer.add_weight..getterzb`caching_device` does not work with mixed precision API. Ignoring user specified `caching_device`.)r{) rlshaperz overwrite initializerrB constraintrk use_resourcerwsynchronization aggregationry:))r*rCr&rBrfloatxr?as_dtype _dtype_policyvariable_dtyper[r Policy base_dtyperlrrOrrVariableSynchronizationON_READ ValueError is_floating is_integer is_unsignedis_boolr make_variable compute_dtyperwarning functoolspartial _add_variable_with_custom_getterfind_handle_weight_regularizationis_split_variabletrack_variabler8r(r9)rjrlrrBr regularizerrkrrrrrnkwargcollections_argr2ryr{rzr name_in_scopevr,rr- add_weight!s5                         zLayer.add_weightcst|jjdd}|j|jd}t|dr8|j|d<t |j |d<t|drv|j rd|j |d<nd|vrv| d| fdd |D}|rt|jd rttd |jjd |d |S)aReturns the config of the layer. A layer config is a Python dictionary (serializable) containing the configuration of a layer. The same layer can be reinstantiated later (without its trained weights) from this configuration. The config of a layer does not include connectivity information, nor the layer class name. These are handled by `Network` (one layer of abstraction above). Note that `get_config()` does not guarantee to return a fresh copy of dict every time it is called. The callers should make a copy of the returned dict if they want to modify it. Returns: Python dictionary. N)rlrkrcr5rBrmcsg|]}|vr|qSr,r,).0arg expected_argsr,r- z$Layer.get_config.. _is_defaultz Layer z has arguments a! in `__init__` and therefore must override `get_config()`. Example: class CustomLayer(keras.layers.Layer): def __init__(self, arg1, arg2): super().__init__() self.arg1 = arg1 self.arg2 = arg2 def get_config(self): config = super().get_config() config.update({ "arg1": self.arg1, "arg2": self.arg2, }) return config)rgetfullargspecrprtrlrkhasattrrcr serializerrmremovekeys get_configNotImplementedErrortextwrapdedent __class____name__)rjall_argsconfig extra_argsr,rr-rs2     zLayer.get_configcCs|fi|S)aCreates a layer from its config. This method is the reverse of `get_config`, capable of instantiating the same layer from the config dictionary. It does not handle layer connectivity (handled by Network), nor weights (handled by `set_weights`). Args: config: A Python dictionary, typically the output of get_config. Returns: A layer instance. r,)clsrr,r,r- from_config,szLayer.from_configc str҈|tjd}tj|tj |dd}fdd}tj ||}z|dd}Wn6t y}zt djj|WYd}~n d}~00Wdn1s0Ytj d d |St d jjdS) a:Computes the output shape of the layer. This method will cause the layer's state to be built, if that has not happened before. This requires that the layer will later be used with inputs that match the input shape provided here. Args: input_shape: Shape tuple (tuple of integers) or list of shape tuples (one per output tensor of the layer). Shape tuples can include None for free dimensions, instead of an integer. Returns: An input shape tuple. _scratch_graphF to_tuplescstj|jd}d|_|SN)rrB)r placeholderrB _keras_mask)rphrjr,r-_make_placeholder_like]sz:Layer.compute_output_shape.._make_placeholder_like)trainingzWe could not automatically infer the static shape of the layer's output. Please implement the `compute_output_shape` method on your layer (%s).NcSs|jSr)rtr,r,r-nrz,Layer.compute_output_shape..z[Please run in eager mode or implement the `compute_output_shape` method on your layer (%s).)r?executing_eagerly _maybe_buildstrrl __internal__ FuncGraph as_defaultrconvert_shapesnest map_structurerCrrr)rjr4 graph_namerrsoutputser,rr-compute_output_shape>s6  6zLayer.compute_output_shapecsbdd}tj||}||}|jdurLddtj|D}|dtjfdd|S) aCompute the output tensor signature of the layer based on the inputs. Unlike a TensorShape object, a TensorSpec object contains both shape and dtype information for a tensor. This method allows layers to provide output dtype information if it is different from the input dtype. For any layer that doesn't implement this function, the framework will fall back to use `compute_output_shape`, and will assume that the output dtype matches the input dtype. Args: input_signature: Single TensorSpec or nested structure of TensorSpec objects, describing a candidate input for the layer. Returns: Single TensorSpec or nested structure of TensorSpec objects, describing how the layer would transform the provided input. Raises: TypeError: If input_signature contains a non-TensorSpec object. cSs"t|tjstd|d|jS)Nz9Only TensorSpec signature types are supported. Received: r)r=r? TensorSpecrCrsr,r,r-check_type_return_shapes z?Layer.compute_output_signature..check_type_return_shapeNcSsg|] }|jqSr,rBrrr,r,r-rrz2Layer.compute_output_signature..rcstj|dS)N)rBr)r?rrrr,r-rrz0Layer.compute_output_signature..)r?rrr_compute_dtypeflatten)rjinput_signaturerr4 output_shape input_dtypesr,rr-compute_output_signaturets  zLayer.compute_output_signaturecCsB|js>tddtj|Dr:td|jdt|dS|S)a Computes an output mask tensor. Args: inputs: Tensor or list of tensors. mask: Tensor or list of tensors. Returns: None or a tensor (or list of tensors, one per output tensor of the layer). css|]}|duVqdSrr,rmr,r,r- rz%Layer.compute_mask..Layer z9 does not support masking, but was passed an input_mask: N)rManyr?rrrCrlr)rjrsmaskr,r,r-rLs  zLayer.compute_maskc Ost|dstd|j||\}}}tj|}t|||||rR|||||St }t dd|Drtj t |}tj|}|||||\}}|jr|r||d<||||\}}}|js|t} |j||| |d~t|j||j| r|j} |j} n|} |} tj| d|jd|jjd d } t !} t"rh| #t$|j%| #t&| |j's|(||j)r|*||}t+,|j-&| |g|Ri|} Wd n1s0Y|j.r|/|| |j0r|1|| || |j2d ur*|3|||| Wd Wd S1sR0YWd n1sr0Yd S) aWraps `call`, applying pre- and post-processing steps. Args: *args: Positional arguments to be passed to `self.call`. **kwargs: Keyword arguments to be passed to `self.call`. Returns: Output tensor(s). Note: - The following optional keyword arguments are reserved for specific uses: * `training`: Boolean scalar tensor of Python boolean indicating whether the `call` is meant for training or inference. * `mask`: Boolean input mask. - If the layer's `call` method takes a `mask` argument (as some Keras layers do), its default value will be set to the mask generated for `inputs` by the previous layer (if `input` did come from a layer that generated a corresponding mask, i.e. if it came from a Keras layer with masking support. - If the layer is not built, the method will call `build`. Raises: ValueError: if the layer's `call` method returns None (an invalid value). RuntimeError: if `super().