Ë
    ÓGìj`Ä  ã                   óÌ  — d dl Z d dlZd dlZd dlZd dlZd dlmZ d dlmZ d dl	m
Z
mZmZmZmZ d dlZd dlZd dlmZmZmZmZ d dlZddlmZmZmZ dej6                  d	ej6                  d
ej6                  fd„Z	 	 d-dej6                  d	ej6                  dee   ded
ej6                  f
d„Z G d„ d«      Z  G d„ d«      Z! G d„ de«      Z" G d„ de jF                  ¬«      Z$ G d„ d«      Z% G d„ de%«      Z& G d„ de%«      Z' G d„ de'«      Z( G d „ d!e'«      Z) G d"„ d#e'«      Z* G d$„ d%e jF                  ¬«      Z+ G d&„ d'e+«      Z,dd(e"jZ                  d)i fd*ee.ef   d+eee.      fd,„Z/y).é    N)ÚEnum)ÚPath)ÚDictÚOptionalÚSequenceÚTupleÚUnion)Ú
ModelProtoÚTensorProtoÚhelperÚnumpy_helperé   )Ú
apply_plotÚload_model_with_shape_inferÚsmooth_distributionÚpkÚqkÚreturnc                 ó  — t        j                  | j                  | j                  ¬«      }| dd t        j                  | dd |dd z  «      z  |dd | dk(  |dk\  z  }d||<   | dkD  |dkD  z  }t         j
                  || <   |S )z‘
    See https://docs.scipy.org/doc/scipy/reference/generated/scipy.special.rel_entr.html#scipy.special.rel_entr.
    Python implementation.
    ©ÚdtypeNr   )ÚnpÚemptyÚshaper   ÚlogÚinf)r   r   ÚresÚc2Úc1s        úfC:\Users\Stabi\Downloads\virtual-tryon(1)\venv\Lib\site-packages\onnxruntime/quantization/calibrate.pyÚrel_entrr!      s„   € ô
 �(‰(�2—8‘8 2§8¡8Ô
,€CØ‘ˆU”R—V‘V˜B™q˜E B¡q E™MÓ*Ñ*€C‰€FØ
�‰'�b˜A‘gÑ	€BØ€Cˆ�GØ
ˆq‰&�R˜!‘VÑ	€BÜ�v‰v€Cˆˆ�HØ€Jó    ÚbaseÚaxisc                 óR  — |�|dkD  sJ d«       ‚|€J d«       ‚t        j                  | «      j                  t         j                  «      } d| z  t        j                  | |d¬«      z  } t        j                  |«      j                  t         j                  «      }t        j
                  | |«      \  } }d|z  t        j                  ||d¬«      z  }t        | |«      }t        j                  ||¬«      }|�|t        j                  |«      z  }|j                  | j                  «      S )zÉ
    Simplifeied version of entropy.
    Source: https://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.entropy.html.
    This avoids taking a dependency on scipy just for this function.
    r   z0base={base} must be a positive number or `None`.z
qk is Noneç      ð?T)r$   Úkeepdims©r$   )	r   ÚasarrayÚastypeÚfloat32ÚsumÚbroadcast_arraysr!   r   r   )r   r   r#   r$   ÚvecÚss         r    Úentropyr0   '   sõ   € ð ˆ<˜4 !š8ÐWÐ%WÓWÐ#Øˆ>Ð'˜<Ó'ˆ>ä	�‰�B‹×	Ñ	œrŸz™zÓ	*€BØ	ˆr‰”B—F‘F˜2 D°4Ô8Ñ	8€Bä	�‰�B‹×	Ñ	œrŸz™zÓ	*€BÜ× Ñ   RÓ(�F€BˆØ	ˆr‰”B—F‘F˜2 D°4Ô8Ñ	8€BÜ
�2�rÓ
€Cä
�‰ˆs˜Ô€AØÐØ	ŒR�V‰V�D‹\ÑˆØ�8‰8�B—H‘HÓÐr"   c                   óZ   — e Zd Z eg d¢«      Z eg d¢«      Zd„ Zed„ «       Zed„ «       Z	y)Ú
TensorData)ÚavgÚstdÚlowestÚhighestÚhistÚ
hist_edgesÚbins)r3   r4   r5   r6   r8   c                 ó°  — |j                  «       D ]Ã  \  }}|t        j                  vr t        d|›dt        j                  › d�«      ‚|t        j                  v rmt        |d«      st        dt        |«      › d|›�«      ‚|j                  t        j                  t        j                  fvrt        d|j                  › d|›�«      ‚t        | ||«       ŒÅ y )NzUnexpected value z not in Ú.r   úUnexpected type z for k=zUnexpected dtype )Úitemsr2   Ú_allowedÚ
ValueErrorÚ_floatsÚhasattrÚtyper   r   Úfloat16r+   Úsetattr)ÚselfÚkwargsÚkÚvs       r    Ú__init__zTensorData.__init__G   s¾   € Ø—L‘L–N‰DˆAˆqØœ
×+Ñ+Ñ+Ü Ð#4°Q°E¸Ä*×BUÑBUÐAVÐVWÐ!XÓYÐYØ”J×&Ñ&Ñ&Ü˜q 'Ô*Ü$Ð'7¼¸Q»°yÀÈÀuÐ%MÓNÐNØ—7‘7¤2§:¡:¬r¯z©zÐ":Ñ:Ü$Ð'8¸¿¹¸	ÀÈÈÐ%NÓOÐOÜ�D˜!˜QÕñ #r"   c                 ó’   — t        | d«      rt        | d«      st        dt        | «      › d�«      ‚| j                  | j                  fS )Nr5   r6   z0Attributes 'lowest' and/or 'highest' missing in r;   )rA   ÚAttributeErrorÚdirr5   r6   ©rE   s    r    Úrange_valuezTensorData.range_valueR   sF   € ä�t˜XÔ&¬g°d¸IÔ.FÜ Ð#SÔTWÐX\ÓT]ÐS^Ð^_Ð!`ÓaÐaØ—‘˜TŸ\™\Ð*Ð*r"   c                 ó’   — t        | d«      rt        | d«      st        dt        | «      › d�«      ‚| j                  | j                  fS )Nr3   r4   z)Attributes 'avg' and/or 'std' missing in r;   )rA   rK   rL   r3   r4   rM   s    r    Úavg_stdzTensorData.avg_stdX   sC   € ä�t˜UÔ#¬7°4¸Ô+?Ü Ð#LÌSÐQUËYÈKÐWXÐ!YÓZÐZØ—‘˜$Ÿ(™(Ð#Ð#r"   N)
Ú__name__Ú
__module__Ú__qualname__Ú	frozensetr>   r@   rI   ÚpropertyrN   rP   © r"   r    r2   r2   C   sE   „ ÙÒZÓ[€HÙÒIÓJ€Gò	 ð ñ+ó ð+ð
 ñ$ó ñ$r"   r2   c                   óJ   — e Zd Zdeeeeef   f   fd„Zd„ Z	d„ Z
d„ Zd„ Zd„ Zy)	ÚTensorsDataÚdatac           
      ób  — || _         i | _        |j                  «       D �]  \  }}t        |t        «      st        dt        |«      › d�«      ‚t        |t        «      r˜|t        j                  k(  r/t        |«      dk(  r!t        |d   |d   ¬«      | j                  |<   Œ�t        |«      dk(  r)t        |d   |d   |d   |d   ¬	«      | j                  |<   Œ¸t        d
|d›dt        |«      › d|› d�«      ‚t        |t        «      st        dt        |«      › d�«      ‚|| j                  |<   �Œ y )NzKeys must be strings not r;   é   r   r   )r5   r6   é   é   )r5   r6   r7   r9   zUnexpected tuple for Úrz	, it has z elements: zValues must be TensorData not )Úcalibration_methodrY   r=   Ú
isinstanceÚstrÚ	TypeErrorrB   ÚtupleÚCalibrationMethodÚMinMaxÚlenr2   )rE   r_   rY   rG   rH   s        r    rI   zTensorsData.__init__`   s  € Ø"4ˆÔØˆŒ	Ø—J‘J—L‰DˆAˆqÜ˜a¤Ô%ÜÐ";¼DÀ»G¸9ÀAÐ FÓGÐGÜ˜!œUÔ#Ø%Ô):×)AÑ)AÒAÄcÈ!ÃfÐPQÂkÜ#-°Q°q±TÀ1ÀQÁ4Ô#H�D—I‘I˜a‘LØÜ�q“6˜Q’;Ü#-°Q°q±TÀ1ÀQÁ4ÈaÐPQÉdÐYZÐ[\ÑY]Ô#^�D—I‘I˜a‘LØÜÐ"7¸¸!°u¸IÄcÈ!ÃfÀXÈ[ÐYZÐX[Ð[\Ð ]Ó^Ð^Ü˜a¤Ô,ÜÐ"@ÄÀaÃÀ	ÈÐ KÓLÐLØˆD�I‰I�a‹Lñ !r"   c              #   ó8   K  — | j                   E d {  –—†  y 7 Œ­w©N©rY   rM   s    r    Ú__iter__zTensorsData.__iter__r   s   è ø€ Ø—9‘9×Òús   ‚’“c                 ó   — || j                   v S rh   ri   ©rE   Úkeys     r    Ú__contains__zTensorsData.__contains__u   s   € Ø�d—i‘iÐÐr"   c                 ó    — | j                   |   S rh   ri   rl   s     r    Ú__getitem__zTensorsData.__getitem__x   s   € Ø�y‰y˜‰~Ðr"   c                 ó\   — || j                   vrt        d|›d�«      ‚|| j                   |<   y )Nz)Only an existing tensor can be modified, z is not.)rY   ÚRuntimeError)rE   rm   Úvalues      r    Ú__setitem__zTensorsData.__setitem__{   s1   € Ø�d—i‘iÑÜÐ!JÈ3È'ÐQYÐZÓ[Ð[Øˆ�	‰	�#Šr"   c                 ó6   — | j                   j                  «       S rh   )rY   ÚvaluesrM   s    r    rv   zTensorsData.values€   s   € Ø�y‰y×ÑÓ!Ð!r"   N)rQ   rR   rS   r   ra   r	   r2   r   rI   rj   rn   rp   rt   rv   rV   r"   r    rX   rX   _   s;   „ ð°°c¸5ÀÈUÐARÑ;SÐ6SÑ1Tó ò$ò òòó
"r"   rX   c                   ó   — e Zd ZdZdZdZdZy)rd   r   r   r[   r]   N)rQ   rR   rS   re   ÚEntropyÚ
PercentileÚDistributionrV   r"   r    rd   rd   „   s   „ Ø€FØ€GØ€JØ�Lr"   rd   c                   óR   — e Zd Zed„ «       Zej                  defd„«       Zd„ Z	d„ Z
