h''SSKJr SSKrSSKJr SSKrSSKJrJr SSK J r SSK J r SS /S /S 4SS jjr SSS jjrg)) annotationsN)Path) IS_JETSONLOGGER) TORCH_2_4) torch2imximagesoutput0Fc [(aSS0O0n[RR"UUU4SUSUUU=(d SS.UD6 g)a8 Export a PyTorch model to ONNX format. Args: torch_model (torch.nn.Module): The PyTorch model to export. im (torch.Tensor): Example input tensor for the model. onnx_file (str): Path to save the exported ONNX file. opset (int): ONNX opset version to use for export. input_names (list[str]): List of input tensor names. output_names (list[str]): List of output tensor names. dynamic (bool | dict, optional): Whether to enable dynamic axes. Notes: Setting `do_constant_folding=True` may cause issues with DNN inference for torch>=1.12. dynamoFTN)verbose opset_versiondo_constant_folding input_names output_names dynamic_axes)rtorchonnxexport) torch_modelim onnx_fileopsetrrdynamickwargss [/home/james-whalen/.local/lib/python3.13/site-packages/ultralytics/utils/export/__init__.py torch2onnxrsV0#,)h F JJ   !_   c l ^^^!SSKm!U=(d [U5RS5nT!RT!RR5n U (a%T!RR R U lT!RU 5n U R5n[T=(d SS-5n[T!RRSS5S5S:nU(a-US:a'URT!RRU5 O US:aXlS[T!R"R$5-nU R'U5nU R(=(a UnU R*=(a UnTb[,(d [/S5e[0R2"U S TS 35 U(dU(d [/S 5eT!R4R6Ul[T5UlUR=T!R>R@5 T!RCUU 5nUREU5(d[GS U35e[IURJ5Vs/sHnURMU5PM nn[IURN5Vs/sHnURQU5PM nnUHBn[0R2"U S URRSURTSURV35 MD UHBn[0R2"U SURRSURTSURV35 MD U(a~U RY5nSUSSS4n/USSQU4SjUSS5Q7nUHnUR[URRUUUS9 M! UR]U5 U(aUR_U5 [0R2"U SU(aSO SU(aSOS-SU35 U(aUR=T!R>R`5 T!RbRdUl3"UU!4SjST!Rh5nU"U[k[U5RS55S9Ul6O,U(a%UR=T!R>Rn5 U(a U RpO U RrnU"UU5n[uUS5nU b^[vRx"U 5n UR{[}U 5RS S!S"S#95 UR{U R55 UR{U(aUOUR55 SSS5 SSS5 gs snfs snf!,(df  N!=f!,(df  g=f)$a Export a YOLO model to TensorRT engine format. Args: onnx_file (str): Path to the ONNX file to be converted. engine_file (str, optional): Path to save the generated TensorRT engine file. workspace (int, optional): Workspace size in GB for TensorRT. half (bool, optional): Enable FP16 precision. int8 (bool, optional): Enable INT8 precision. dynamic (bool, optional): Enable dynamic input shapes. shape (tuple[int, int, int, int], optional): Input shape (batch, channels, height, width). dla (int, optional): DLA core to use (Jetson devices only). dataset (ultralytics.data.build.InfiniteDataLoader, optional): Dataset for INT8 calibration. metadata (dict, optional): Metadata to include in the engine file. verbose (bool, optional): Enable verbose logging. prefix (str, optional): Prefix for log messages. Raises: ValueError: If DLA is enabled on non-Jetson devices or required precision is not set. RuntimeError: If the ONNX file cannot be parsed. Notes: TensorRT version compatibility is handled for workspace size and engine building. INT8 calibration requires a dataset and generates a calibration cache. Metadata is serialized and written to the engine file if provided. rNz.enginei@.r z.DLA is only available on NVIDIA Jetson devicesz enabling DLA on core z...ztDLA requires either 'half=True' (FP16) or 'int8=True' (INT8) to be enabled. Please enable one of them and try again.zfailed to load ONNX file: z input "z " with shape z output " c3f># UH&n[[ST=(d S5U-5v M( g7f)r&N)intmax).0d workspaces r onnx2engine..s)"VIq3s1in1'='A#B#BIs.1)minoptr)z building INT8FP1632z engine as cj>\rSrSrSrS S UU4SjjjrS SjrSSjrSSjrSSjr SSjr S r g )%onnx2engine..EngineCalibratoraQ Custom INT8 calibrator for TensorRT engine optimization. This calibrator provides the necessary interface for TensorRT to perform INT8 quantization calibration using a dataset. It handles batch generation, caching, and calibration algorithm selection. Attributes: dataset: Dataset for calibration. data_iter: Iterator over the calibration dataset. algo (trt.CalibrationAlgoType): Calibration algorithm type. batch (int): Batch size for calibration. cache (Path): Path to save the calibration cache. Methods: get_algorithm: Get the calibration algorithm to use. get_batch_size: Get the batch size to use for calibration. get_batch: Get the next batch to use for calibration. read_calibration_cache: Use existing cache instead of calibrating again. write_calibration_cache: Write calibration cache to disk. c>TRRU5 Xl[U5UlTbTR R OTR RUlURUl [U5Ul g)z;Initialize the INT8 calibrator with dataset and cache path.N) IInt8Calibrator__init__datasetiter data_iterCalibrationAlgoTypeENTROPY_CALIBRATION_2MINMAX_CALIBRATIONalgo batch_sizebatchrcache)selfr;rDdlatrts rr:.onnx2engine..EngineCalibrator.__init__sq ##,,T2& !%g++AA00CC %// !%[ r cUR$)z%Get the calibration algorithm to use.)rArEs r get_algorithm3onnx2engine..EngineCalibrator.get_algorithms yy r c,UR=(d S$)z*Get the batch size to use for calibration.r)rCrJs rget_batch_size4onnx2engine..EngineCalibrator.get_batch_sizeszzQ&r c[UR5SS- nURRS:XaUR S5OUn[ UR 55/$![a gf=f)zOGet the next batch to use for calibration, as a list of device memory pointers.imggo@cpucudaN)nextr=devicetypetor(data_ptr StopIteration)rEnamesim0ss r get_batch/onnx2engine..EngineCalibrator.get_batchse /6>D.2kk.>.>%.G4776?TD 011$  sA!A$$ A10A1cURR5(a5URRS:XaURR5$gg)zSUse existing cache instead of calibrating again, otherwise, implicitly return None..cacheN)rDexistssuffix read_bytesrJs rread_calibration_cache.EngineCalibrator.read_calibration_cachesB::$$&&4::+<+<+H::0022,I&r c:URRU5ng)z Write calibration cache to disk.N)rD write_bytes)rErD_s rwrite_calibration_cache=onnx2engine..EngineCalibrator.write_calibration_cachesJJ**51r )rArCrDr=r;N))rDstrreturnNone)rlztrt.CalibrationAlgoType)rlr()rlzlist[int] | None)rlz bytes | None)rDbytesrlrm) __name__ __module__ __qualname____firstlineno____doc__r:rKrNr\rcrh__static_attributes__)rFrGsrEngineCalibratorr6sD 0  ) )  ) )" ! '  3  2r rur_)r;rDwblittleT) byteordersigned)Btensorrtr with_suffixLoggerINFOSeverityVERBOSE min_severityBuildercreate_builder_configr( __version__splitset_memory_pool_limitMemoryPoolType WORKSPACEmax_workspace_sizeNetworkDefinitionCreationFlagEXPLICIT_BATCHcreate_networkplatform_has_fast_fp16platform_has_fast_int8r ValueErrorrinfo DeviceTypeDLAdefault_device_typeDLA_coreset_flag BuilderFlag GPU_FALLBACK OnnxParserparse_from_file RuntimeErrorrange num_inputs get_input num_outputs get_outputnameshapedtypecreate_optimization_profile set_shapeadd_optimization_profileset_calibration_profiler1ProfilingVerbosityDETAILEDprofiling_verbosityr9rkint8_calibratorFP16build_serialized_network build_engineopenjsondumpswritelento_bytesencode serialize)"r engine_filer,halfint8rrrFr;metadatarprefixloggerbuilderconfigworkspace_bytesis_trt10flagnetworkparseriinputsoutputsinpoutprofile min_shape max_shaperubuildenginetmetarGs" ` ` @r onnx2enginer7sPGi!g67O3??((a034:HOa'$$S%7%7%A%A?S 1 $3! C55DDE ED$$T*G  ) ) 2dD  ) ) 2dD yMN N vh4SE=>DG &)^^%7%7"c(445^^GV ,F  ! !) , ,7 {CDD-2'2D2D,E F,Eqg",EF F.3G4G4G.HI.Hw!!!$.HGI vhhsxxj SYYKq TU vhizcii[#))UV557a"b) WeBQiW"VERSRTI"VW C   chhI5i  P''0   * *7 3 KK6(*tVSW9X$YYdepdqrs ,,-%(%;%;%D%D"@ 2@ 2s22@ 2F"2d9o11(;<"  ,,-19G , ,g>R>RE w 64 T+Ba  ::h'D GGCI&&qHT&J K GGDKKM " ((8(8(:;,C  GGID,C+B  s1'V V" V%/B V9V% V" V%% V3)rztorch.nn.Modulerz torch.Tensorrrkrr(r list[str]rrrz bool | dictrlrm) NNFFF)rrNNNFrj)rrkrz str | Noner, int | Nonerboolrrrrrztuple[int, int, int, int]rFrrz dict | Nonerrrrkrlrm) __future__rrpathlibrrultralytics.utilsrrultralytics.utils.torch_utilsrimxr rrr rrs)#  /3&Z(k $ $$$  $  $  $$ $R# '7  {<{<{<{<  {<  {<  {< %{< {<{<{< {< {