__init__()` was not called in the constructor. rUz.rlayerrs build_graphrlayer "z" " f"(type ) object_nameN)4r RuntimeError _call_specsplit_out_first_argr?rr _in_functional_construction_mode_functional_construction_callr call_contextrr_convert_numpy_or_python_types_get_input_masks_expects_mask_arg_set_training_modein_call _clear_lossesrenterrassert_input_compatibilityrlru_name_get_unnested_name_scope_autographed_callr!inject_argument_info_in_tracebackrr contextlib ExitStack+_is_name_scope_on_model_declaration_enabled enter_contextr.rh name_scoperFrr]_maybe_cast_inputsr enable_auto_cast_variables_compute_dtype_objectrP_handle_activity_regularizationrM_set_mask_metadatarI_set_save_spec)rjrtrnrs input_listr input_masksmask_is_implicit training_modeeagercall_fnrZnamescope_stackrr,r,r-__call__s             6   zLayer.__call__c Cstrt|jp}tdt|g}d|d}|dkr>|j}t||}Wdn1sf0YWdq1s0Yn|}|SNr%) rr.rhfilterr?rgjoinr _name_scope)rjZ"relative_name_scope_on_declarationr+current_name_scoperr,r,r-r Vs$ FzLayer._get_unnested_name_scopecCs|jjS)zThe dtype of the layer weights. This is equivalent to `Layer.dtype_policy.variable_dtype`. Unless mixed precision is used, this is the same as `Layer.compute_dtype`, the dtype of the layer's computations. rrrr,r,r-rBmsz Layer.dtypecCs|jS)z3Name of the layer (string), set in the constructor.)r rr,r,r-rlwsz Layer.namecCs|jS)zBWhether this layer supports computing a mask using `compute_mask`.rMrr,r,r-supports_masking|szLayer.supports_maskingcCs ||_dSrr&rjr!r,r,r-r'scCstdd|DS)zBWhether the layer is dynamic (eager-only); set in the constructor.css|] }|jVqdSr)rarrr,r,r-rrz Layer.dynamic..r_flatten_layersrr,r,r-rmsz Layer.dynamiccCstdd|DS)Ncss|] }|jVqdSrrEr)r,r,r-rrz!Layer.stateful..r*rr,r,r-statefulszLayer.statefulcCs ||_dSrr,r(r,r,r-r-scCs|jSr)rDrr,r,r-rkszLayer.trainablecCs|D] }||_qdS)aSets trainable attribute for the layer and its sublayers. When this value is changed during training (e.g. with a `tf.keras.callbacks.Callback`) you need to call the parent `tf.keras.Model.make_train_function` with `force=True` in order to recompile the training graph. Args: value: Boolean with the desired state for the layer's trainable attribute. N)r+rD)rjr!rr,r,r-rks cCs|jS);Optional regularizer function for the output of this layer.rPrr,r,r-r6szLayer.activity_regularizercCs ||_dS)r.Nr/)rjrr,r,r-r6scCs|jS)a|`InputSpec` instance(s) describing the input format for this layer. When you create a layer subclass, you can set `self.input_spec` to enable the layer to run input compatibility checks when it is called. Consider a `Conv2D` layer: it can only be called on a single input tensor of rank 4. As such, you can set, in `__init__()`: ```python self.input_spec = tf.keras.layers.InputSpec(ndim=4) ``` Now, if you try to call the layer on an input that isn't rank 4 (for instance, an input of shape `(2,)`, it will raise a nicely-formatted error: ``` ValueError: Input 0 of layer conv2d is incompatible with the layer: expected ndim=4, found ndim=1. Full shape received: [2] ``` Input checks that can be specified via `input_spec` include: - Structure (e.g. a single input, a list of 2 inputs, etc) - Shape - Rank (ndim) - Dtype For more information, see `tf.keras.layers.InputSpec`. Returns: A `tf.keras.layers.InputSpec` instance, or nested structure thereof. )rGrr,r,r-rs!zLayer.input_speccCs>tj|D]&}|dur t|tjs td|q ||_dS)Nz:Layer input_spec must be an instance of InputSpec. Got: {}) r?rrr=r InputSpecrCformatrG)rjr!rr,r,r-rscCs(|jr |d}||j|SgSdS)zList of all trainable weights tracked by this layer. Trainable weights are updated via gradient descent during training. Returns: A list of trainable variables. trainable_variablesN)rk_gather_children_attribute_dedup_weightsr8)rjchildren_weightsr,r,r-trainable_weightss zLayer.trainable_weightscCs@|jr|d}|j|}n|d}|j|j|}||S)aList of all non-trainable weights tracked by this layer. Non-trainable weights are *not* updated during training. They are expected to be updated manually in `call()`. Returns: A list of non-trainable variables. non_trainable_variables variables)rkr3r9r8r4)rjr5non_trainable_weightsr,r,r-r9s  zLayer.non_trainable_weightscCs |j|jS)zjReturns the list of all layer variables/weights. Returns: A list of variables. )r6r9rr,r,r-r1sz Layer.weightscCstjdddgS)Nz`layer.updates` will be removed in a future version. This property should not be used in TensorFlow 2.0, as `updates` are applied automatically. stacklevel)warningswarnrr,r,r-updatess z Layer.updatescCsjg}|D]X}|jr4|jdtjur@||jn ||j|jD]}|}|durF||qFq |S)aList of losses added using the `add_loss()` API. Variable regularization tensors are created when this property is accessed, so it is eager safe: accessing `losses` under a `tf.GradientTape` will propagate gradients back to the corresponding variables. Examples: >>> class MyLayer(tf.keras.layers.Layer): ... def call(self, inputs): ... self.add_loss(tf.abs(tf.reduce_mean(inputs))) ... return inputs >>> l = MyLayer() >>> l(np.ones((10, 1))) >>> l.losses [1.0] >>> inputs = tf.keras.Input(shape=(10,)) >>> x = tf.keras.layers.Dense(10)(inputs) >>> outputs = tf.keras.layers.Dense(1)(x) >>> model = tf.keras.Model(inputs, outputs) >>> # Activity regularization. >>> len(model.losses) 0 >>> model.add_loss(tf.abs(tf.reduce_mean(x))) >>> len(model.losses) 1 >>> inputs = tf.keras.Input(shape=(10,)) >>> d = tf.keras.layers.Dense(10, kernel_initializer='ones') >>> x = d(inputs) >>> outputs = tf.keras.layers.Dense(1)(x) >>> model = tf.keras.Model(inputs, outputs) >>> # Weight regularization. >>> model.add_loss(lambda: tf.reduce_mean(d.kernel)) >>> model.losses [] Returns: A list of tensors. rN)r+ _eager_lossesrREVIVED_LOSS_PLACEHOLDERextendrWrVr()rjcollected_lossesrr loss_tensorr,r,r-losses"s,   z Layer.lossesc Ks@|dd|r"td|fdd}tj|}g}g}g}|D]}t|rf|t ||qF|durpqFt |st |t j stj|td}t|st |t j rts||qFt |rF||qF|j|tj}|r|std|j||D],} t|dd r,|| n |j| qdS) a Add loss tensor(s), potentially dependent on layer inputs. Some losses (for instance, activity regularization losses) may be dependent on the inputs passed when calling a layer. Hence, when reusing the same layer on different inputs `a` and `b`, some entries in `layer.losses` may be dependent on `a` and some on `b`. This method automatically keeps track of