y)ÚCalibrationDataReaderc                 óX   — t        |d«      xr t        |j                  «      xs t        S )NÚget_next)rA   Úcallabler~   ÚNotImplemented)ÚclsÚsubclasss     r    Ú__subclasshook__z&CalibrationDataReader.__subclasshook__Œ   s%   € ä�x Ó,ÒL´¸(×:KÑ:KÓ1LÒ^ÔP^Ð^r"   r   c                 ó   — t         ‚)z9generate the input data dict for ONNXinferenceSession run©ÚNotImplementedErrorrM   s    r    r~   zCalibrationDataReader.get_next�   s
   € ô "Ð!r"   c                 ó   — | S rh   rV   rM   s    r    rj   zCalibrationDataReader.__iter__•   s   € Øˆr"   c                 ó6   — | j                  «       }|€t        ‚|S rh   )r~   ÚStopIteration)rE   Úresults     r    Ú__next__zCalibrationDataReader.__next__˜   s   € Ø—‘“ˆØˆ>ÜÐØˆr"   N)rQ   rR   rS   Úclassmethodrƒ   ÚabcÚabstractmethodÚdictr~   rj   r‹   rV   r"   r    r|   r|   ‹   sA   „ Øñ_ó ð_ð 	×Ñð"˜$ò "ó ð"òór"   r|   )Ú	metaclassc                   ó~   — e Zd Z	 	 	 	 ddeeef   deee      fd„Zdgfd„Z	d„ Z
defd	„Zd
„ Zd„ Zdefd„Zdefd„Zy)ÚCalibraterBaseNÚ
model_pathÚop_types_to_calibratec                 ó  — t        |t        «      rt        t        |«      «      | _        n,t        |t        «      rt        |«      | _        nt        d«      ‚|| _        || _        || _        || _	        d| _
        d| _        dg| _        y)aÓ  
        :param model_path: ONNX model to calibrate. It should be a model file path
        :param op_types_to_calibrate: operator types to calibrate. By default, calibrate all the float32/float16 tensors.
        :param augmented_model_path: save augmented model to this path.
        :param symmetric: make range of tensor symmetric (central point is 0).
        :param use_external_data_format: use external data format to store model which size is >= 2Gb
        z model_path should be model path.NÚCPUExecutionProvider)r`   ra   r   r   Úmodelr?   r”   Úaugmented_model_pathÚ	symmetricÚuse_external_data_formatÚaugment_modelÚinfer_sessionÚexecution_providers)rE   r“   r”   r˜   r™   rš   s         r    rI   zCalibraterBase.__init__    s{   € ô �j¤#Ô&Ü4´T¸*Ó5EÓFˆD�JÜ˜
¤DÔ)Ü4°ZÓ@ˆD�JäÐ?Ó@Ð@à%:ˆÔ"Ø$8ˆÔ!Ø"ˆŒØ(@ˆÔ%à!ˆÔØ!ˆÔØ$:Ð#;ˆÕ r"   r–   c                 ó2   — || _         | j                  «        y)zz
        reset the execution providers to execute the collect_data. It triggers to re-creating inference session.
        N)r�   Úcreate_inference_session)rE   r�   s     r    Úset_execution_providersz&CalibraterBase.set_execution_providers¿   s   € ð $7ˆÔ Ø×%Ñ%Õ'r"   c                 óÌ   — t        j                  «       }t         j                  j                  |_        t        j
                  | j                  || j                  ¬«      | _        y)z9
        create an OnnxRuntime InferenceSession.
        )Úsess_optionsÚ	providersN)	ÚonnxruntimeÚSessionOptionsÚGraphOptimizationLevelÚORT_DISABLE_ALLÚgraph_optimization_levelÚInferenceSessionr˜   r�   rœ   )rE   r¢   s     r    rŸ   z'CalibraterBase.create_inference_sessionÆ   sN   € ô #×1Ñ1Ó3ˆÜ0;×0RÑ0R×0bÑ0bˆÔ-Ü(×9Ñ9Ø×%Ñ%Ø%Ø×.Ñ.ô
ˆÕr"   r—   c                 óÐ  — |j                   j                  D �ci c]  }|j                  |“Œ }}|j                  |j                   j                  D �ci c]  }|j                  |“Œ c}«       |j                  |j                   j
                  D �ci c]  }|j                  |“Œ c}«       |j                   j                  D �ch c]  }|j                  ’Œ }}t        «       }t        j                  t        j                  h}	|j                   j                  D ]µ  }
| j                  r|
j                  | j                  v sŒ(t        j                  |
j
                  |
j                  «      D ]a  }||v sŒ||   }|j                   j#                  d«      sŒ)|j                   j$                  j&                  |	v sŒL||vsŒQ|j)                  |«       Œc Œ· ||fS c c}w c c}w c c}w c c}w )zÉ
        select input/output tensors of candidate nodes to calibrate.
        returns:
            tensors (set): set of tensor name.
            value_infos (dict): tensor name to value info.
        Útensor_type)ÚgraphÚ
value_infoÚnameÚupdateÚoutputÚinputÚinitializerÚsetr   ÚFLOATÚFLOAT16Únoder”   Úop_typeÚ	itertoolsÚchainrB   ÚHasFieldr«   Ú	elem_typeÚadd)rE   r—   ÚviÚvalue_infosÚotÚitÚinitr²   Útensors_to_calibrateÚtensor_type_to_calibrater¶   Útensor_names               r    Úselect_tensors_to_calibratez*CalibraterBase.select_tensors_to_calibrateÒ   sŒ  € ð .3¯[©[×-CÒ-CÓDÑ-C r�r—w‘w ‘{Ð-CˆÐDØ×Ñ°%·+±+×2DÒ2DÓEÑ2D¨B˜BŸG™G R™KÐ2DÑEÔFØ×Ñ°%·+±+×2CÒ2CÓDÑ2C¨B˜BŸG™G R™KÐ2CÑDÔEØ-2¯[©[×-DÒ-DÓEÑ-D T�t—y“yÐ-DˆÐEä"›uÐÜ$/×$5Ñ$5´{×7JÑ7JÐ#KÐ à—K‘K×$Ô$ˆDØ×-Ò-°·±À×A[ÑA[Ò1[Ü#,§?¡?°4·:±:¸t¿{¹{Ö#K�KØ" kÒ1Ø(¨Ñ5˜àŸG™G×,Ñ,¨]Õ;Ø!#§¡×!4Ñ!4×!>Ñ!>ÐBZÒ!ZØ!,°KÒ!?à0×4Ñ4°[ÕAñ $Lð %ð $ [Ð0Ð0ùò) EùÚEùÚDùÚEs   ™GÁGÂGÃG#c                 ó   — | j                   S )zP
        return: augmented onnx model. Call after calling augment_graph
        )r—   rM   s    r    Úget_augment_modelz CalibraterBase.get_augment_modelï   s   € ð �z‰zÐr"   c                 ó   — t         ‚)zï
        abstract method: augment the input model to prepare for collecting data. It will:
            1. augment the model to be able to collect desired statistics data
            2. save augmented model to augmented_model_paths
        r…   rM   s    r    Úaugment_graphzCalibraterBase.augment_graphõ   s
   € ô "Ð!r"   Údata_readerc                 ó   — t         ‚)z€
        abstract method: collect the tensors that will be used for range computation. It can be called multiple times.
        r…   )rE   rÊ   s     r    Úcollect_datazCalibraterBase.collect_dataý   ó
   € ô "Ð!r"   r   c                 ó   — t         ‚)ze
        abstract method: compute data based on the calibration method stored in TensorsData
        r…   rM   s    r    Úcompute_datazCalibraterBase.compute_data  rÍ   r"   )Núaugmented_model.onnxFF)rQ   rR   rS   r	   ra   r   r   r   rI   r    rŸ   r
   rÅ   rÇ   rÉ   r|   rÌ   rX   rÏ   rV   r"   r    r’   r’   Ÿ   s{   „ ð :>Ø3ØØ!&ñ<à˜#˜t˜)Ñ$ð<ð  (¨°©Ñ6ó<ð> <RÐ:Ró (ò

ð1°ó 1ò:ò"ð"Ð(=ó "ð"˜kô "r"   r’   c                   óx   ‡ — e Zd Z	 	 	 	 	 	 	 ddeeef   deee      fˆ fd„Zd„ Z	d„ Z
defd„Zd„ Zd	efd
„Zˆ xZS )ÚMinMaxCalibraterr“   r”   c	                 ó‚  •— t         ‰
| �  |||||¬«       g | _        d| _        t	        | j
                  j                  j                  «      | _        | j
                  j                  j                  D �	ch c]  }	|	j                  ’Œ c}	| _
        || _        |r|dk  s|dkD  rt        d«      ‚|| _        || _        yc c}	w )a/  
        :param model_path: ONNX model to calibrate. It is a model path
        :param op_types_to_calibrate: operator types to calibrate. By default, calibrate all the float32/float16 tensors.
