a hN @sddlZddlmZddlmZmZddlmZddlm Z ddl m Z m Z ddl mZmZddlmZdd lmZmZmZe d Zgd ZGd d d eZeeedddZe dddZddZeeeeeeeeefdddZ d9eeee eedddZ!ee"ee#dddZ$ee"edd d!Z%d:eeee"e ee#ed$d%d&Z&eed'd(d)Z'eed'd*d+Z(ed,d-d.Z)e*d/kreZ+e+,Z-ee-j./e-_.ze0d0e&e-j1e-j2e-j.e-j3e-j4e-j5e-je-j(reee-j1d1kre0d2e0d3e'e-j.e-_6e(e-j6e-_7e-j8rXe0d4e)e-j.e9e-d5rBe)e-j6e9e-d6rXe)e-j7Wn<e:yZ;z"e0d7e;e= than 2Gbz --quantizez/Quantize the neural network to be run with int8output)super__init__ add_argumentstrSUPPORTED_PIPELINESint)self __class__^/var/www/html/assistant/venv/lib/python3.9/site-packages/transformers/convert_graph_to_onnx.pyr3sJ z$OnnxConverterArgumentParser.__init__)__name__ __module__ __qualname____doc__r __classcell__r%r%r#r&r.sr)filename identifierreturncCs|j|j||jS)aE Append a string-identifier at the end (before the extension, if any) to the provided filepath Args: filename: pathlib.Path The actual path object we would like to add an identifier suffix identifier: The suffix to add Returns: String with concatenated identifier at the end of the filename )parentjoinpathstem with_suffixsuffix)r,r-r%r%r&generate_identified_filename\s r4)minimum_versioncCsVz6ddl}t|j}|tkr4td|jd|dWntyPtdYn0dS)z Check onnxruntime is installed and if the installed version match is recent enough Raises: ImportError: If onnxruntime is not installed or too old version is found rNz*We found an older version of onnxruntime (z&) but we require onnxruntime to be >= zp to enable all the conversions options. Please update onnxruntime by running `pip install --upgrade onnxruntime`zonnxruntime doesn't seem to be currently installed. Please install the onnxruntime by running `pip install onnxruntime` and relaunch the conversion.) onnxruntimer __version__ORT_QUANTIZE_MINIMUM_VERSION ImportError)r5r6Z ort_versionr%r%r&check_onnxruntime_requirementsis   r:cCs|td|jjj}gg}}|ddD]8}||vrN|||||q(t|dqbq(td||t|fS)a Ensure inputs are presented in the correct order, without any Non Args: model: The model used to forward the input data tokens: BatchEncoding holding the input data input_names: The name of the inputs Returns: Tuple z$Ensuring inputs are in correct orderNz, is not present in the generated input list.zGenerated inputs order: )printforward__code__ co_varnamesappendtuple)modeltokens input_namesZmodel_args_name model_argsordered_input_namesZarg_namer%r%r&ensure_valid_inputs    rG)nlp frameworkr.c stttdfdd |jd|d}|jjd|dkrH|jfi|n||}t|trd| }t|t t fsx|f}t | }fdd | D}g}|D](}t|t t fr||q||qd d tt|D}fd d t||D} t|fi| } ||| |fS) a? Attempt to infer the static vs dynamic axes for each input and output tensors for a specific model Args: nlp: The pipeline object holding the model to be exported framework: The framework identifier to dispatch to the correct inference scheme (pt/tf) Returns: - List of the inferred input variable names - List of the inferred output variable names - Dictionary with input/output variables names as key and shape tensor as value - a BatchEncoding reference which was used to infer all the above information )nameis_inputseq_lencst|ttfr&fdd|DSddt|jDddi}rvt|jdkr^d|d<qtd t|jd n*fd dt|jD}|t |dt d rd nddd||S)Ncsg|]}|qSr%r%).0t)build_shape_dictrKrJrLr%r& z:infer_shapes..build_shape_dict..cSsg|]\}}|dkr|qS)r;r%)rMZaxisZnumelr%r%r&rPrQrbatchsequencer;zUnable to infer tensor axes ()csg|]\}}|kr|qSr%r%)rMdimshape)rLr%r&rPrQzFound inputr z with shape: ) isinstancerAlist enumeraterWlen ValueErrorupdatedictfromkeysr<)rJtensorrKrLZaxesZseq_axes)rO)rKrJrLr&rOs "z&infer_shapes..build_shape_dictzThis is a sample output)Zreturn_tensorsrcs i|]\}}|||dqS)Tr%rMkvrOrLr%r& rQz infer_shapes..cSsg|]}d|qS)Zoutput_r%)rMir%r%r&rPrQz infer_shapes..cs i|]\}}|||dqS)Fr%rdrgr%r&rhrQ)rboolr! tokenizerZ input_idsrWrBrZr Zto_tupler[rAkeysitemsextendr@ranger]zipr`) rHrIrCoutputsZ input_varsZinput_dynamic_axesZ outputs_flatr output_namesZoutput_dynamic_axes dynamic_axesr%rgr& infer_shapess& "    rt) pipeline_namerIrBrkr.cKs`|dur |}|dkr"ts"td|dkr8ts8tdtd|d|dt|||||d S) a Convert the set of arguments provided through the CLI to an actual pipeline reference (tokenizer + model Args: pipeline_name: The kind of pipeline to use (ner, question-answering, etc.) framework: The actual model to convert the pipeline from ("pt" or "tf") model: The model name which will be loaded by the pipeline tokenizer: The tokenizer name which will be loaded by the pipeline, default to the model's value Returns: Pipeline object NrLCannot convert because PyTorch is not installed. Please install torch first.rLCannot convert because TF is not installed. Please install tensorflow first.zLoading pipeline (model: z , tokenizer: rU)rBrkrI model_kwargs)r Exceptionr r<r )rurIrBrkZ models_kwargsr%r%r&load_graph_from_argssrz)rHopsetruse_external_formatc Cststdddl}ddlm}td|j|Rt|d\}}}} t |j | |\} } ||j | | | ||d|dWdn1s0YdS) a Export a PyTorch backed pipeline to ONNX Intermediate Representation (IR Args: nlp: The pipeline to be exported opset: The actual version of the ONNX operator set to use output: Path where will be stored the generated ONNX model use_external_format: Split the model definition from its parameters to allow model bigger than 2GB Returns: rvrN)exportzUsing framework PyTorch: rT)frDrrrsZdo_constant_foldingZ opset_version) rrytorchZ torch.onnxr}r<r7Zno_gradrtrGrBas_posix) rHr{rr|rr}rDrrrsrCrFrEr%r%r&convert_pytorchs$   r)rHr{rc ststdtdzddlddl}ddlm}tdjjd|t|d\}}}}|j |j fd d | D} |j j|j | ||d \} } WnBty} z*td | jd | jd| WYd} ~ n d} ~ 00dS)av Export a TensorFlow backed pipeline to ONNX Intermediate Representation (IR) Args: nlp: The pipeline to be exported opset: The actual version of the ONNX operator set to use output: Path where will be stored the generated ONNX model Notes: TensorFlow cannot export model bigger than 2GB due to internal constraint from TensorFlow rwzD/!\ Please note TensorFlow doesn't support exporting model > 2Gb /!\rN)r7zUsing framework TensorFlow: z , tf2onnx: rcs g|]\}}jj||dqS))rJ)Z TensorSpecZ from_tensor)rMkeyrbrr%r&rPCrQz&convert_tensorflow..)r{Z output_pathzCannot import z6 required to convert TF model to ONNX. Please install z first. )r ryr<Z tensorflowtf2onnxr7versionVERSIONrtrBZpredictdatarmconvertZ from_kerasrr9rJ) rHr{rrZt2ovrDrrrsrCZinput_signatureZ model_proto_er%rr&convert_tensorflow&s$  rFr)rIrBrr{rkr|ruc Kstdttd|t||||fi|}|jsZtd|jt|jn,t t |jdkrt d|jd|dkrt ||||n t |||dS) a Convert the pipeline object to the ONNX Intermediate Representation (IR) format Args: framework: The framework the pipeline is backed by ("pt" or "tf") model: The name of the model to load for the pipeline output: The path where the ONNX graph will be stored opset: The actual version of the ONNX operator set to use tokenizer: The name of the model to load for the pipeline, default to the model's name if