dependencies. This method can be used inside a subclassed layer or model's `call` function, in which case `losses` should be a Tensor or list of Tensors. Example: ```python class MyLayer(tf.keras.layers.Layer): def call(self, inputs): self.add_loss(tf.abs(tf.reduce_mean(inputs))) return inputs ``` This method can also be called directly on a Functional Model during construction. In this case, any loss Tensors passed to this Model must be symbolic and be able to be traced back to the model's `Input`s. These losses become part of the model's topology and are tracked in `get_config`. Example: ```python inputs = tf.keras.Input(shape=(10,)) x = tf.keras.layers.Dense(10)(inputs) outputs = tf.keras.layers.Dense(1)(x) model = tf.keras.Model(inputs, outputs) # Activity regularization. model.add_loss(tf.abs(tf.reduce_mean(x))) ``` If this is not the case for your loss (if, for example, your loss references a `Variable` of one of the model's layers), you can wrap your loss in a zero-argument lambda. These losses are not tracked as part of the model's topology since they can't be serialized. Example: ```python inputs = tf.keras.Input(shape=(10,)) d = tf.keras.layers.Dense(10) x = d(inputs) outputs = tf.keras.layers.Dense(1)(x) model = tf.keras.Model(inputs, outputs) # Weight regularization. model.add_loss(lambda: tf.reduce_mean(d.kernel)) ``` Args: losses: Loss tensor, or list/tuple of tensors. Rather than tensors, losses may also be zero-argument callables which create a loss tensor. **kwargs: Used for backwards compatibility only. rsNzUnknown keyword arguments: %scSsjt|r8td|}Wdn1s.0Y|durDdSt|s`tj|td}d|_|S)z3Tags callable loss tensor as `_unconditional_loss`.NrT) callabler rr? is_tensorconvert_to_tensorrr_unconditional_loss)lossr,r,r- _tag_callables $ z%Layer.add_loss.._tag_callablerz|Expected a symbolic Tensors or a callable for the loss value. Please wrap your loss computation in a zero argument `lambda`._is_graph_networkF)r*rCrr?rrrFr(rrrGr=r KerasTensorrHrrris_symbolic_tensorris_in_tf_functionrVrBrrrr@r&_graph_network_add_lossrW) rjrErnrKcallable_losses eager_lossessymbolic_lossesrJin_call_context symbolic_lossr,r,r-add_losscsL=           zLayer.add_lossc CsTg}|D]B}t|dsq |j||jWdq 1sD0Yq |S)aList of metrics added using the `add_metric()` API. Example: >>> input = tf.keras.layers.Input(shape=(3,)) >>> d = tf.keras.layers.Dense(2) >>> output = d(input) >>> d.add_metric(tf.reduce_max(output), name='max') >>> d.add_metric(tf.reduce_min(output), name='min') >>> [m.name for m in d.metrics] ['max', 'min'] Returns: A list of `Metric` objects. rZN)r+rrZrBrX)rjcollected_metricsrr,r,r-metricss  ,z Layer.metricsc Ks~t|}t|dks0t|dkrD|ddkrDtd|dt|d}t|tj}t j }|durz|szt dn |r|j j }|s|st d t||st|d d sPt|dd}| } |r|j n|}|jX||} | r| }n4|r|j|n"tj|t|d dd }|j|Wdn1s60Y| rz||n*|r^t d|rhdnd} ||| |dS)aMAdds metric tensor to the layer. This method can be used inside the `call()` method of a subclassed layer or model. ```python class MyMetricLayer(tf.keras.layers.Layer): def __init__(self): super(MyMetricLayer, self).__init__(name='my_metric_layer') self.mean = tf.keras.metrics.Mean(name='metric_1') def call(self, inputs): self.add_metric(self.mean(inputs)) self.add_metric(tf.reduce_sum(inputs), name='metric_2') return inputs ``` This method can also be called directly on a Functional Model during construction. In this case, any tensor passed to this Model must be symbolic and be able to be traced back to the model's `Input`s. These metrics become part of the model's topology and are tracked when you save the model via `save()`. ```python inputs = tf.keras.Input(shape=(10,)) x = tf.keras.layers.Dense(10)(inputs) outputs = tf.keras.layers.Dense(1)(x) model = tf.keras.Model(inputs, outputs) model.add_metric(math_ops.reduce_sum(x), name='metric_1') ``` Note: Calling `add_metric()` with the result of a metric object on a Functional Model, as shown in the example below, is not supported. This is because we cannot trace the metric result tensor back to the model's inputs. ```python inputs = tf.keras.Input(shape=(10,)) x = tf.keras.layers.Dense(10)(inputs) outputs = tf.keras.layers.Dense(1)(x) model = tf.keras.Model(inputs, outputs) model.add_metric(tf.keras.metrics.Mean()(x), name='metric_1') ``` Args: value: Metric tensor. name: String metric name. **kwargs: Additional keyword arguments for backward compatibility. Accepted values: `aggregation` - When the `value` tensor provided is not the result of calling a `keras.Metric` instance, it will be aggregated by default using a `keras.Metric.Mean`. rrrzUnknown keyword arguments: z. Expected `aggregation`. _metric_objNzkPlease provide a name for your metric like `self.add_metric(tf.reduce_sum(inputs), name='mean_activation')`z;Expected a symbolic Tensor for the metric value, received: rLFrB)rlrBzUsing the result of calling a `Metric` object when calling `add_metric` on a Functional Model is not supported. Please pass the Tensor to monitor directly.mean)listrlenrCrr=r rMrrrrrYrlrr&rZ_get_existing_metricrXr(rMean_graph_network_add_metric) rjr!rlrn kwargs_keysfrom_metric_obj is_symbolicrT metric_objshould_update_statematchrr,r,r- add_metrics\6           , zLayer.add_metriccCs<t}|jrdS|js8tj|D]}t|r$|q$dS)aAdd update op(s), potentially dependent on layer inputs. Weight updates (for instance, the updates of the moving mean and variance in a BatchNormalization layer) may be dependent on the inputs passed when calling a layer. Hence, when reusing the same layer on different inputs `a` and `b`, some entries in `layer.updates` may be dependent on `a` and some on `b`. This method automatically keeps track of dependencies. This call is ignored when eager execution is enabled (in that case, variable updates are run on the fly and thus do not need to be tracked for later execution). Args: updates: Update op, or list/tuple of update ops, or zero-arg callable that returns an update op. A zero-arg callable should be passed in order to disable running the updates by setting `trainable=False` on this Layer, when executing in Eager mode. N)rrin_keras_graphfrozenr?rrrF)rjr?rupdater,r,r- add_update}szLayer.add_updatec Cs4|j}d}|D]$}t|tjr*||j7}q|d7}q|t|krftd|jt||t|ddfd}g}|D]}t|tjr|j}||||}| |||7}qr||} t | dr| j nd} |j } | | std|jd | d | d | || f|d7}qrt||D]} | q dS) aSets the weights of the layer, from NumPy