        :param augmented_model_path: save augmented model to this path.
        :param symmetric: make range of tensor symmetric (central point is 0).
        :param use_external_data_format: use external data format to store model which size is >= 2Gb
        :param moving_average: compute the moving average of the minimum and maximum values instead of the global minimum and maximum.
        :param averaging_constant: constant smoothing factor to use when computing the moving average.
        :param max_intermediate_outputs: maximum number of intermediate outputs before an intermediate range is computed.
        ©r”   r˜   r™   rš   Nr   r   z;Invalid averaging constant, which should not be < 0 or > 1.)ÚsuperrI   Úintermediate_outputsÚcalibrate_tensors_rangerf   r—   r¬   r°   Únum_model_outputsr®   Úmodel_original_outputsÚmoving_averager?   Úaveraging_constantÚmax_intermediate_outputs)rE   r“   r”   r˜   r™   rš   rÚ   rÛ   rÜ   r°   Ú	__class__s             €r    rI   zMinMaxCalibrater.__init__  s¿   ø€ ô* 	‰ÑØØ"7Ø!5ØØ%=ð 	ô 	
ð %'ˆÔ!Ø'+ˆÔ$Ü!$ T§Z¡Z×%5Ñ%5×%<Ñ%<Ó!=ˆÔØAEÇÁ×AQÑAQ×AXÒAXÓ&YÑAX°v v§{£{ÐAXÑ&YˆÔ#Ø,ˆÔÙÐ1°AÒ5Ð9KÈaÒ9OÜÐZÓ[Ð[Ø"4ˆÔØ(@ˆÕ%ùò 'Zs   Á4B<c                 óú  ‡ ‡— ‰ j                  ‰ j                  «      \  }}t        t        j                  «       «      Št        j                  t        j                  dgt        j                  ¬«      ‰«      }‰ j                  j                  j                  j                  |«       ˆˆ fd„}|D ]  } ||d«        ||d«       Œ t        j                  ‰ j                  ‰ j                  ‰ j                   ¬«       y)zÙ
        Adds ReduceMin and ReduceMax nodes to all quantization_candidates op type nodes in
        model and ensures their outputs are stored as part of the graph output
        :return: augmented ONNX model
        r   r   c                 óÄ  •— d}| dz   |z   }|dz   }t         j                  j                  || g|g||¬«      }t         j                  j                  d|‰g|g|¬«      }‰j                  j                  j
                  j                  ||g«       ‰j                  j                  j                  D �ci c]  }|j                  |“Œ }}|j                  ‰j                  j                  j                  D �	ci c]  }	|	j                  |	“Œ c}	«       |j                  ‰j                  j                  j                  D �
ci c]  }
|
j                  |
“Œ c}
«       | |v r$||    j                  j                  j                  }nt        d| ›d�«      ‚‰j                  j                  j                  j!                  t        j"                  ||dg«      «       y c c}w c c}	w c c}
w )	Nr   Ú_Ú_Reshape)r'   r®   ÚReshape)ÚinputsÚoutputsr®   z'Unable to guess tensor type for tensor zE, running shape inference before quantization may resolve this issue.)Úonnxr   Ú	make_noder—   r¬   r¶   Úextendr­   r®   r¯   r°   r±   rB   r«   r»   r?   ÚappendÚmake_tensor_value_info)rÄ   Úreduce_op_namer'   Úreduce_outputÚintermediate_outputÚreduce_nodeÚreshape_noder½   r¾   ÚoÚiÚ	onnx_typeÚreshape_shape_namerE   s               €€r    Úadd_reduce_min_maxz:MinMaxCalibrater.augment_graph.<locals>.add_reduce_min_max<  sÆ  ø€ ð ˆHð (¨#Ñ-°Ñ>ˆMØ"/°*Ñ"<ÐÜŸ+™+×/Ñ/Ø  Ð0CÐ/DÈxÐ^kð 0ó ˆKô  Ÿ;™;×0Ñ0ØØ+Ð-?Ð@Ø&˜Ø(ð	 1ó ˆLð �J‰J×Ñ×!Ñ!×(Ñ(¨+°|Ð)DÔEØ15·±×1AÑ1A×1LÒ1LÓMÑ1L¨2˜2Ÿ7™7 B™;Ð1LˆKÐMØ×Ñ°4·:±:×3CÑ3C×3JÒ3JÓKÑ3J¨a §¡¨¡	Ð3JÑKÔLØ×Ñ°4·:±:×3CÑ3C×3IÒ3IÓJÑ3I¨a §¡¨¡	Ð3IÑJÔKØ˜kÑ)Ø'¨Ñ4×9Ñ9×EÑE×OÑO‘	ä Ø=¸k¸_ð MZð [óð ð �J‰J×Ñ×#Ñ#×*Ñ*¬6×+HÑ+HÈÐXaÐdeÐcfÓ+gÕhùò NùÚKùÚJs   Â0GÃ3GÄ;GÚ	ReduceMinÚ	ReduceMax©Úsave_as_external_dataN)rÅ   r—   ra   ÚuuidÚuuid4r   Ú
from_arrayr   ÚarrayÚint64r¬   r²   rè   rå   Úsaver˜   rš   )rE   Útensorsrà   Úreshape_shaperó   Útensorrò   s   `     @r    rÉ   zMinMaxCalibrater.augment_graph1  s¾   ù€ ð ×5Ñ5°d·j±jÓA‰
ˆ�Ü ¤§¡£Ó.ÐÜ$×/Ñ/´·±¸!¸ÄBÇHÁHÔ0MÐOaÓbˆØ�
‰
×Ñ×$Ñ$×+Ñ+¨MÔ:õ	ió@ ˆFÙ˜v {Ô3Ù˜v {Õ3ð ô 	�	‰	Ø�J‰JØ×%Ñ%Ø"&×"?Ñ"?ö	
r"   c                 ó   — g | _         y rh   ©rÖ   rM   s    r    Úclear_collected_dataz%MinMaxCalibrater.clear_collected_dataf  ó
   € Ø$&ˆÕ!r"   rÊ   c                 ó  — 	 |j                  «       }|snt| j                  j                  | j                  j	                  d |«      «       | j
                  �2t        | j                  «      | j
                  k(  r| j                  «        Œ‡t        | j                  «      dk(  r| j                  €t        d«      ‚| j                  «       }t        |t        «      st        dt        |«      › d�«      ‚| j                  «        y )Nr   úNo data is collected.z+compute_data must return a TensorsData not r;   )r~   rÖ   rè   rœ   ÚrunrÜ   rf   r  r×   r?   rÏ   r`   rX   rb   rB   )rE   rÊ   rã   Úts       r    rÌ   zMinMaxCalibrater.collect_datai  sà   € ØØ ×)Ñ)Ó+ˆFÙØØ×%Ñ%×,Ñ,¨T×-?Ñ-?×-CÑ-CÀDÈ&Ó-QÔRà×-Ñ-Ð9Ü˜×1Ñ1Ó2°d×6SÑ6SÒSà×)Ñ)Ô+ð ô ˆt×(Ñ(Ó)¨QÒ.°4×3OÑ3OÐ3WÜÐ4Ó5Ð5à×ÑÓˆÜ˜!œ[Ô)ÜÐIÌ$ÈqË'ÈÐRSÐTÓUÐUØ×!Ñ!Õ#r"   c                 ó>  — |s|S |j                  «       D ]…  \  }}| j                  rC|d   | j                  ||   d   |d   z
  z  z   }|d   | j                  ||   d   |d   z
  z  z   }n*t        |d   ||   d   «      }t	        |d   ||   d   «      }||f||<   Œ‡ |S )Nr   r   )r=   rÚ   rÛ   ÚminÚmax)rE   Ú	old_rangeÚ	new_rangerm   rs   Ú	min_valueÚ	max_values          r    Úmerge_rangezMinMaxCalibrater.merge_range}  sÏ   € ÙØÐà#Ÿ/™/Ö+‰JˆC�Ø×"Ò"Ø! !™H t×'>Ñ'>À)ÈCÁ.ÐQRÑBSÐV[Ð\]ÑV^ÑB^Ñ'_Ñ_�	Ø! !™H t×'>Ñ'>À)ÈCÁ.ÐQRÑBSÐV[Ð\]ÑV^ÑB^Ñ'_Ñ_‘	ä  a¡¨)°C©.¸Ñ*;Ó<�	Ü  a¡¨)°C©.¸Ñ*;Ó<�	Ø'¨Ð3ˆI�cŠNð ,ð Ðr"   r   c           	      óô  — t        | j                  «      dk(  r| j                  S t        t        | j                  d   «      «      D �cg c])  }| j                  j                  «       |   j                  ‘Œ+ }}| j                  D �cg c]  }t        t        ||«      «      ‘Œ }}i }|D ];  }|j                  «       D ]&  \  }}|j                  |g «      j                  |«       Œ( Œ= || j                  d }	t        dt        |	«      d«      D �cg c]  }|	|   j                  d«      d   ‘Œ }
