o bi @sddlZddlmZddlmZddlmZddlmZddlmZddlm Z ddlm Z dd l m Z dd l mZ   dd d Zd dZddZdS)N)backend)treeconvert_spec_to_tensor)get_input_signature)make_tf_tensor_spec)DEFAULT_ENDPOINT_NAME) ExportArchive) patch_tf2onnx)io_utilsc Ks|}|durd}|durt|}|r|jstdddt|D}tdvrKddlm}tt |}t ||fi|} t |j j | |||d ntd krddl} td d |} t| } td d| Drntdt|drw|t^tjdddtjdddtjdddtjdddtjdddz| jj|| |||dd} t| dr| | |Wn|durd}| jj|| ||||dYWdn1swYntd|rtd|ddSdS)aExport the model as a ONNX artifact for inference. This method lets you export a model to a lightweight ONNX artifact that contains the model's forward pass only (its `call()` method) and can be served via e.g. ONNX Runtime. The original code of the model (including any custom layers you may have used) is *no longer* necessary to reload the artifact -- it is entirely standalone. Args: filepath: `str` or `pathlib.Path` object. The path to save the artifact. verbose: `bool`. Whether to print a message during export. Defaults to `None`, which uses the default value set by different backends and formats. input_signature: Optional. Specifies the shape and dtype of the model inputs. Can be a structure of `keras.InputSpec`, `tf.TensorSpec`, `backend.KerasTensor`, or backend tensor. If not provided, it will be automatically computed. Defaults to `None`. opset_version: Optional. An integer value that specifies the ONNX opset version. If not provided, the default version for the backend will be used. Defaults to `None`. **kwargs: Additional keyword arguments. **Note:** This feature is currently supported only with TensorFlow, JAX and Torch backends. **Note:** The dtype policy must be "float32" for the model. You can further optimize the ONNX artifact using the ONNX toolkit. Learn more here: [https://onnxruntime.ai/docs/performance/](https://onnxruntime.ai/docs/performance/). **Note:** The dynamic shape feature is not yet supported with Torch backend. As a result, you must fully define the shapes of the inputs using `input_signature`. If `input_signature` is not provided, all instances of `None` (such as the batch size) will be replaced with `1`. Example: ```python # Export the model as a ONNX artifact model.export("path/to/location", format="onnx") # Load the artifact in a different process/environment ort_session = onnxruntime.InferenceSession("path/to/location") ort_inputs = { k.name: v for k, v in zip(ort_session.get_inputs(), input_data) } predictions = ort_session.run(None, ort_inputs) ``` NTzSThe model provided has never called. It must be called at least once before export.cSs&g|]\}}t|ddpd|qS)nameNinput_)getattr).0ispecrI/home/ubuntu/.local/lib/python3.10/site-packages/keras/src/export/onnx.py Sszexport_onnx..) tensorflowjaxr)tf2onnx)opset output_pathtorchcSs t|ddS)N)replace_none_numberr)xrrrms zexport_onnx..css|]}t|tVqdS)N) isinstancedict)rrrrr rszexport_onnx..zUCurrently, `export_onnx` in the torch backend doesn't support dictionaries as inputs.evalignorez4.*\n.*\n*.*\n*.*export will treat it as a constant.*)messagez+.*not properly registered as a submodule,.*z3.*which is what 'get_attr' Nodes typically target.*z2.*underlying reference in the owning GraphModule.*z(.*suppressed about get_attr references.*)verbose opset_version input_namesdynamooptimizeF)r%r&r'zI`export_onnx` is only compatible with TensorFlow, JAX and Torch backends.zSaved artifact at 'z'.)r_called ValueError enumeraterkeras.src.utils.module_utilsrr map_structurerget_concrete_fnr convert from_functionrtupleanyhasattrr"warningscatch_warningsfilterwarningsonnxexportr)saveNotImplementedErrorr print_msg) modelfilepathr%input_signaturer&kwargsactual_verboser'r decorated_fnr sample_inputs onnx_programrrr export_onnxs:       2rEcCsv|}d|vr d|d<d|vrddd|d<|ddur!td|dddur-td|dd dur9td |S) N is_staticT jax2tf_kwargsF) enable_xlanative_serializationzB`is_static` must be `True` in `kwargs` when using the jax backend.rHzU`enable_xla` must be `False` in `kwargs['jax2tf_kwargs']` when using the jax backend.rIz_`native_serialization` must be `False` in `kwargs['jax2tf_kwargs']` when using the jax backend.)copyr+)r@rrr_check_jax_kwargss( rKcKsNtdkr t|}t}|jt||fi|tdkr"||tS)z0Get the `tf.function` associated with the model.rr)rrKr track_and_add_endpointr_filter_and_track_resources_get_concrete_fn)r=r?r@export_archiverrrr/s   r/)NNN)r5 keras.srcrrkeras.src.export.export_utilsrrrkeras.src.export.saved_modelrr keras.src.export.tf2onnx_libr keras.src.utilsr rErKr/rrrrs"          (