arrays. The weights of a layer represent the state of the layer. This function sets the weight values from numpy arrays. The weight values should be passed in the order they are created by the layer. Note that the layer's weights must be instantiated before calling this function, by calling the layer. For example, a `Dense` layer returns a list of two values: the kernel matrix and the bias vector. These can be used to set the weights of another `Dense` layer: >>> layer_a = tf.keras.layers.Dense(1, ... kernel_initializer=tf.constant_initializer(1.)) >>> a_out = layer_a(tf.convert_to_tensor([[1., 2., 3.]])) >>> layer_a.get_weights() [array([[1.], [1.], [1.]], dtype=float32), array([0.], dtype=float32)] >>> layer_b = tf.keras.layers.Dense(1, ... kernel_initializer=tf.constant_initializer(2.)) >>> b_out = layer_b(tf.convert_to_tensor([[10., 20., 30.]])) >>> layer_b.get_weights() [array([[2.], [2.], [2.]], dtype=float32), array([0.], dtype=float32)] >>> layer_b.set_weights(layer_a.get_weights()) >>> layer_b.get_weights() [array([[1.], [1.], [1.]], dtype=float32), array([0.], dtype=float32)] Args: weights: a list of NumPy arrays. The number of arrays and their shape must match number of the dimensions of the weights of the layer (i.e. it should match the output of `get_weights`). Raises: ValueError: If the provided weights list does not match the layer's specifications. rrzYou called `set_weights(weights)` on layer "%s" with a weight list of length %s, but the layer was expecting %s weights. Provided weights: %s...N2rr,rz weight shape z. is not compatible with provided weight shape r)r1r=rTrackableWeightHandler num_tensorsr\rrlr set_weightsrris_compatible_withr(rbatch_set_valuer+finalize_state) rjr1paramsexpected_num_weightsparam weight_indexweight_value_tuplesrmtensorsweight weight_shape ref_shaperr,r,r-rnsL,            zLayer.set_weightscCsD|j}g}|D]*}t|tjr.||q||qt|S)aReturns the current weights of the layer, as NumPy arrays. The weights of a layer represent the state of the layer. This function returns both trainable and non-trainable weight values associated with this layer as a list of NumPy arrays, which can in turn be used to load state into similarly parameterized layers. For example, a `Dense` layer returns a list of two values: the kernel matrix and the bias vector. These can be used to set the weights of another `Dense` layer: >>> layer_a = tf.keras.layers.Dense(1, ... kernel_initializer=tf.constant_initializer(1.)) >>> a_out = layer_a(tf.convert_to_tensor([[1., 2., 3.]])) >>> layer_a.get_weights() [array([[1.], [1.], [1.]], dtype=float32), array([0.], dtype=float32)] >>> layer_b = tf.keras.layers.Dense(1, ... kernel_initializer=tf.constant_initializer(2.)) >>> b_out = layer_b(tf.convert_to_tensor([[10., 20., 30.]])) >>> layer_b.get_weights() [array([[2.], [2.], [2.]], dtype=float32), array([0.], dtype=float32)] >>> layer_b.set_weights(layer_a.get_weights()) >>> layer_b.get_weights() [array([[1.], [1.], [1.]], dtype=float32), array([0.], dtype=float32)] Returns: Weights values as a list of NumPy arrays. ) r1r=rrlrB get_tensorsr(rbatch_get_value)rjr1output_weightsrxr,r,r- get_weightss#  zLayer.get_weightscCsdS)azFinalizes the layers state after updating layer weights. This function can be subclassed in a layer and will be called after updating a layer weights. It can be overridden to finalize any additional layer state after a weight update. This function will be called after weights of a layer have been restored from a loaded model. Nr,rr,r,r-rq&s zLayer.finalize_statecCs2||}t|tr"dd|DSt|ddSdS)aRetrieves the input mask tensor(s) of a layer at a given node. Args: node_index: Integer, index of the node from which to retrieve the attribute. E.g. `node_index=0` will correspond to the first time the layer was called. Returns: A mask tensor (or list of tensors if the layer has multiple inputs). cSsg|]}t|ddqSrNr&rr,r,r-rCrz+Layer.get_input_mask_at..rN) get_input_atr=r[r&)rj node_indexrsr,r,r-get_input_mask_at3s  zLayer.get_input_mask_atcCs2||}t|tr"dd|DSt|ddSdS)aRetrieves the output mask tensor(s) of a layer at a given node. Args: node_index: Integer, index of the node from which to retrieve the attribute. E.g. `node_index=0` will correspond to the first time the layer was called. Returns: A mask tensor (or list of tensors if the layer has multiple outputs). cSsg|]}t|ddqSrrrr,r,r-rWrz,Layer.get_output_mask_at..rN) get_output_atr=r[r&)rjroutputr,r,r-get_output_mask_atGs  zLayer.get_output_mask_atcCs.|j}t|trdd|DSt|ddSdS)aRetrieves the input mask tensor(s) of a layer. Only applicable if the layer has exactly one inbound node, i.e. if it is connected to one incoming layer. Returns: Input mask tensor (potentially None) or list of input mask tensors. Raises: AttributeError: if the layer is connected to more than one incoming layers. cSsg|]}t|ddqSrrrr,r,r-rmrz$Layer.input_mask..rN)inputr=r[r&rjrsr,r,r- input_mask[s zLayer.input_maskcCs.|j}t|trdd|DSt|ddSdS)aRetrieves the output mask tensor(s) of a layer. Only applicable if the layer has exactly one inbound node, i.e. if it is connected to one incoming layer. Returns: Output mask tensor (potentially None) or list of output mask tensors. Raises: AttributeError: if the layer is connected to more than one incoming layers. cSsg|]}t|ddqSrrrr,r,r-rrz%Layer.output_mask..rN)rr=r[r&)rjrr,r,r- output_maskqs zLayer.output_maskcCs||ddS)aRetrieves the input shape(s) of a layer at a given node. Args: node_index: Integer, index of the node from which to retrieve the attribute. E.g. `node_index=0` will correspond to the first time the layer was called. Returns: A shape tuple (or list of shape tuples if the layer has multiple inputs). Raises: RuntimeError: If called in Eager mode. input_shapesz input shape_get_node_attribute_at_indexrjrr,r,r-get_input_shape_atszLayer.get_input_shape_atcCs||ddS)aRetrieves the output shape(s) of a layer at a given node. Args: node_index: Integer, index of the node from which to retrieve the attribute. E.g. `node_index=0` will correspond to the first time the layer was called. Returns: A shape tuple (or list of shape tuples if the layer has multiple outputs). Raises: RuntimeError: If called in Eager mode. output_shapesz output shaperrr,r,r-get_output_shape_atszLayer.get_output_shape_atcCs||ddS)aRetrieves the input tensor(s) of a layer