}|D �ci c]  }|| j                  vsŒ|||   “Œ }}g }t        dt        |	«      d«      D �]  }| j                  r>t!        j"                  ||	|      d¬«      }t!        j"                  ||	|dz         d¬«      }n=t!        j$                  ||	|      d¬«      }t!        j&                  ||	|dz         d¬«      }| j(                  rPt'        t!        j*                  |«      t!        j*                  |«      «      }|j                  t-        | |g«      «       Œç|j                  t-        ||g«      «       �Œ t/        t0        j2                  t        t        |
|«      «      «      }| j                  r-| j5                  | j                  |«      | _        | j                  S || _        | j                  S c c}w c c}w c c}w c c}w )zŒ
        Compute the min-max range of tensor
        :return: dictionary mapping: {added node names: (ReduceMin, ReduceMax) pairs }
        r   Nr[   rà   r(   r   )rf   rÖ   r×   Úrangerœ   Úget_outputsr®   r�   Úzipr=   Ú
setdefaultrè   rØ   Ú
rpartitionrÙ   rÚ   r   Úmeanr
  r  r™   Úabsrc   rX   rd   re   r  )rE   rð   Úoutput_namesrì   Úoutput_dicts_listÚmerged_output_dictÚdrG   rH   Úadded_output_namesÚcalibrate_tensor_namesÚmerged_added_output_dictÚpairsÚmin_value_arrayÚmax_value_arrayÚmax_absolute_valueÚnew_calibrate_tensors_ranges                    r    rÏ   zMinMaxCalibrater.compute_dataŒ  s÷  € ô ˆt×(Ñ(Ó)¨QÒ.Ø×/Ñ/Ð/äJOÔPSÐTX×TmÑTmÐnoÑTpÓPqÔJrÓsÑJrÀQ˜×*Ñ*×6Ñ6Ó8¸Ñ;×@Ó@ÐJrˆÐsàTX×TmÒTmó
ÙTmÐ=PŒD”�\Ð#6Ó7Õ8ÐTmð 	ð 
ð  ÐÛ"ˆAØŸ™ž	‘��1Ø"×-Ñ-¨a°Ó4×;Ñ;¸AÕ>ñ "ð #ð *¨$×*@Ñ*@Ð*BÐCÐä>CÀAÄsÐK]ÓG^Ð`aÔ>bó"
Ù>b¸Ð˜qÑ!×,Ñ,¨SÓ1°!Ó4Ð>bð 	ð "
ñ
 /Aó$
Ù.@¨ÀAÈT×MhÑMhÒDhˆAÐ! !Ñ$Ñ$Ð.@ð 	!ð $
ð ˆÜ�qœ#Ð0Ó1°1×5ˆAØ×"Ò"Ü"$§'¡'Ð*BÐCUÐVWÑCXÑ*YÐ`aÔ"b�Ü"$§'¡'Ð*BÐCUÐVWÐZ[ÑV[ÑC\Ñ*]ÐdeÔ"f‘ä"$§&¡&Ð)AÐBTÐUVÑBWÑ)XÐ_`Ô"a�Ü"$§&¡&Ð)AÐBTÐUVÐYZÑUZÑB[Ñ)\ÐcdÔ"e�à�~Š~Ü%(¬¯©°Ó)@Ä"Ç&Á&ÈÓBYÓ%ZÐ"Ø—‘œUÐ%7Ð$7Ð9KÐ#LÓMÕNà—‘œU O°_Ð#EÓFÖGð 6ô '2Ô2C×2JÑ2JÌDÔQTÐUkÐmrÓQsÓLtÓ&uÐ#Ø×'Ò'Ø+/×+;Ñ+;¸D×<XÑ<XÐZuÓ+vˆDÔ(ð ×+Ñ+Ð+ð ,GˆDÔ(à×+Ñ+Ð+ùòO tùò
ùò"
ùò$
s   Á.K&ÂK+ÄK0Ä1K5ÅK5)NrÐ   FFFç{®Gáz„?N)rQ   rR   rS   r	   ra   r   r   r   rI   rÉ   r  r|   rÌ   r  rX   rÏ   Ú__classcell__©rÝ   s   @r    rÒ   rÒ   
  sp   ø„ ð :>Ø3ØØ!&ØØØ!%ñ$Aà˜#˜t˜)Ñ$ð$Að  (¨°©Ñ6õ$AòL3
òj'ð$Ð(=ó $ò(ð0,˜k÷ 0,r"   rÒ   c                   óv   ‡ — e Zd Z	 	 	 	 	 	 	 	 	 d
deeef   deee      fˆ fd„Zd„ Z	d„ Z
defd„Zdefd	„Zˆ xZS )ÚHistogramCalibraterr“   r”   c                 óŒ  •— t         ‰| �  |||||¬«       g | _        d| _        t	        | j
                  j                  j                  «      | _        | j
                  j                  j                  D �ch c]  }|j                  ’Œ c}| _
        d| _        || _        || _        || _        |	| _        d| _        |
| _        yc c}w )a=  
        :param model_path: ONNX model to calibrate. It is a model path.
        :param op_types_to_calibrate: operator types to calibrate. By default, calibrate all the float32/float16 tensors.
        :param augmented_model_path: save augmented model to this path.
        :param use_external_data_format: use external data format to store model which size is >= 2Gb
        :param method: A string. One of ['entropy', 'percentile'].
        :param symmetric: make range of tensor symmetric (central point is 0).
        :param num_bins: number of bins to create a new histogram for collecting tensor values.
        :param num_quantized_bins: number of quantized bins. Default 128.
        :param percentile: A float number between [0, 100]. Default 99.99.
        :param scenario: see :class:`DistributionCalibrater`
        rÔ   N)rÕ   rI   rÖ   r×   rf   r—   r¬   r°   rØ   r®   rÙ   Ú	collectorÚmethodÚnum_binsÚnum_quantized_binsÚ
percentilerÂ   Úscenario)rE   r“   r”   r˜   rš   r,  r™   r-  r.  r/  r0  r°   rÝ   s               €r    rI   zHistogramCalibrater.__init__À  s»   ø€ ô2 	‰ÑØØ"7Ø!5ØØ%=ð 	ô 	
ð %'ˆÔ!Ø'+ˆÔ$Ü!$ T§Z¡Z×%5Ñ%5×%<Ñ%<Ó!=ˆÔØAEÇÁ×AQÑAQ×AXÒAXÓ&YÑAX°v v§{£{ÐAXÑ&YˆÔ#ØˆŒØˆŒØ ˆŒØ"4ˆÔØ$ˆŒØ$(ˆÔ!Ø ˆ�ùò 'Zs   Á4Cc                 óZ  — | j                  | j                  «      \  | _        }| j                  D ]C  }|| j                  vsŒ| j                  j                  j
                  j                  ||   «       ŒE t        j                  | j                  | j                  | j                  ¬«       y)zƒ
        make all quantization_candidates op type nodes as part of the graph output.
        :return: augmented ONNX model
        rö   N)rÅ   r—   rÂ   rÙ   r¬   r°   rè   rå   rý   r˜   rš   )rE   r¾   r   s      r    rÉ   z!HistogramCalibrater.augment_graphì  s‹   € ð
 26×1QÑ1QÐRV×R\ÑR\Ó1]Ñ.ˆÔ! ;Ø×/Ô/ˆFØ˜T×8Ñ8Ò8Ø—
‘
× Ñ ×'Ñ'×.Ñ.¨{¸6Ñ/BÕCð 0ô 	�	‰	Ø�J‰JØ×%Ñ%Ø"&×"?Ñ"?ö	
r"   c                 ó   — g | _         y rh   r  rM   s    r    r  z(HistogramCalibrater.clear_collected_dataü  r  r"   rÊ   c           	      óÔ  — 	 |j                  «       }|sn6| j                  j                  | j                  j	                  d|«      «       ŒIt        | j                  «      dk(  rt        d«      ‚t        t        | j                  d   «      «      D �cg c])  }| j                  j                  «       |   j                  ‘Œ+ }}| j                  D �cg c]  }t        t        ||«      «      ‘Œ }}i }|D ];  }|j                  «       D ]&  \  }	}
|j                  |	g «      j                  |
«       Œ( Œ= |D �ci c]  }|| j                  v sŒ|||   “Œ }}| j                  sRt!        | j"                  | j$                  | j&                  | j(                  | j*                  | j,                  ¬«      | _        | j                  j/                  |«       | j1                  «        yc c}w c c}w c c}w )zy
        Entropy Calibrator collects operators' tensors as well as generates tensor histogram for each operator.
