ijyddlmZddlmZddlmZddlmZddlmZddl m Z edGdd ejZ y ) )backend)layers) keras_export)_list_variables_used_by_fns)serialization_lib) tensorflowzkeras.layers.TFSMLayercXeZdZdZ dfd ZdZddZfdZed dZ xZ S) TFSMLayeragReload a Keras model/layer that was saved via SavedModel / ExportArchive. Arguments: filepath: `str` or `pathlib.Path` object. The path to the SavedModel. call_endpoint: Name of the endpoint to use as the `call()` method of the reloaded layer. If the SavedModel was created via `model.export()`, then the default endpoint name is `'serve'`. In other cases it may be named `'serving_default'`. Example: ```python model.export("path/to/artifact") reloaded_layer = TFSMLayer("path/to/artifact") outputs = reloaded_layer(inputs) ``` The reloaded object can be used like a regular Keras layer, and supports training/fine-tuning of its trainable weights. Note that the reloaded object retains none of the internal structure or custom methods of the original object -- it's a brand new layer created around the saved function. **Limitations:** * Only call endpoints with a single `inputs` tensor argument (which may optionally be a dict/tuple/list of tensors) are supported. For endpoints with multiple separate input tensor arguments, consider subclassing `TFSMLayer` and implementing a `call()` method with a custom signature. * If you need training-time behavior to differ from inference-time behavior (i.e. if you need the reloaded object to support a `training=True` argument in `__call__()`), make sure that the training-time call function is saved as a standalone endpoint in the artifact, and provide its name to the `TFSMLayer` via the `call_training_endpoint` argument. c htjdk7r tdt | |||tj j ||_||_||_ ||_ t|j|rt|j||_ ns||jjvr|jj||_ n,> $ND ! d00;; ;$($6$6$A$A-$PD ! 0$ **55::<=> @  "t))+AB18&&(>2.(4+=+=+H+HH151C1C1N1N*2.!$%;$<=( D..99>>@AB D(() ! NN499 :/8 T )A  % %a ( ) )A  % %a ( ) c tj||j|j|j|j j dd}|j|y)zTracks an existing weight./_) initializerr rshaper N)rVariabler rr0r replace_track_variable)r$weightvariables r*r"zTFSMLayer._add_existing_weight{sP##&&,,,,$$S#.   X&r+c j|r|jr|j|fi|S|j|fi|S)N)rr r)r$inputstrainingkwargss r*callzTFSMLayer.calls> **5t55fGGG$t$$V6v66r+ctt|}|j|j|jd}i||S)N)rrr)r get_configrrr)r$ base_configconfigr)s r*r<zTFSMLayer.get_configsDg(*  !//&*&A&A   )+(((r+c`||ntj}|dur td|di|S)a/Creates a TFSMLayer from its config. Args: config: A Python dictionary, typically the output of `get_config`. custom_objects: Optional dictionary mapping names to custom objects. safe_mode: Boolean, whether to disallow loading TFSMLayer. When `safe_mode=True`, loading is disallowed because TFSMLayer loads external SavedModels that may contain attacker-controlled executable graph code. Defaults to `True`. Returns: A TFSMLayer instance. FahRequested the deserialization of a `TFSMLayer`, which loads an external SavedModel. This carries a potential risk of arbitrary code execution and thus it is disallowed by default. If you trust the source of the artifact, you can override this error by passing `safe_mode=False` to the loading function, or calling `keras.config.enable_unsafe_deserialization().)r in_safe_moder)clsr>custom_objects safe_modeeffective_safe_modes r* from_configzTFSMLayer.from_configsK $ "//1  e +A }V}r+)serveNTNN)F)NN) __name__ __module__ __qualname____doc__rr"r:r< classmethodrF __classcell__)r)s@r*r r sD$R# HT '7 )r+r N) keras.srcrrkeras.src.api_exportrkeras.src.export.saved_modelrkeras.src.savingrkeras.src.utils.module_utilsrrLayerr r@r+r*rTs?-D.9&'l l(lr+