at a given node. Args: node_index: Integer, index of the node from which to retrieve the attribute. E.g. `node_index=0` will correspond to the first input node of the layer. Returns: A tensor (or list of tensors if the layer has multiple inputs). Raises: RuntimeError: If called in Eager mode. input_tensorsrrrr,r,r-rszLayer.get_input_atcCs||ddS)aRetrieves the output tensor(s) of a layer at a given node. Args: node_index: Integer, index of the node from which to retrieve the attribute. E.g. `node_index=0` will correspond to the first output node of the layer. Returns: A tensor (or list of tensors if the layer has multiple outputs). Raises: RuntimeError: If called in Eager mode. output_tensorsrrrr,r,r-rszLayer.get_output_atcCs&|jstd|jd|dddS)afRetrieves the input tensor(s) of a layer. Only applicable if the layer has exactly one input, i.e. if it is connected to one incoming layer. Returns: Input tensor or list of input tensors. Raises: RuntimeError: If called in Eager mode. AttributeError: If no inbound nodes are found. rz& is not connected, no input to return.rrr_inbound_nodesAttributeErrorrlrrr,r,r-rs  z Layer.inputcCs&|jstd|jd|dddS)aRetrieves the output tensor(s) of a layer. Only applicable if the layer has exactly one output, i.e. if it is connected to one incoming layer. Returns: Output tensor or list of output tensors. Raises: AttributeError: if the layer is connected to more than one incoming layers. RuntimeError: if called in Eager mode. rz has no inbound nodes.rrrrrr,r,r-rs  z Layer.outputcCs^|jstd|jdtdd|jD}t|dkrD|jdjStdt|jddS) aRetrieves the input shape(s) of a layer. Only applicable if the layer has exactly one input, i.e. if it is connected to one incoming layer, or if all inputs have the same shape. Returns: Input shape, as an integer shape tuple (or list of shape tuples, one tuple per input tensor). Raises: AttributeError: if the layer has no defined input_shape. RuntimeError: if called in Eager mode. The layer "z" has never been called and thus has no defined input shape. Note that the `input_shape` property is only available for Functional and Sequential models.cSsg|]}t|jqSr,)rrrr r,r,r-rrz%Layer.input_shape..rrz" has multiple inbound nodes, with different input shapes. Hence the notion of "input shape" is ill-defined for the layer. Use `get_input_shape_at(node_index)` instead.N)rrrlsetr\rr)rjall_input_shapesr,r,r-r4s    zLayer.input_shapecCsp|jsdt|ddrJt|||jWdqd1s>0Yntd|jd|jdt |j S)zCount the total number of scalars composing the weights. Returns: An integer count. Raises: ValueError: if the layer isn't yet built (in which case its weights aren't yet defined). rLFNz*You tried to call `count_params` on layer z=, but the layer isn't built. You can build it manually via: `z.build(batch_input_shape)`.) rFr&rmaybe_init_scoperrsrrlr count_paramsr1rr,r,r-r+s   ,zLayer.count_paramscCsV|jstd|jdtdd|jD}t|dkrD|jdjStd|jdS) aRetrieves the output shape(s) of a layer. Only applicable if the layer has one output, or if all outputs have the same shape. Returns: Output shape, as an integer shape tuple (or list of shape tuples, one tuple per output tensor). Raises: AttributeError: if the layer has no defined output shape. RuntimeError: if called in Eager mode. rz=" has never been called and thus has no defined output shape.cSsg|]}t|jqSr,)rrrr,r,r-rYrz&Layer.output_shape..rrzThe layer "%s" has multiple inbound nodes, with different output shapes. Hence the notion of "output shape" is ill-defined for the layer. Use `get_output_shape_at(node_index)` instead.N)rrrlrr\r)rjall_output_shapesr,r,r-rCs   zLayer.output_shapecCs|jS)zzThe dtype policy associated with this layer. This is an instance of a `tf.keras.mixed_precision.Policy`. )rrr,r,r- dtype_policyhszLayer.dtype_policycCs|jjS)a-The dtype of the layer's computations. This is equivalent to `Layer.dtype_policy.compute_dtype`. Unless mixed precision is used, this is the same as `Layer.dtype`, the dtype of the weights. Layers automatically cast their inputs to the compute dtype, which causes computations and the output to be in the compute dtype as well. This is done by the base Layer class in `Layer.__call__`, so you do not have to insert these casts if implementing your own layer. Layers often perform certain internal computations in higher precision when `compute_dtype` is float16 or bfloat16 for numeric stability. The output will still typically be float16 or bfloat16 in such cases. Returns: The layer's compute dtype. rrrr,r,r-rpszLayer.compute_dtypecCs|jS)z1Alias of `Layer.dtype`, the dtype of the weights.rrr,r,r-rszLayer.variable_dtypecCs|jS)z3Return Functional API nodes upstream of this layer.)rrr,r,r- inbound_nodesszLayer.inbound_nodescCs|jS)z5Return Functional API nodes downstream of this layer.)_outbound_nodesrr,r,r-outbound_nodesszLayer.outbound_nodescCs|jS)aReturns the list of all layer variables/weights. Alias of `self.weights`. Note: This will not track the weights of nested `tf.Modules` that are not themselves Keras layers. Returns: A list of variables. )r1rr,r,r-r8s zLayer.variablescCs|jSr)r6rr,r,r-r2szLayer.trainable_variablescCs|jSr)r9rr,r,r-r7szLayer.non_trainable_variablescOstjddd|j|i|S)z/Deprecated, do NOT use! Alias for `add_weight`.z`layer.add_variable` is deprecated and will be removed in a future version. Please use the `layer.add_weight()` method instead.r:r;)r=r>r)rjrtrnr,r,r- add_variables zLayer.add_variable)_obj_reference_counts_dictcCs6t|jddd}|dur"d|S|jjd|jjS)NkerasT)api_nameadd_prefix_to_v1_namesztf.{}r)rrr1 __module__r)rjcanonical_namer,r,r-_get_cell_names  zLayer._get_cell_namecCsd|_d|_d|_t|ddsxtddd|_t|ddr\t|dd|_qt |dd|_ntdddS)NF_disable_keras_instrumentationrT_is_model_for_instrumentationZ legacy_layer) _instrumented_keras_api_instrumented_keras_layer_class_instrumented_keras_model_classr&keras_api_gaugeget_cellrkeras_models_gaugerkeras_layers_gaugerr,r,r-r;s  z Layer._instrument_layer_creationcCs>t|tjr|}n t|}|r.