        Nr   r  )r,  r™   r-  r.  r/  r0  )r~   rÖ   rè   rœ   r  rf   r?   r  r  r®   r�   r  r=   r  rÂ   r+  ÚHistogramCollectorr,  r™   r-  r.  r/  r0  Úcollectr  )rE   rÊ   rã   rð   r  rì   r  Úmerged_dictr  rG   rH   Úclean_merged_dicts               r    rÌ   z HistogramCalibrater.collect_dataÿ  sÀ  € ð Ø ×)Ñ)Ó+ˆFÙØØ×%Ñ%×,Ñ,¨T×-?Ñ-?×-CÑ-CÀDÈ&Ó-QÔRð	 ô ˆt×(Ñ(Ó)¨QÒ.ÜÐ4Ó5Ð5äJOÔPSÐTX×TmÑTmÐnoÑTpÓPqÔJrÓsÑJrÀQ˜×*Ñ*×6Ñ6Ó8¸Ñ;×@Ó@ÐJrˆÐsàTX×TmÒTmó
ÙTmÐ=PŒD”�\Ð#6Ó7Õ8ÐTmð 	ð 
ð ˆÛ"ˆAØŸ™ž	‘��1Ø×&Ñ& q¨"Ó-×4Ñ4°QÕ7ñ "ð #ñ 9DÓf¹°1ÀqÈD×LeÑLeÒGe˜Q ¨A¡Ñ.¸ÐÐfà�~Š~Ü/Ø—{‘{ØŸ.™.ØŸ™Ø#'×#:Ñ#:ØŸ?™?ØŸ™ôˆDŒNð 	�‰×ÑÐ0Ô1à×!Ñ!Õ#ùò/ tùò
ùò gs   Â.GÃG Ä3G%ÅG%r   c                 ón  — | j                   st        d«      ‚t        | t        «      rt        j
                  }nZt        | t        «      rt        j                  }n9t        | t        «      rt        j                  }nt        dt        | «      › d�«      ‚t        || j                   j                  «       «      S )z€
        Compute the min-max range of tensor
        :return: dictionary mapping: {tensor name: (min value, max value)}
        z9No collector created and can't generate calibration data.zUnknown calibrater z". This method must be overwritten.)r+  r?   r`   ÚEntropyCalibraterrd   rx   ÚPercentileCalibraterry   ÚDistributionCalibraterrz   rb   rB   rX   Úcompute_collection_result)rE   Úcals     r    rÏ   z HistogramCalibrater.compute_data%  s�   € ð
 �~Š~ÜÐXÓYÐYä�dÔ-Ô.Ü#×+Ñ+‰CÜ˜Ô2Ô3Ü#×.Ñ.‰CÜ˜Ô4Ô5Ü#×0Ñ0‰CäÐ1´$°t³*°Ð=_Ð`ÓaÐaÜ˜3 §¡× HÑ HÓ JÓKÐKr"   )	NrÐ   Fr/  Fé€   é   ç-²�ïÿX@Úsame)rQ   rR   rS   r	   ra   r   r   r   rI   rÉ   r  r|   rÌ   rX   rÏ   r&  r'  s   @r    r)  r)  ¿  sp   ø„ ð :>Ø3Ø!&ØØØØØØñ*!à˜#˜t˜)Ñ$ð*!ð  (¨°©Ñ6õ*!òX
ò 'ð$$Ð(=ó $$ðLL˜k÷ Lr"   r)  c                   óN   ‡ — e Zd Z	 	 	 	 	 	 	 ddeeef   deee      fˆ fd„Zˆ xZ	S )r9  r“   r”   c	           
      ó4   •— t         ‰	| �  ||||||||¬«       y)aÄ  
        :param model_path: ONNX model to calibrate. It is a model path
        :param op_types_to_calibrate: operator types to calibrate. By default, calibrate all the float32/float16 tensors.
        :param augmented_model_path: save augmented model to this path.
        :param use_external_data_format: use external data format to store model which size is >= 2Gb
        :param method: A string. One of ['entropy', 'percentile', 'distribution'].
        :param symmetric: make range of tensor symmetric (central point is 0).
        :param num_bins: number of bins to create a new histogram for collecting tensor values.
        :param num_quantized_bins: number of quantized bins. Default 128.
        )r,  r™   r-  r.  N©rÕ   rI   )
rE   r“   r”   r˜   rš   r,  r™   r-  r.  rÝ   s
            €r    rI   zEntropyCalibrater.__init__9  s/   ø€ ô* 	‰ÑØØ!Ø Ø$ØØØØ1ð 	õ 		
r"   )NrÐ   Fr0   Fr>  r>  ©
rQ   rR   rS   r	   ra   r   r   r   rI   r&  r'  s   @r    r9  r9  8  sH   ø„ ð :>Ø3Ø!&ØØØØñ
à˜#˜t˜)Ñ$ð
ð  (¨°©Ñ6÷
ñ 
r"   r9  c                   óN   ‡ — e Zd Z	 	 	 	 	 	 	 ddeeef   deee      fˆ fd„Zˆ xZ	S )r:  r“   r”   c	           
      ó4   •— t         ‰	| �  ||||||||¬«       y)a¯  
        :param model_path: ONNX model to calibrate. It is a model path
        :param op_types_to_calibrate: operator types to calibrate. By default, calibrate all the float32/float16 tensors.
        :param augmented_model_path: save augmented model to this path.
        :param use_external_data_format: use external data format to store model which size is >= 2Gb
        :param method: A string. One of ['entropy', 'percentile', 'distribution'].
        :param symmetric: make range of tensor symmetric (central point is 0).
        :param num_quantized_bins: number of quantized bins. Default 128.
        :param percentile: A float number between [0, 100]. Default 99.99.
        )r,  r™   r-  r/  NrD  )
rE   r“   r”   r˜   rš   r,  r™   r-  r/  rÝ   s
            €r    rI   zPercentileCalibrater.__init__[  s/   ø€ ô* 	‰ÑØØ!Ø Ø$ØØØØ!ð 	õ 		
r"   )NrÐ   Fr/  Fr?  r@  rE  r'  s   @r    r:  r:  Z  sH   ø„ ð :>Ø3Ø!&ØØØØñ
à˜#˜t˜)Ñ$ð
ð  (¨°©Ñ6÷
ñ 
r"   r:  c                   óL   ‡ — e Zd Z	 	 	 	 	 	 ddeeef   deee      fˆ fd„Zˆ xZ	S )r;  r“   r”   c           	      ó2   •— t         ‰| �  |||||||¬«       y)aŒ  
        :param model_path: ONNX model to calibrate. It is a model path
        :param op_types_to_calibrate: operator types to calibrate. By default, calibrate all the float32/float16 tensors.
        :param augmented_model_path: save augmented model to this path.
        :param use_external_data_format: use external data format to store model which size is >= 2Gb
        :param method: A string. One of ['entropy', 'percentile', 'distribution'].
        :param symmetric: make range of tensor symmetric (central point is 0).
        :param num_bins: number of bins to create a new histogram for collecting tensor values.
        :param scenario: for float 8 only, if `scenario="same"`,
            the algorithm weights and float 8 follow the same distribution,
            if `scenario="p3"`, it assumes the weights follow
            a gaussian law and float 8 ~ X^3 where X is a gaussian law
        )r,  r-  r0  NrD  )	rE   r“   r”   r˜   rš   r,  r-  r0  rÝ   s	           €r    rI   zDistributionCalibrater.__init__}  s,   ø€ ô. 	‰ÑØØ!Ø Ø$ØØØð 	õ 	
r"   )NrÐ   FÚdistributionr>  rA  rE  r'  s   @r    r;  r;  |  sE   ø„ ð :>Ø3Ø!&ØØØñ
à˜#˜t˜)Ñ$ð
ð  (¨°©Ñ6÷
ñ 
r"   r;  c                   óX   — e Zd ZdZej
                  d„ «       Zej
                  d„ «       Zy)ÚCalibrationDataCollectorzL
    Base class for collecting data for calibration-based quantization.
    c                 ó   — t         ‚)z‹
        Generate informative data based on given data.
            name_to_arr : dict
                tensor name to NDArray data
        r…   ©rE   Úname_to_arrs     r    r5  z CalibrationDataCollector.collect¤  s
   € ô "Ð!r"   c                 ó   — t         ‚)z?
        Get the optimal result among collection data.
        r…   rM   s    r    r<  z2CalibrationDataCollector.compute_collection_result­  s
   € ô
 "Ð!r"   N)rQ   rR   rS   Ú__doc__r�   rŽ   r5  r<  rV   r"   r    rL  rL  Ÿ  s;   „ ñð 	×Ññ"ó ð"ð 	×Ññ"ó ñ"r"   rL  c                   ód   — e Zd ZdZd„ Zd„ Zd„ Zd„ Zd„ Zd„ Z	d„ Z
d	„ Zd
„ Zedd„«       Zd„ Zd„ Zy)r4  a`  
    Collecting histogram for each tensor. Percentile and Entropy method are supported.