not provided use_external_format: Split the model definition from its parameters to allow model bigger than 2GB (PyTorch only) pipeline_name: The kind of pipeline to instantiate (ner, question-answering, etc.) model_kwargs: Keyword arguments to be forwarded to the model constructor Returns: zoThe `transformers.convert_graph_to_onnx` package is deprecated and will be removed in version 5 of TransformerszONNX opset version set to: zCreating folder rzFolder z" is not empty, aborting conversionrN)warningswarn FutureWarningr<rzr/existsrrr]rryrr) rIrBrr{rkr|rurxrHr%r%r&rNs r)onnx_model_pathr.cCsTddlm}m}t|d}|}||_|||}td|dtd|S)a> Load the model at the specified path and let onnxruntime look at transformations on the graph to enable all the optimizations possible Args: onnx_model_path: filepath where the model binary description is stored Returns: Path where the optimized model binary description has been saved rInferenceSessionSessionOptionsz -optimizedz$Optimized model has been written at : ✔zY/!\ Optimized model contains hardware specific operators which might not be portable. /!\)r6rrr4rZoptimized_model_filepathr<)rrrZopt_model_pathZ sess_optionrr%r%r&optimizes   rc Csddl}ddl}ddlm}ddlm}ddlm}ddlm }| | }t |j t dkrhtd|}||t |j t d kr||d d |jd d d dddt|d } n$||d d |jd d d dddt|d } | t|d} td| d|| jj| | S)z Quantize the weights of the model from float32 to in8 to allow very efficient inference on modern CPU Args: onnx_model_path: Path to location the exported ONNX model is stored Returns: The Path generated for the quantized rN) ModelProto)QuantizationMode) ONNXQuantizer)IntegerOpsRegistryz1.5.0zpModels larger than 2GB will fail to quantize due to protobuf constraint. Please upgrade to onnxruntime >= 1.5.0.z1.13.1FT) rB per_channel reduce_rangemodestatic weight_qTypeZ input_qType tensors_rangenodes_to_quantizenodes_to_excludeop_types_to_quantize) rBrrrrrZactivation_qTyperrrrz -quantizedz$Quantized model has been written at r)onnxr6Z onnx.onnx_pbrZonnxruntime.quantizationrZ'onnxruntime.quantization.onnx_quantizerrZ!onnxruntime.quantization.registryrloadrrr7r<ZCopyFromZ IntegerOpsr[Zquantize_modelr4Z save_modelrB) rrr6rrrrZ onnx_modelZ copy_modelZ quantizerZquantized_model_pathr%r%r&quantizesZ       r)pathc Csddlm}m}ddlm}td|dz.|}|||dgd}td|d Wn4|y}ztd |d WYd}~n d}~00dS) Nrr)RuntimeExceptionz"Checking ONNX model loading from: z ...ZCPUExecutionProvider) providerszModel u correctly loaded: ✔zError while loading the model u: ✘)r6rrZ+onnxruntime.capi.onnxruntime_pybind11_staterr<r)rrrrZ onnx_optionsrrer%r%r&verifys r__main__z' ====== Converting model to ONNX ======raV Using TensorFlow might not provide the same optimization level compared to PyTorch. For TensorFlow users you can try optimizing the model directly through onnxruntime_tools. For more information, please refer to the onnxruntime documentation: https://github.com/microsoft/onnxruntime/tree/master/onnxruntime/python/tools/transformers z$ ====== Optimizing ONNX model ======z+ ====== Check exported ONNX model(s) ======optimized_outputquantized_outputz"Error while converting the model: r;)N)NFr)=rargparserosrrpathlibrtypingrZpackaging.versionrrZtransformers.pipelinesr r Ztransformers.tokenization_utilsr Ztransformers.utilsr r rr8r rrr4r:rGrAr[r`rtrzr!rjrrrrrrr'parser parse_argsargsrabsoluter<rIrBr{rkr|rrZ check_loadinghasattrryrexitr%r%r%r&s     . &A  %- 2K