|j|n |j||S)aAdds a Trackable object to this layer's state. Args: trackable_object: The tf.tracking.Trackable object to add. trainable: Boolean, whether the variable should be part of the layer's "trainable_variables" (e.g. variables, biases) or "non_trainable_variables" (e.g. BatchNorm mean and variance). Returns: The TrackableWeightHandler used to track this object. )r=rrlr8r(r9)rjtrackable_objectrkhandlerr,r,r-_add_trackables   zLayer._add_trackablecCs0t|ddsg|j_n|D] }g|j_qdS)z)Used every step in eager to reset losses.r:N)r&rUr@r+)rjrr,r,r-r s   zLayer._clear_lossescCsF|jr2tjdd|}||}tjtj|S|||||SdS)NcSstj|j|jdSr)r?rrrBrr,r,r-r" rz3Layer._keras_tensor_symbolic_call..)rmr?rrrr rM_infer_output_signature)rjrsrrtrnroutput_signaturer,r,r-_keras_tensor_symbolic_call s z!Layer._keras_tensor_symbolic_callc Cs|}|j}t|r8t|s8tjj|jtjj}t j |d|j d|j j dd}tjt|j d}|tjtj|}tjtj|}tjtj|}tjtj|}t|tt|j:||||}||g|Ri|}Wdn1s0Y|||Wdn1sF0Y|j|||ddtjtj|}Wdn1s0Y| |||t!|d r|j"s|#||~|S) zBCall the layer on input KerasTensors, returns output KerasTensors.rz" (type rrrNF)r _set_inputs)$rur is_subclassedfrom_saved_modelr?r autograph tf_convertcontrol_status_ctxrr rlrrrrrrrr keras_tensor_to_placeholderrrr#r rrrrrrkeras_tensor_from_tensorrrrsr) rjrsrtrnrZkeras_tensor_inputsr scratch_graphrr,r,r-r- sb     6,$ zLayer._infer_output_signaturec Cst}tdd|Dr.cSs$t|tjtjttfr t|S|Srr=r?r@rrrrrHrr,r,r-_convert_non_tensor s z@Layer._functional_construction_call.._convert_non_tensorFrTrrzVA layer's `call` method should return a Tensor or a list of Tensors, not None (layer: z).)pop_kwarg_if_none)rrrr?rrrrrrarg_was_passed get_arg_value_expects_training_argr*rrglobal_learning_phase_is_setlearning_phaserGcastr>default_training_arg set_arg_valuerrrrl_set_connectivity_metadata) rjrsrtrnrrrmask_arg_passed_by_frameworkrrtraining_value training_arg_passed_by_frameworkrr,r,r-rw sx           z#Layer._functional_construction_callcCsd}|jr|jd||r*|jd||}|dur|j}|durF|}nHtrt}t|t rbqt |r|t |t j }qt |}n|jj }|jd|||\}}nd|vr|d}n|j}|||fS)Nr)rrrrrrrrr=r>r?rGrrrr*)rjrtrnrrcall_ctx_trainingr,r,r-r s2     zLayer._set_training_modecCs8t|r.t|s.tjj|jtjjS|jSdSr) rrrr?rrrrurrr,r,r-r  szLayer._autographed_callcCs|jSrr^rr,r,r-r szLayer._inbound_nodescCs ||_dSrrr(r,r,r-r scCs|jSrr_rr,r,r-r szLayer._outbound_nodescCs ||_dSrrr(r,r,r-r$ scCst|tjr||_n\t|tr,t||_nDt|trL|dvrLt||_n$|rftt|j |_n t |_|jj dkrt stj }td|jj|jj f|jjrt|jj|_nd|_dS)zSets self._dtype_policy.) mixed_float16mixed_bfloat16rzMixed precision is not supported with the tf.distribute.Strategy: %s. Either stop using mixed precision by removing the use of the "%s" policy or use a different Strategy, e.g. a MirroredStrategy.N)r=r rrdict deserializerr?rrl global_policyr strategy_supports_loss_scaling distribute get_strategyrrrrr)rjrBstrategyr,r,r-r[) s0      zLayer._set_dtype_policycCs|jjS)z$Deprecated alias of `compute_dtype`.rrr,r,r-rV szLayer._compute_dtypecCsR|stj|}|j}|jo$|o$|j}|rJtt|j|rJtj |j |S|SdS)aMaybe casts the inputs to the compute dtype. If self._compute_dtype is floating-point, and self_autocast is True, floating-point inputs are casted to self._compute_dtype. Args: inputs: Input tensor, or structure of input tensors. input_list: Flat list of input tensors. Returns: `inputs`, but tensors may have been casted to self._compute_dtype N) r?rrrr]rrmap_should_cast_single_inputr_cast_single_input)rjrsrcompute_dtype_objectshould_autocastr,r,r-r[ s   zLayer._maybe_cast_inputscCs(t|tr$|jo"|j|jko"|jjSdS)NF)r=_AUTOCAST_TYPESrrBrrjrr,r,r-rz s  zLayer._should_cast_single_inputcCs ||rt||jS|SdS)z8Cast a single Tensor or TensorSpec to the compute dtype.N)rr?rrrr,r,r-r s zLayer._cast_single_inputcCs|jjSrr%rr,r,r-_dtype sz Layer._dtypecCs t|j}|t|dSr)r?rrlr[r rr(r,r,r-r s cCsBtjjs|jS|j}tj}|r2|d|}|r>|d7}|Sr )r?rtf2enabledrlget_name_scope)rjrr$r,r,r-r# s   zLayer._name_scopecCsR|dur$tjt|jj|d|_n*t|tr@t |||_nt d|dS)N) zero_basedz2Expected `name` argument to be a string, but got: ) runique_object_namerrrrr r=robserve_object_namerC)rjrlrr,r,r-rN s    zLayer._init_set_namecsDfdd|jD}|sdSt|dkr.rzfPlease provide different names for the metrics you have added. We found {} metrics with the name: "{}"r)rXr\rr1)rjrlrer,rr-r] s zLayer._get_existing_metriccslfdd}t|r6|D]}|t||qn2t|tjrV|j |fn|t||dS)z3Create lambdas which compute regularization losses.cs:td|}Wdn1s,0Y|S)z8Creates a regularization loss `Tensor` for variable `v`.z /RegularizerN)rr)rregularizationrlrr,r-_loss_for_variable s&z?Layer._handle_weight_regularization.._loss_for_variableN) rrrVrrr=rLazyInitVariablerir()rjrlrrrrr,rr-r s  z#Layer._handle_weight_regularizationcCs|jrtj|}tdT|D]>}t||}tt|d|j }||}| |q"Wdn1sv0YdS)NActivityRegularizerr) rPr?rrrrrHrrrBrV)rjrsr output_listr activity_lossr7mean_activity_lossr,r,r-r s  z%Layer._handle_activity_regularizationc Cs|js dStj|}t|ddp2tdd|D}|rJ|rF||dS|||}|durbdStj|}t||D]&\} } z | | _ Wqxt yYqx0qx|r||dS)N _compute_output_and_mask_jointlyFcss|]}t|ddduVqdSrrrr,r,r-r sz+Layer._set_mask_metadata..) rMr?rrr&all_set_mask_keras_history_checkedrLziprr) rjrsr previous_maskr flat_outputsmask_already_computed output_masks flat_maskstensorrr,r,r-r s2      zLayer._set_mask_metadatacCs&|D]}t|dddurd|j_qdS)NrT)r&r_keras_history_checked)rjrrr,r,r-r sz%Layer._set_mask_keras_history_checkedcCs|js|jsd}d}nb|jd||r<|jd||}d}n.css|]}|duVqdSrr,)rrr,r,r-r rz)Layer._get_input_masks..T) rMrrrrrr?rpack_sequence_as)rjrsrrtrnr implicit_maskr,r,r-r s zLayer._get_input_masksc Cstj|}tj||f}dd|D}g}|D]P}t||vrxt|jt|}Wdn1sn0Y||q2tj ||}t j ||||d|S)NcSsh|] }t|qSr,)id)rir,r,r- ) rz3Layer._set_connectivity_metadata..) call_args call_kwargsr) r?rrrrrrlidentityr(r node_moduleNode) rjrtrnrr flat_inputs input_ids_set outputs_copyrr,r,r-r$ s  ( z Layer._set_connectivity_metadatacCs|jstd|jd|dt|j|ksNtd|d|dt|jdt|j||}t|tr|t|dkr||d S|Sd S) aPrivate utility to retrieves an attribute (e.g. inputs) from a node. This is used to implement the methods: - get_input_shape_at - get_output_shape_at - get_input_at etc... Args: node_index: Integer index of the node from which to retrieve the attribute. attr: Exact node attribute name. attr_name: Human-readable attribute name, for error messages. Returns: The layer's attribute `attr` at the node of index `node_index`. Raises: RuntimeError: If the layer has no inbound nodes, or if called in Eager mode. ValueError: If the index provided does not match any node. z The layer z/ has never been called and thus has no defined rz Asked to get z at node z, but the layer has only z inbound nodes.rrN)rrrlr\rr&r=r[)rjrattr attr_namevaluesr,r,r-r< s$ z"Layer._get_node_attribute_at_indexcCsJ|jst|j||jtj|}|rj|jjdurjz|dj j j}Wnt yXYn0| t |d}tdd|Drt|}n&ztj|dd}WntyYn0t|jdst|||Wdn1s0Yt|||jdurFt||jWdn1s60Yd|_dS)Nrcss|]}t|dVqdS)rN)rrr,r,r-rt rz%Layer._maybe_build..Frr)rFrr rlr?rrrrrBrrr[r rrr get_shapesrrrrrrr/rd init_scopern)rjrsrrBrr,r,r-rd s8        (   ,zLayer._maybe_buildcCs$t}|D]}|j||<q|S)zGet the `trainable` state of each sublayer. Returns: A dict mapping all sublayers to their `trainable` value. )weakrefWeakKeyDictionaryr+rkrjtrainable_staterr,r,r-_get_trainable_state s  zLayer._get_trainable_statecCs$|D]}||vr|||_qdS)z(Set `trainable` state for each sublayer.N)r+rkrr,r,r-_set_trainable_state s zLayer._set_trainable_statecCs|dt|jS)z?A dict counting the number of attributes referencing an object.r)rQrObjectIdentityDictionaryrrr,r,r-_obj_reference_counts s zLayer._obj_reference_countscCst||s|||dS)a+Create the attribute with the default value if it hasn't been created. This is useful for fields that is used for tracking purpose, _trainable_weights, or _layers. Note that user could create a layer subclass and assign an internal field before invoking the Layer.