    ref: https://github.com//apache/incubator-mxnet/blob/master/python/mxnet/contrib/quantization.py
    ref: https://docs.nvidia.com/deeplearning/tensorrt/pytorch-quantization-toolkit/docs/_modules/
                 pytorch_quantization/calib/histogram.html
    c                 óf   — i | _         || _        || _        || _        || _        || _        || _        y rh   )Úhistogram_dictr,  r™   r-  r.  r/  r0  )rE   r,  r™   r-  r.  r/  r0  s          r    rI   zHistogramCollector.__init__¾  s5   € Ø ˆÔØˆŒØ"ˆŒØ ˆŒØ"4ˆÔØ$ˆŒØ ˆ�r"   c                 ó   — | j                   S rh   )rT  rM   s    r    Úget_histogram_dictz%HistogramCollector.get_histogram_dictÇ  s   € Ø×"Ñ"Ð"r"   c                 óæ   — t        d«       | j                  dv r| j                  |«      S | j                  dk(  r.| j                  r| j	                  |«      S | j                  |«      S t        d«      ‚)Nz/Collecting tensor data and making histogram ...>   r0   rJ  r/  úDOnly 'entropy', 'percentile' or 'distribution' methods are supported)Úprintr,  Úcollect_valuer™   Úcollect_absolute_valuer?   rN  s     r    r5  zHistogramCollector.collectÊ  sl   € ÜÐ?Ô@ð �;‰;Ð5Ñ5Ø×%Ñ% kÓ2Ð2Ø�[‰[˜LÒ(Ø�~Š~Ø×2Ñ2°;Ó?Ð?à×)Ñ)¨+Ó6Ð6äÐcÓdÐdr"   c                 ó´  — |j                  «       D �]D  \  }}t        |t        «      rz|D ]2  }t        |t        j                  «      rŒJ dt        |«      › d|›�«       ‚ t        d„ |D «       «      }t        |«      dk(  sJ d|› d|›�«       ‚t        j                  |«      }n6t        |t        j                  «      st        dt        |«      › d|›�«      ‚|}|j                  «       }|j                  dkD  r+t        j                  |«      }t        j                  |«      }nBt        j                  d|j                  ¬«      }t        j                  d|j                  ¬«      }t        j                   |«      }|| j"                  vrxt        j$                  || j&                  ¬«      \  }	}
|
j)                  |j                  «      }
|j                  t        j*                  k7  sJ d	«       ‚|	|
||f| j"                  |<   �Œî| j"                  |   }|d
   }|d   }t-        |d«      sJ dt        |«      › �«       ‚t-        |d«      sJ dt        |«      › �«       ‚|d   }|d   }t        j                  |«      }||d   kD  rB|d   |d   z
  }t        j.                  |d   |z   ||z   |«      }t        j0                  ||f«      }t        j$                  ||¬«      \  }	}
|
j)                  |j                  «      }
|	dt        |«      xxx |z  ccc |j                  t        j*                  k7  sJ d	«       ‚|	|
t        ||«      t        ||«      f| j"                  |<   �ŒG y)z5
        Collect histogram on absolute value
        r<   z for tensor=c              3   ó4   K  — | ]  }|j                   –— Œ y ­wrh   r   )Ú.0Úas     r    Ú	<genexpr>z<HistogramCollector.collect_absolute_value.<locals>.<genexpr>á  s   è ø€ Ð7©h¨˜QŸW�W©hùs   ‚r   z6The calibration expects only one element type but got r   r   )r9   zMonly float32 or float16 is supported, every constant must be explicetly typedr[   r]   r   z'old_min should be a numpy array but is éÿÿÿÿN)r=   r`   Úlistr   ÚndarrayrB   r³   rf   r)   r?   ÚflattenÚsizer
  r  rû   r   ÚabsoluterT  Ú	histogramr-  r*   Úfloat64rA   ÚarangeÚhstack)rE   rO  r   Údata_arrÚarrÚdtypesÚdata_arr_npr  r  r7   r8   Úold_histogramÚold_minÚold_maxÚold_histÚold_hist_edgesÚ	temp_amaxÚwidthÚnew_bin_edgess                      r    r[  z)HistogramCollector.collect_absolute_valueÙ  s@  € ð !,× 1Ñ 1× 3ÑˆF�HÜ˜(¤DÔ)Û#�CÜ% c¬2¯:©:Õ6ÐlÐ:JÌ4ÐPSË9È+ÐUaÐbhÐakÐ8lÓlÐ6ð $äÑ7©hÓ7Ó7�ä˜“K 1Ò$ðkàKÈFÈ8ÐS_Ð`fÐ_iÐjókØ$ä Ÿj™j¨Ó2‘Ü ¬"¯*©*Ô5Ü Ð#3´D¸³NÐ3CÀ<ÐPVÈzÐ!ZÓ[Ð[à&�Ø%×-Ñ-Ó/ˆKØ×Ñ !Ò#ÜŸF™F ;Ó/�	ÜŸF™F ;Ó/‘	äŸH™H Q¨k×.?Ñ.?Ô@�	ÜŸH™H Q¨k×.?Ñ.?Ô@�	äŸ+™+ kÓ2ˆKà˜T×0Ñ0Ñ0ä#%§<¡<°À$Ç-Á-Ô#PÑ ��jØ'×.Ñ.¨{×/@Ñ/@ÓA�
à×%Ñ%¬¯©Ò3ðcàbócØ3à/3°ZÀÈIÐ.V�×#Ñ# FÓ+à $× 3Ñ 3°FÑ ;�Ø'¨Ñ*�Ø'¨Ñ*�Ü˜w¨Ô0ÐkÐ4[Ô\`ÐahÓ\iÐ[jÐ2kÓkÐ0Ü˜w¨Ô0ÐkÐ4[Ô\`ÐahÓ\iÐ[jÐ2kÓkÐ0Ø(¨Ñ+�Ø!.¨qÑ!1�ÜŸF™F ;Ó/�	Ø˜~¨bÑ1Ò1à*¨1Ñ-°¸qÑ0AÑA�Eä$&§I¡I¨n¸RÑ.@À5Ñ.HÈ)ÐV[ÑJ[Ð]bÓ$c�MÜ%'§Y¡Y°ÀÐ/NÓ%O�NÜ#%§<¡<°À.Ô#QÑ ��jØ'×.Ñ.¨{×/@Ñ/@ÓA�
Ø�_”s˜8“}Ó%¨Ñ1Ó%à×%Ñ%¬¯©Ò3ðcàbócØ3à/3°ZÄÀWÈiÓAXÔZ]Ð^eÐgpÓZqÐ.r�×#Ñ# FÓ+ñi !4r"   c           	      óî  — |j                  «       D �]a  \  }}t        j                  |«      }|j                  «       }|j                  dkD  r+t        j
                  |«      }t        j                  |«      }nBt        j                  d|j                  ¬«      }t        j                  d|j                  ¬«      }t        j                  t        t        |«      t        |«      «      |j                  ¬«      }|| j                  v r3| j                  |   }| j                  |||||«      | j                  |<   �Œ&t        j                  || j                  | |f¬«      \  }}	||	|||f| j                  |<   �Œd y)z1
        Collect histogram on real value
        r   r   ©r  N)r=   r   r)   rd  re  r
  r  rû   r   r  rT  Úmerge_histogramrg  r-  )
rE   rO  r   rk  r  r  Ú	thresholdro  r7   r8   s
             r    rZ  z HistogramCollector.collect_value  s7  € ð !,× 1Ñ 1× 3ÑˆF�HÜ—z‘z (Ó+ˆHØ×'Ñ'Ó)ˆHà�}‰}˜qÒ ÜŸF™F 8Ó,�	ÜŸF™F 8Ó,‘	äŸH™H Q¨h¯n©nÔ=�	ÜŸH™H Q¨h¯n©nÔ=�	äŸ™¤¤S¨£^´S¸³^Ó!DÈHÏNÉNÔ[ˆIà˜×,Ñ,Ñ,Ø $× 3Ñ 3°FÑ ;�Ø.2×.BÑ.BØ! 8¨Y¸	À9ó/�×#Ñ# FÓ+ô $&§<¡<°¸$¿-¹-ÐQZÐPZÐ\eÐOfÔ#gÑ ��jàØØØØð/�×#Ñ# FÓ+ñ) !4r"   c                 ó  — |\  }}}}	}
||
k  rEt        j                  |t        |«      |
 |
f¬«      \  }}||z   |t        ||«      t	        |	|«      |
fS |
dk(  r-t        j                  |t        |«      | |f¬«      \  }}||z  }nft        |«      }d|
z  |z  }t        ||
z
  |z  dz   «      }|d|z  z   }||z  |
z   }t        j                  ||| |f¬«      \  }}||||z
  xxx |z  ccc ||t        ||«      t	        |	|«      |fS )Nrx  r   r[   r   )r   rg  rf   r
  r  Úint)rE   ro  rk  Únew_minÚnew_maxÚnew_thresholdrr  rs  rp  rq  Úold_thresholdÚnew_histrà   r7   r8   Úold_num_binsÚ
old_strideÚhalf_increased_binsÚnew_num_binss                      r    ry  z"HistogramCollector.merge_histogram3  sT  € ØFSÑCˆ�> 7¨G°]à˜MÒ)ÜŸ,™, x´°X³ÈÀ~ÐWdÐFeÔf‰KˆH�aà˜8Ñ#ØÜ�G˜WÓ%Ü�G˜WÓ%Øðð ð  Ò!Ü#%§<¡<°¼#¸h»-ÐQ^ÐP^Ð`mÐOnÔ#oÑ ��jØ˜Ñ ‘ä" 8›}�Ø Ñ.°Ñ=�
Ü&)¨=¸=Ñ+HÈZÑ*WÐZ[Ñ*[Ó&\Ð#Ø+¨aÐ2EÑ.EÑE�Ø 3°jÑ @À=Ñ P�Ü#%§<¡<°¸,ÐP]È~Ð_lÐNmÔ#nÑ ��jØÐ(¨<Ð:MÑ+MÓNÐRZÑZÓNàØÜ�G˜WÓ%Ü�G˜WÓ%Øðð r"   c                 ób  — | j                   rt        | j                   «      dk(  rt        d«      ‚t        d| j                  ›d�«       | j                  dk(  r| j                  «       S | j                  dk(  r| j                  «       S | j                  dk(  r| j                  «       S t        d«      ‚)	Nr   z=Histogram has not been collected. Please run collect() first.z0Finding optimal threshold for each tensor using z algorithm ...r0   r/  rJ  rX  )rT  rf   r?   rY  r,  Úcompute_entropyÚcompute_percentileÚcompute_distributionrM   s    r    r<  z,HistogramCollector.compute_collection_resultS  sœ   € Ø×"Ò"¤c¨$×*=Ñ*=Ó&>À!Ò&CÜÐ\Ó]Ð]ÜÐ@ÀÇÁÀÈ~Ð^Ô_à�;‰;˜)Ò#Ø×'Ñ'Ó)Ð)Ø�[‰[˜LÒ(Ø×*Ñ*Ó,Ð,Ø�[‰[˜NÒ*Ø×,Ñ,Ó.Ð.äÐcÓdÐdr"   c                 óˆ  — | j                   dk  s| j                   dkD  rt        d«      ‚| j                  }| j                   }i }t        dt	        |«      › �«       t        d| j