__init__(), the __setattr__() need to create the tracking fields and __init__() need to not override them. Args: name: String, the name of the attribute. default_value: Object, the default value of the attribute. N)r __setattr__)rjrl default_valuer,r,r-rQ s zLayer._maybe_create_attributecs$t||d|j}|vr4ttjjj||dS|}|dkrj|d|<ttjjj||dS|=ttjjj||tt st rttjjj| dfdd|j Dttjr ttjjj| dfdd|jDttjjj| dfdd|jDdS) Nrr:csg|]}|ur|qSr,r,)rlexisting_valuer,r-r sz%Layer.__delattr__..r8csg|]}|ur|qSr,r,rwrr,r-r rr9csg|]}|ur|qSr,r,r rr,r-r s)r&rsuperr?rtracking AutoTrackable __delattr__r=r/r has_weightsrr:rAr8r9)rjrlreference_countsreference_countrrr-r% sH    zLayer.__delattr__cs|dks t|ddr t|j|rbzttjjj||Wn t y\t d |Yn0dStjjj ||d|j }| dd|<z||Wnt yYn0tjD]&ttjrt|dr|jqt|ddrRttjstrR|d gtfd d |jDsR|jtd rRd_tjjdd D]ttjsxqb|dg|dgjrtfdd |jDrqb|jn*tfdd |j Drqb|j t!"qbttjjj||dS)N_self_setattr_trackingTzCan't set the attribute "{}", likely because it conflicts with an existing read-only @property of the object. Please choose a different name.) trackabler!rlrrrXrer:c3s|]}|uVqdSrr,r))r!r,r-r. rz$Layer.__setattr__.._use_resource_variables)expand_compositesr8r9c3s|]}|uVqdSrr,r valr,r-rB rc3s|]}|uVqdSrr,r r.r,r-rF r)#r&rrr"r?rr#r$rrr1sticky_attribute_assignmentrrOr%rrr=rMetricrXr(Modulerr&rQrr:r,rArkr8r9rr)rjrlr!r'r))r/r!r-r sx            zLayer.__setattr__csFdvs Jt|drB|jddd}ttjfdd|DSgS)N>r2r7r8r:F) include_self recursivec3s|]}t|VqdSrrr) attributer,r-r] sz3Layer._gather_children_attribute..)r_flatten_modulesr[ itertoolschain from_iterable)rjr6 nested_layersr,r5r-r3R s  z Layer._gather_children_attributeccs(|j||dD]}t|tr|VqdS)N)r4r3)r7r=r/)rjr4r3rr,r,r-r+c s   zLayer._flatten_layersc cs|r |Vt|dd}|rt}t|}|r|}t|}||vrHq*||t|tj rt|t j s|V|rt|dd}|r| t |q*t|tjjjr*|j} | r*| t | q*dS)a Flattens `tf.Module` instances (excluding `Metrics`). Args: recursive: Whether to recursively flatten through submodules. include_self: Whether to include this `Layer` instance. Yields: `tf.Module` instance tracked by this `Layer`. r:N)r&rrwdequepopleftraddr=r?r2rr1 extendleftreversedrr#TrackableDataStructure_values) rjr4r3 trackablesseen_object_idsr< trackable_obj trackable_id subtrackablestracked_valuesr,r,r-r7j s:    zLayer._flatten_modulescCsdS)NTr,rr,r,r- _is_layer szLayer._is_layercCs(tt|j|_|dur$||j_dSr)rCallFunctionSpecrrrurexpects_training_arg)rjrKr,r,r-r` s  zLayer._init_call_fn_argscCs|jjS)z9Whether the call function uses 'training' as a parameter.)rrKrr,r,r-r szLayer._expects_training_argcCs|jjSr)rexpects_mask_argrr,r,r-r szLayer._expects_mask_argcCst|jdsg|j_|jjS)Nr@)rrUr@rr,r,r-r@ s zLayer._eager_lossescCs ||j_dSr)rUr@)rjrEr,r,r-r@ scCs>gt}}|D](}t||vr|||t|q|S)z;Dedupe weights while maintaining order as much as possible.)rrr(r>)rjr1rseen_idsr!r,r,r-r4 s    zLayer._dedup_weightsc Cs|jdurdStjtj|}tjtj|p.g}i}|D]H\}}tj|} dd| D} tdd| Drtq>tj || ||<q>||_|gt ||f|_ dS)aDefines the save spec so that serialization is able to trace layer call. The TensorSpecs of the call function `inputs`, `args`, and `kwargs` are saved into a tuple of `([inputs] + args, kwargs)`. Args: inputs: possibly nested inputs passed into the call function. args: a list of positional arguments passed into call. kwargs: a dictionary of keyword arguments passed into call. NcSsg|]}t|qSr,rget_tensor_specrr,r,r-r rz(Layer._set_save_spec..css|]}|duVqdSrr,rr,r,r-r rz'Layer._set_save_spec..) rIr?rrrrOitemsrrrr[rJ) rjrsrtrnZ inputs_spec args_specZ kwargs_speckeyrZ flat_kwarg flat_specsr,r,r-r s   zLayer._set_save_speccs:|jdurdStjfdd|j}|r6|ddS|S)Ncstj|dS)N dynamic_batchrNrrTr,r-r rz&Layer._get_save_spec..r)rIr?rrrJ)rjrU inputs_onlyspecr,rTr-_get_save_spec s  zLayer._get_save_speccCs t|Sr)rLayerSavedModelSaverrr,r,r-_trackable_saved_model_saver sz"Layer._trackable_saved_model_savercCs|jjSr)rZobject_identifierrr,r,r-_object_identifier szLayer._object_identifiercCs|jjS)z6Info about this layer to be saved into the SavedModel.)rZtracking_metadatarr,r,r-_tracking_metadata szLayer._tracking_metadata checkpointc s@|dkr|d}|j|}ni}|tj|fi||S)N savedmodelcache)rZtrackable_childrenrir"_trackable_childrenrj save_typernrachildrenr)r,r-rc szLayer._trackable_childrencCstt|ddduS)Nr)r)rr rr,r,r-!_use_input_spec_as_call_signature sz'Layer._use_input_spec_as_call_signaturecCs&|j}|dd|dd|S)NrUrZ)__dict__copyr*rjstater,r,r- __getstate__ s   zLayer.__getstate__cCs*t|d<t|d<t|d|dS)NrUrZrh)rSrTrYobjectrrjr,r,r- __setstate__' s  zLayer.__setstate__)TNNF)N)N)N)T)N)TT)TT)N)NN)TT)r_)rr __qualname____doc__r?rr# no_automatic_dependency_trackingrprdefaultrrrfor_subclass_implementersrurAUTOVariableAggregationNONErr classmethodrrrrLrfilter_tracebackrr propertyrBrlr'setterrmdo_not_doc_inheritabler-rkr6rr6r9r1do_not_generate_docsr?rErVrXrfrjrnr~rqrrrrrrrrrrr4rrrrrrrr8r2r7r frozensetr8r9r2_TF_MODULE_IGNORED_PROPERTIES_must_restore_from_configrr;rrrrrrr rrr[rrrrrr#rNr]rrrrrrrrrrrrQr%rr3r+r7rIr`rrr@r4rrXrZr\r^rcrgrlrn __classcell__r,r,r)r-r/s7 "  / I A 6 ,              "      @z   ],            ' #           Ja/    -        "(/    = W  2             r/cs^eZdZdZejjjdfdd ZddZ dd Z d d Z ej d d Z fddZZS)TensorFlowOpLayeraWraps a TensorFlow Operation in a Layer. This class is used internally by the Functional API. When a user uses a raw TensorFlow Operation on symbolic tensors originating from an `Input` Layer, the resultant operation will be wrapped with this Layer object in order to make the operation compatible with the Keras API. This Layer will create a new, identical operation (except for inputs and outputs) every time it is called. If `run_eagerly` is `True`, the op creation and calculation will happen inside an Eager function. Instances of this Layer are created when `autolambda` is called, which is whenever a Layer's `__call__` encounters symbolic inputs that do not have Keras metadata, or when a Network's `__init__` encounters outputs that do not have Keras metadata. Attributes: node_def: String, the serialized NodeDef of the Op this layer will wrap. name: String, the name of the Layer. constants: Dict of NumPy arrays, the values of any Tensors needed for this Operation that do not originate from a Keras `Input` Layer. Since all placeholders must come from Keras `Input` Layers, these Tensors must be treated as constant in the Functional API. trainable: Bool, whether this Layer is trainable. Currently Variables are not supported, and so this parameter has no effect. dtype: The default dtype of this Layer. Inherited from `Layer` and has no effect on this class, however is used in `get_config`. NTcstt|jt|||ddt|tr>t|tj j |_ n&t|t sR|d}tj j j ||_ |dur~dd|Dni|_d|_d|_dS)NF)rlrkrBr2zutf-8cSsi|]\}}t||qSr,)r)rindexconstantr,r,r- c rz.TensorFlowOpLayer.__init__..T)r"rrp_TF_OP_LAYER_NAME_PREFIXr=rr ParseDictr?compatv1NodeDefnode_defbytesencode FromStringrP constantsrFr)rjrrlrrkrBr)r,r-rpM s&     zTensorFlowOpLayer.__init__cCstr||S||Sr)r?r _defun_call_make_oprr,r,r-run s zTensorFlowOpLayer.callcCs6tjj}||jd|jd_||j |_ |S)NT_cloned) r?rrrCopyFromrrb unique_namerl)rjgraphrr,r,r-_make_node_defs s    z TensorFlowOpLayer._make_node_defc CsRtj|}|dj}||}||jD]:\}}t|}|durdtj ||j |d}| ||q6tj j |||gd}||}|tj|jj} dd|jjD} g} | D]} | | | || qt| } tj | |j| |jt|jdkr(|jdWdS|jWdS1sD0YdS)Nrr)control_inputscSsg|]}tj|jqSr,)r?ras_strrl)rrr,r,r-r sz.TensorFlowOpLayer._make_op..r)r?rrrrrrrPget_static_valuerrinsertr create_c_op_create_op_from_tf_operation_control_flow_post_processingrrop_defrlrr(get_attrrbrecord_gradientrsrr\) rjrsrrrrr!c_opopop_type attr_namesattrsrr,r,r-r| s:        zTensorFlowOpLayer._make_opcCs ||S)z@Wraps op creation method in an Eager function for `run_eagerly`.)rrr,r,r-r szTensorFlowOpLayer._defun_callcsLtt|}||dttdt|jdd|j Dd|S)NrlcSsi|]\}}|t|qSr,)r get_value)rrcr,r,r-r sz0TensorFlowOpLayer.get_config..)rlrr) r"rrrir\rr MessageToDictrrrPrjrr)r,r-r s  zTensorFlowOpLayer.get_config)NTN)rrrorpr?rr#rqrprurrfunctionrrrr,r,r)r-r. s  & rcs4eZdZdZfddZddZfddZZS)AddLosszAdds its inputs as a loss. Attributes: unconditional: Whether or not the loss should be conditioned on the inputs. c s(d|d<tt|jfi|||_dS)NFr2)r"rrp unconditional)rjrrnr)r,r-rp szAddLoss.__init__cCs|j||j d|S)N)rs)rVrrr,r,r-ru sz AddLoss.callcs"tt|}|d|ji|S)Nr)r"rrrirrr)r,r-r szAddLoss.get_configrrrorprprurrr,r,r)r-r s rcs6eZdZdZd fdd ZddZfddZZS) AddMetriczAdds its inputs as a metric. Attributes: aggregation: 'mean' or None. How the inputs should be aggregated. metric_name: The name to use for this metric. Nc s&tt|jfi|||_||_dSr)r"rrpr metric_name)rjrrrnr)r,r-rp szAddMetric.__init__cCs|j||j|jd|S)N)rrl)rfrrrr,r,r-ru s zAddMetric.callcs&tt|}||j|jd|S)N)rr)r"rrrirrrr)r,r-r s  zAddMetric.get_config)NNrr,r,r)r-r srcCs tddtj|||gDS)zECheck the arguments to see if we are constructing a functional model.css|]}t|tjVqdSr)r=r rM)rrr,r,r-r sz3_in_functional_construction_mode..)rr?rr)rrsrtrnrr,r,r-r srcCs$t|tjtjttfr t|S|Srrrr,r,r-r s rz8keras.__internal__.apply_name_scope_on_model_declaration)rcCs t|tstd||adS)a<Apply `with tf.name_scope(...)` on model declaration. ```python tf.keras.__internal__.apply_name_scope_on_model_declaration(True) inputs = input_layer.Input((3,)) with tf.name_scope('MyScope'): outputs = layers.Dense(10, name='MyDense')(inputs) model = tf.keras.Model(inputs, outputs) # with `tf.keras.__internal__.apply_name_scope_on_model_declaration(True)`, # The name of the dense layer is "model/MyScope/MyDense/*", and without, # "model/MyDense/*" ``` Args: enable: Enables if `True`, disables if `False`. z3`enable` argument must be `True` or `False`, got {}N)r=r>rCr1r)enabler,r,r-&_apply_name_scope_on_model_declaration s  rz)keras.__internal__.layers.BaseRandomLayercsReZdZdZejjjd fdd ZfddZ dfd d Z fd d Z Z S)BaseRandomLayerzAA layer handle the random number creation and savemodel behavior.NFc s(tjfi|tj|||d|_dS)aInitialize the BaseRandomLayer. Note that the constructor is annotated with @no_automatic_dependency_tracking. This is to skip the auto tracking of self._random_generator instance, which is an AutoTrackable. The backend.RandomGenerator could contain a tf.random.Generator instance which will have tf.Variable as the internal state. We want to avoid saving that state into model.weights and checkpoints for backward compatibility reason. In the meantime, we still need to make them visible to SavedModel when it is tracing the tf.function for the `call()`. See _list_extra_dependencies_for_serialization below for more details. Args: seed: optional integer, used to create RandomGenerator. force_generator: boolean, default to False, whether to force the RandomGenerator to use the code branch of tf.random.Generator. rng_type: string, the rng type that will be passed to backend RandomGenerator. Default to `None`, which will allow RandomGenerator to choose types by itself. Valid values are "stateful", "stateless", "legacy_stateful". **kwargs: other keyword arguments that will be passed to the parent *class )force_generatorrng_typeN)r"rprRandomGenerator_random_generator)rjseedrrrnr)r,r-rpszBaseRandomLayer.__init__cst||jdSr)r"rrr _maybe_initrqr)r,r-rr=s zBaseRandomLayer.buildr_c sJ|dkr(|d}|j|}|j|d<ni}|tj|fi||S)Nr`rar)rZrbrrir"rcrdr)r,r-rcAs z#BaseRandomLayer._trackable_childrencs|dkr|jSt|SdS)Nr)rr"_lookup_dependency)rjrlr)r,r-rRsz"BaseRandomLayer._lookup_dependency)NFN)r_) rrrorpr?rr#rqrprrrcrrr,r,r)r-rs r)Urprwrrr8rrSr=rnumpyrtensorflow.compat.v2rv2r?rrrrr keras.dtensorr keras.enginerrr r r keras.mixed_precisionr r r keras.saving.saved_modelr keras.utilsrrrrrrrkeras.utils.generic_utilsrkeras.utils.tf_utilsrgoogle.protobufrtensorflow.python.platformr tensorflow.python.util.tf_exportrrtensorflow.tools.docsr LazyLoaderglobalsrrr@ SparseTensor RaggedTensorrr monitoring BoolGaugerrrZkeras_premade_model_gaugerrTr'contextmanagerr.r2LayerVersionSelectorr/rrrrrrrr,r,r,r-s                            !D