                  › �«       t        dd|z
  › d|› d	�«       |j                  «       D �]¦  \  }}|d   }|d
   }|j                  «       }t        j                  ||z  «      }	| j                  rft        j                  |	|dz  «      }
t        j                  ||
   |j                  ¬«       t        j                  ||
   |j                  ¬«      f||<   n‚d|z
  dz  }t        j                  |	d|z
  «      }
t        j                  |	|«      }t        j                  ||   |j                  ¬«      t        j                  ||
   |j                  ¬«      f||<   |d   }|d   }||   d   |k  r|||   d
   f||<   ||   d
   |kD  r||   d   |f||<   g ||   ¢|d d ¢­||<   t        j                  j!                  dd«      dv s�Œ›t#        ||«       �Œ© |S )Nr   éd   z<Invalid percentile. Must be in range 0 <= percentile <= 100.úNumber of tensors : úNumber of histogram bins : zPercentile : (g      Y@Ú,Ú)r   r   g      i@r&   r[   r]   ÚQUANTIZATION_DEBUG©r   Ú1)r/  r?   rT  rY  rf   r-  r=   r,   r   Úcumsumr™   Úsearchsortedrû   r   ÚosÚenvironÚgetr   )rE   rT  r/  Úthresholds_dictr   rg  r7   r8   ÚtotalÚcdfÚ	idx_rightÚpercent_to_cut_one_sideÚidx_leftr  r  s                  r    rˆ  z%HistogramCollector.compute_percentilea  sU  € Ø�?‰?˜QÒ $§/¡/°CÒ"7ÜÐ[Ó\Ð\à×,Ñ,ˆØ—_‘_ˆ
àˆäÐ$¤S¨Ó%8Ð$9Ð:Ô;ÜÐ+¨D¯M©M¨?Ð;Ô<Ü�˜u zÑ1Ð2°!°J°<¸qÐAÔBà!/×!5Ñ!5×!7ÑˆF�IØ˜Q‘<ˆDØ" 1™ˆJØ—H‘H“JˆEÜ—)‘)˜D 5™LÓ)ˆCØ�~Š~ÜŸO™O¨C°¸eÑ1CÓD�	ô —X‘X˜j¨Ñ3¸:×;KÑ;KÔLÐLÜ—H‘H˜Z¨	Ñ2¸*×:JÑ:JÔKð+� Ò'ð
 ,1°:Ñ+=ÀÑ*FÐ'ÜŸO™O¨C°Ð7NÑ1NÓO�	ÜŸ?™?¨3Ð0GÓH�ä—H‘H˜Z¨Ñ1¸×9IÑ9IÔJÜ—H‘H˜Z¨	Ñ2¸*×:JÑ:JÔKð+� Ñ'ð " !™ˆIØ! !™ˆIØ˜vÑ& qÑ)¨IÒ5Ø+4°oÀfÑ6MÈaÑ6PÐ*Q� Ñ'Ø˜vÑ& qÑ)¨IÒ5Ø+:¸6Ñ+BÀ1Ñ+EÀyÐ*Q� Ñ'Ø&K¨¸Ñ(?Ð&KÀ$ÀrÈÀ(Ñ&KˆO˜FÑ#ä�z‰z�~‰~Ð2°AÓ6¸(ÓBÜ˜4 Ö,ð; "8ð> Ðr"   c                 ó¬  — | j                   }| j                  }i }t        dt        |«      › �«       t        d| j                  › d�«       t        d| j                  › �«       |j                  «       D ]^  \  }}| j                  ||«      }|||<   g |¢|d d ¢­||<   t        j                  j                  dd«      dv sŒMt        |d   |d	   «       Œ` |S )
NrŒ  r�  z: (The number may increase depends on the data it collects)zNumber of quantized bins : r[   r�  r   r‘  r   )rT  r.  rY  rf   r-  r=   Úget_entropy_thresholdr•  r–  r—  r   )rE   rT  r.  r˜  r   rg  Úoptimal_thresholds          r    r‡  z"HistogramCollector.compute_entropy�  sç   € Ø×,Ñ,ˆØ!×4Ñ4ÐàˆäÐ$¤S¨Ó%8Ð$9Ð:Ô;ÜÐ+¨D¯M©M¨?Ð:tÐuÔvÜÐ+¨D×,CÑ,CÐ+DÐEÔFà!/×!5Ñ!5Ö!7ÑˆF�IØ $× :Ñ :¸9ÐFXÓ YÐØ&7ˆO˜FÑ#Ø&JÐ(9Ð&J¸IÀbÀq¸MÑ&JˆO˜FÑ#ô �z‰z�~‰~Ð2°AÓ6¸(ÒBÜ˜9 Q™<¨°1©Õ6ð "8ð Ðr"   c                 óÞ  — |dk  rt        d|› d�«      ‚|d d |dd  z   dz  }|dk(  r–| |z  j                  «       | j                  «       z  }| |dz  z  j                  «       | j                  «       z  |dz  z
  dz  }t        j                  ||j                  ¬«      t        j                  ||j                  ¬«      fS t        |«      |k(  rªt        |«      dz  dk(  r™| ||z  z  j                  «       | j                  «       z  }| ||z  |z
  dz  z  j                  «       | j                  «       z  dz  }t        j                  ||j                  ¬«      t        j                  ||j                  ¬«      fS t        j                  |«      |z  }d|t        j                  |«      <   d|t        j                  |«      <   t        j                  |«      |z  |z  }| |z  j                  «       | j                  «       z  }| |dz  z  j                  «       | j                  «       z  |dz  z
  dz  }t        j                  ||j                  ¬«      t        j                  ||j                  ¬«      fS )	Nr   zpower=z <= 0 is invalid.ra  r   g      à?r[   r   )	r?   r,   r   rû   r   r|  r  ÚisnanÚisinf)r7   r8   Úpowerrv   r3   r4   Úfacts          r    Ú_avg_stdzHistogramCollector._avg_std¤  s  € à�AŠ:Ü˜v e WÐ,=Ð>Ó?Ð?Ø˜S˜b�/ J¨q¨r NÑ2°cÑ9ˆØ�AŠ:Ø˜&‘=×%Ñ%Ó'¨$¯(©(«*Ñ4ˆCØ˜6 1™9Ñ$×)Ñ)Ó+¨d¯h©h«jÑ8¸3À¹6ÑAÀcÑIˆCÜ—8‘8˜C z×'7Ñ'7Ô8¼"¿(¹(À3Èj×N^ÑN^Ô:_Ð_Ð_Üˆu‹:˜Ò¤3 u£:°¡>°QÒ#6Ø˜& %™-Ñ'×,Ñ,Ó.°·±³Ñ;ˆCØ˜F E™M¨CÑ/°AÑ5Ñ5×:Ñ:Ó<¸t¿x¹x»zÑIÈcÑQˆCÜ—8‘8˜C z×'7Ñ'7Ô8¼"¿(¹(À3Èj×N^ÑN^Ô:_Ð_Ð_ä�v‰v�f‹~ Ñ&ˆØ ˆŒR�X‰X�d‹^ÑØ ˆŒR�X‰X�d‹^ÑÜ—‘˜“ 5Ñ(¨4Ñ/ˆØ�f‰}×!Ñ!Ó# d§h¡h£jÑ0ˆØ�v˜q‘yÑ ×%Ñ%Ó'¨$¯(©(«*Ñ4°s¸A±vÑ=À#ÑEˆÜ�x‰x˜ :×#3Ñ#3Ô4´b·h±h¸sÈ*×JZÑJZÔ6[Ð[Ð[r"   c           
      ó˜  — | j                   dk  rt        d«      ‚| j                  }i }t        dt	        |«      › �«       t        d| j                   › �«       t        d| j
                  ›d�«       |j                  «       D �]E  \  }}|d   }|d   }|j                  t        j                  k7  sJ ‚| j
                  d	k(  r| j                  ||d¬
«      \  }}n2| j
                  dk(  r| j                  ||d¬
«      \  }}nt        d«      ‚|j                  t        j                  k7  sJ ‚|j                  t        j                  k7  sJ ‚|j                  t        j                  k7  sJ ‚t        |||||j                  «       |j                  «       ¬«      ||<   t        j                  j!                  dd«      dv s�Œ:t#        ||«       �ŒH |S )Ni   z3Invalid num_bins. Must be in range 512 <= num_bins.rŒ  r�  zScenario : r�  r   r   rA  )r¤  Úp3gUUUUUUÕ?z,Invalid scenario. Must be in {'same', 'p3'}.)r3   r4   r7   r8   r5   r6   r�  r‘  )r-  r?   rT  rY  rf   r0  r=   r   r   rh  r¦  r2   r
  r  r•  r–  r—  r   )	rE   rT  r˜  r   rg  r7   r8   Úavg_coefÚstd_coefs	            r    r‰  z'HistogramCollector.compute_distributionº  s¡  € Ø�=‰=˜3ÒÜÐRÓSÐSà×,Ñ,ˆØˆäÐ$¤S¨Ó%8Ð$9Ð:Ô;ÜÐ+¨D¯M©M¨?Ð;Ô<Ü�˜DŸM™MÐ,¨AÐ.Ô/à!/×!5Ñ!5×!7ÑˆF�IØ˜Q‘<ˆDØ" 1™ˆJà×#Ñ#¤r§z¡zÒ1Ð1Ð1Ø�}‰} Ò&Ø%)§]¡]°4¸È1 ]Ó%MÑ"�™(Ø—‘ $Ò&Ø%)§]¡]°4¸È9 ]Ó%UÑ"�™(ä Ð!OÓPÐPØ—>‘>¤R§Z¡ZÒ/Ð/Ð/Ø—>‘>¤R§Z¡ZÒ/Ð/Ð/Ø×#Ñ#¤r§z¡zÒ1Ð1Ð1Ü&0ØØØØ%Ø!—~‘~Ó'Ø"Ÿ™Ó(ô'ˆO˜FÑ#ô �z‰z�~‰~Ð2°AÓ6¸(ÓBÜ˜4 Ö,ð3 "8ð6 Ðr"   c           	      ó"  — |d   }|d   }|j                   }|dz  }|dz  }|d   j                  }t        j                  ||z
  dz   «      }	t	        |	j                   «      D �
cg c]0  }
t        j
                  d|¬«      t        j
                  d|¬«      f‘Œ2 }}
t	        ||dz   d«      D �]ù  }
||
z
  }||
z   dz   |k  r||
z   dz   n|}||   ||   f||
|z
  <   t        j                  ||| «      }|j                  «       }t        |d| «      }t        ||d «      }|dxx   |z  cc<   |dxx   |z  cc<   |dk7  j                  t        j                  «      }t        j                  |t        j                  ¬«      }|j                   |z  }t	        |«      D ]  }||z  }||z   }t        ||| «      ||<   Œ |dxx   t        |||z  d «      z  cc<   t        j                  |j                   t        j                  ¬«      }t	        |«      D ]+  }||z  }||z   }t        ||| «      }|dk7  sŒ!||   |z  ||| Œ- t        |«      }t        |«      }|�|€&t        j
                  t        j                  |¬«      }n!t        j
                  t        ||«      |¬«      }||	|
|z
  <   �Œü t        j                  |	«      }||   }|d   }|d   }|d   |k  r||d   f}|d   |kD  r|d   |f}t        |d   d«      sJ ‚t        |d   d«      sJ ‚|S c c}
w )	aF  Given a dataset, find the optimal threshold for quantizing it.
        The reference distribution is `q`, and the candidate distribution is `p`.
        `q` is a truncated version of the original distribution.
        Ref: http://on-demand.gputechconf.com/gtc/2017/presentation/s7310-8-bit-inference-with-tensorrt.pdf
        r   r   r[   r   Nra  r]   r   )re  r   r   Úzerosr  rû   ÚcopyÚdeepcopyr,   r*   rü   r   r   r0   ÚargminrA   )rE   rg  r.  r7   r8   r-  Úzero_bin_indexÚnum_half_quantized_binr   Úkl_divergencerð   Ú
thresholdsÚstart_indexÚ	end_indexÚsliced_distributionÚpÚleft_outliers_countÚright_outliers_countÚnonzerosÚquantized_binsÚnum_merged_binsÚindexÚstartÚendÚqÚnormÚdivÚmin_kl_divergence_idxr   r  r  s                                  r    rŸ  z(HistogramCollector.get_entropy_thresholdâ  su  € ð ˜‰|ˆØ˜q‘\ˆ
Ø—9‘9ˆØ! Q™ˆØ!3°qÑ!8Ðà˜!‘×"Ñ"ˆÜŸ™ Ð2HÑ!HÈ1Ñ!LÓMˆÜTYÐZg×ZlÑZlÔTmÓnÑTmÈq”r—x‘x ¨Ô/´·±¸!À5Ô1IÒJÐTmˆ
Ðnô  Ð-¨~ÀÑ/AÀ1×EˆAØ(¨1Ñ,ˆKØ3AÀAÑ3EÈÑ3IÈhÒ2V˜¨Ñ*¨QÒ.Ð\dˆIà6@ÀÑ6MÈzÐZcÑOdÐ5eˆJ�qÐ1Ñ1Ñ2ä"&§-¡-°°[ÀÐ0KÓ"LÐð $×(Ñ(Ó*ˆAÜ"% d¨<¨KÐ&8Ó"9ÐÜ#& t¨I¨JÐ'7Ó#8Ð Øˆa‹DÐ'Ñ'‹DØˆb‹EÐ)Ñ)‹Eð ˜Q™—‘¤r§x¡xÓ0ˆHô  ŸX™XÐ&8ÄÇÁÔIˆNØ1×6Ñ6Ð:LÑLˆOô Ð1Ö2�Ø Ñ/�Ø˜oÑ-�Ü(+Ð,?ÀÀcÐ,JÓ(K�˜uÒ%ð 3ð ˜2Ó¤#Ð&9Ð:LÈÑ:^Ð:`Ð&aÓ"bÑbÓô —‘˜Ÿ™¤r§x¡xÔ0ˆAÜÐ1Ö2�Ø Ñ/�Ø˜oÑ-�ä˜8 E¨#Ð.Ó/�Ø˜1“9Ø#1°%Ñ#8¸4Ñ#?�A�e˜C‘Lð 3ô $ AÓ&ˆAÜ# AÓ&ˆAØˆy˜A˜IÜ—h‘hœrŸv™v¨UÔ3‘ä—h‘hœw q¨!›}°EÔ:�Ø8;ˆM˜!Ð4Ñ4Ó5ð] Fô` !#§	¡	¨-Ó 8ÐØ&Ð'<Ñ=ÐØ˜a‘Lˆ	Ø˜a‘Lˆ	Ø˜QÑ )Ò+Ø!*Ð,=¸aÑ,@Ð AÐØ˜QÑ )Ò+Ø!2°1Ñ!5°yÐ AÐÜÐ(¨Ñ+¨WÔ5Ð5Ð5ÜÐ(¨Ñ+¨WÔ5Ð5Ð5Ø Ð ùòU os   Á"5LN)r   )rQ   rR   rS   rQ  rI   rV  r5  r[  rZ  ry  r<  rˆ  r‡  Ústaticmethodr¦  r‰  rŸ  rV   r"   r    r4  r4  µ  s]   „ ñò!ò#òeò8sòtò@ò@eò,ò\ð* ò\ó ð\ò*&óPX!r"   r4  rÐ   Fr—   r”   c           
      óJ  — d }|t         j                  k(  r]|j                  dd«      }|j                  dd«      }|j                  dd«      }	|j                  dd «      }
t        | ||||||	|
¬«      }�n |t         j                  k(  rI|j                  dd	«      }|j                  d
d	«      }|j                  dd«      }t        | ||||||¬«      }n¤|t         j                  k(  rI|j                  dd«      }|j                  dd«      }|j                  dd«      }t        | ||||||¬«      }nH|t         j                  k(  r5|j                  dd«      }|j                  dd«      }t        | |||||¬«      }|r"|j                  «        |j                  «        |S t        d|› �«      ‚)Nr™   FrÚ   rÛ   r%  rÜ   )rš   r™   rÚ   rÛ   rÜ   r-  r>  r.  )rš   r™   r-  r.  r?  r/  r@  T)rš   r™   r-  r/  r0  rA  )rš   r-  r0  zUnsupported calibration method )rd   re   r—  rÒ   rx   r9  ry   r:  rz   r;  rÉ   rŸ   r?   )r—   r”   r˜   Úcalibrate_methodrš   Úextra_optionsÚ
calibratorr™   rÚ   rÛ   rÜ   r-  r.  r/  r0  s                  r    Úcreate_calibratorrÉ  =  sß  € ð €JØÔ,×3Ñ3Ò3à!×%Ñ% k°5Ó9ˆ	Ø&×*Ñ*Ð+;¸UÓCˆØ*×.Ñ.Ð/CÀTÓJÐØ#0×#4Ñ#4Ð5OÐQUÓ#VÐ Ü%ØØ!Ø Ø%=ØØ)Ø1Ø%=ô	
Š
ð 
Ô.×6Ñ6Ò	6à ×$Ñ$ Z°Ó5ˆØ*×.Ñ.Ð/CÀSÓIÐØ!×%Ñ% k°5Ó9ˆ	Ü&ØØ!Ø Ø%=ØØØ1ô
‰
ð 
Ô.×9Ñ9Ò	9à ×$Ñ$ Z°Ó6ˆØ"×&Ñ& |°VÓ<ˆ
Ø!×%Ñ% k°4Ó8ˆ	Ü)ØØ!Ø Ø%=ØØØ!ô
‰
ð 
Ô.×;Ñ;Ò	;à ×$Ñ$ Z°Ó6ˆØ ×$Ñ$ Z°Ó8ˆä+ØØ!Ø Ø%=ØØô
ˆ
ñ Ø× Ñ Ô"Ø×+Ñ+Ô-ØÐä
Ð6Ð7GÐ6HÐIÓ
JÐJr"   )Nr   )0r�   r­  r¸   r•  rø   Úenumr   Úpathlibr   Útypingr   r   r   r   r	   Únumpyr   rå   r
   r   r   r   r¤   Úquant_utilsr   r   r   rc  r!   Úfloatr|  r0   r2   rX   rd   ÚABCMetar|   r’   rÒ   r)  r9  r:  r;  rL  r4  re   ra   rÉ  rV   r"   r    Ú<module>rÑ     s”  ðó Û Û Û 	Û Ý Ý ß 9Õ 9ã Û ß >Ó >ã ç UÑ Uð�—‘ð  §¡ð °·
±
ó ð" !Øñ	Ø
�
‰
ðà
�
‰
ðð �5‰/ðð ð	ð
 ‡Z�Zó÷8$ñ $÷8""ñ ""ôJ˜ô ô c§k¡kõ ÷(h"ñ h"ôVr,�~ô r,ôjvL˜.ô vLôr
Ð+ô 
ôD
Ð.ô 
ôD 
Ð0ô  
ôF"¨¯©õ "ô,E!Ð1ô E!ðT 6:Ø/Ø&×-Ñ-Ø"ØñIKØ��d�ÑðIKà# H¨S¡MÑ2ôIKr"   