`i2SSKrSSKrSSKrSSKJr SSKJr SSKJr Sr Sr Sr Sr S S S S S SSSSSSSSSSS.r "SS\5r"SS\5r"SS\5r"SS \5r\R$R&\R("S!5\R$R*\R("S"50rS7S#jrS$rS%rS8S&jr"S'S(5r"S)S*5r"S+S,5rS-rS.rS/r S0r!S1\S2\"S3\4S4jr#S5r$S9S6jr%g):N)Enum)onnx_pb)Pathz onnx.quantizez0.1.0zai.onnxz com.microsoftFLOATUINT8INT8UINT16INT16INT32INT64STRINGBOOLFLOAT16DOUBLEUINT32UINT64 COMPLEX64 COMPLEX128) c2\rSrSrSrSrSr\S5rSr g)QuantizationMode%rrcUR$Nnameselfs i/var/www/html/ai-image-ml/venv/lib/python3.13/site-packages/onnxruntime_tools/quantization/quant_utils.py__str__QuantizationMode.__str__) yycF[U$![a [5ef=fr()r%KeyError ValueError)modes r- from_stringQuantizationMode.from_string,s) #D) ) ,    N) __name__ __module__ __qualname____firstlineno__ IntegerOps QLinearOpsr. staticmethodr6__static_attributes__r9r1r-r%r%%s%JJr1r%c2\rSrSrSrSrSr\S5rSr g)QuantizedValueType4rrcUR$r(r)r+s r-r.QuantizedValueType.__str__8r0r1cF[U$![a [5ef=fr()rCr3r4)vs r-r6QuantizedValueType.from_string;s) %a( ( ,  r8r9N) r:r;r<r=Input Initializerr.r@r6rAr9r1r-rCrC4s% EKr1rCc2\rSrSrSrSrSr\S5rSr g) QuantTypeCrrcUR$r(r)r+s r-r.QuantType.__str__Gr0r1cF[U$![a [5ef=fr()rMr3r4)ts r-r6QuantType.from_stringJs( Q<  ,  r8r9N) r:r;r<r=QInt8QUInt8r.r@r6rAr9r1r-rMrMCs% E Fr1rMc2\rSrSrSrSrSr\S5rSr g) QuantFormatRrrcUR$r(r)r+s r-r.QuantFormat.__str__Vr0r1cF[U$![a [5ef=fr()rWr3r4)formats r-r6QuantFormat.from_stringYs) v& & ,  r8r9N) r:r;r<r= QOperatorQDQr.r@r6rAr9r1r-rWrWRs%I Cr1rWint8uint8cU[;dSRU55e[Un[U[R:XaSOSUcSOU5n[ U[R:XaSOSUcSOU5n[R "UR[R5U- R5U-5n [R"XXS9 U RU5$)NEUnexpected data type {} requested. Only INT8 and UINT8 are supported.ri)out) ONNX_TYPE_TO_NP_TYPEr\maxnumpyraminasarrayastypefloat32roundclip) qTypearrscale zero_pointlowhighdtypecliplowcliphigharr_fp32s r-quantize_nparrayrzes ( (^OVVW\]^ (  'Eu +!s{tPSTG%5;;.3C RVWH}}cjj7%?FFH:UVH JJx(9 ??5 !!r1c~U[RR:Xa<[[ U5[ U55nUS:a[ U5S-U- OSnSnXe/$U[RR :Xa,X:wa[ U5U- U- OSn[SU- U- 5nXe/$[SRU55e)Nrrrrc) onnx_proto TensorProtorrhabsfloatrrnr4r\)rminrmaxrpquantize_range max_rangerrrss r-compute_scale_zprps &&+++D 3t9- ;Dq=y!A%7a    *((.. .9=tt#~51AH-.   `gghmnoor1c[[U5S5n[[U5S5n[X4X!5upV[U[R "U5Xe5nX4XVU4$)a :parameter data: data to quantize :parameter quantize_range: list of data to weight pack. :parameter qType: data type to quantize to. Supported types UINT8 and INT8 :return: minimum, maximum, zero point, scale, and quantized weights To pack weights, we compute a linear transformation - when data type == uint8 mode, from [rmin, rmax] -> [0, 2^{b-1}] and - when data type == int8, from [-m , m] -> [-(2^{b-1}-1), 2^{b-1}-1] where m = max(abs(rmin), abs(rmax)) and add necessary intermediate nodes to trasnform quantized weight to full weight using the equation r = S(q-z), where r: real original value q: quantized value S: scale z: zero point r)rjrhrrzrirk)datarrprrrsrrquantized_datas r- quantize_datar~sW" s4y! D s4y! D(UKJ%eU]]4-@%TN z. 88r1cU[RR:Xa U(aS$S$U[RR:Xa U(aS$S$[ S5e)z Helper function to get the quantization range for a type. parameter qType: quantization type. return: quantization range. rerdz"unsupported quantization data type)r|r}rrr4)rp reduce_ranges r-get_qrange_for_qTypersR  &&,,,"s++ *((-- -"s++=>>r1c(\rSrSrSr//S4SjrSrg)QuantizedInitializerzB Represents a linearly quantized weight input from ONNX operators Nc pXlX lX0lX@lXPlX`lXplXlXlgr() r* initializerrminsrmaxs zero_pointsscalesrraxis) r,r*rrrrrrrrs r-__init__QuantizedInitializer.__init__s4 &  &  , r1) rrrr*rrrrrr:r;r<r=__doc__rrAr9r1r-rrs "r1rc$\rSrSrSrSSjrSrg)QuantizedValuezA Represents a linearly quantized value (input\output\intializer) NcLXlX lX0lX@lXPlX`lgr() original_nameq_name scale_namezp_name value_typer)r,r*new_quantized_namerzero_point_namequantized_value_typers r-rQuantizedValue.__init__s#"( $& . r1)rrrrrrr(rr9r1r-rrs r1rc\rSrSrSrSrSrg)BiasToQuantizez# Represents a bias to be quantized c(XlX lX0lgr( bias_name input_name weight_name)r,rrrs r-rBiasToQuantize.__init__s"$&r1rNrr9r1r-rrs 'r1rc,URS:Xa$[SRUR55eURS:XaURnGO5URS:XaUR nGOURS:Xa UR nOURS:Xa URnOURS:Xa URnOURS:Xa URnOURS :Xa URnOURS :Xa URnOiURS :Xa URnOLURS :Xa URnO/[S RURUR55eURU0$)z Convert attribute to kwarg format for use with onnx.helper.make_node. :parameter attribute: attribute in AttributeProto format. :return: attribute in {key: value} format. rz*attribute {} does not have type specified.rrrrrrrrrrz%attribute {} has unsupported type {}.)typer4r\r*fisrRgfloatsintsstringstensorsgraphs) attributevalues r-attribute_to_kwargrs: !ELLY^^\]] !  ..A   ..A   ..A   ..A   ..A    ..A  ..A !! ..A !! ..B   @GG XaXfXfghh NNE ""r1c~UVs/sHo"RU:XdMUPM nn[U5S:aUS$S$s snf)z Helper function to find item by name in a list. parameter item_name: name of the item. parameter item_list: list of items. return: item if found. None otherwise. rN)r*len) item_name item_listitemitemss r- find_by_namers@( Bid99 +ATiE B5zA~58/4/ Cs::cZSn[S[U55HnXU:XdM UnM U$)z; Helper function to return index of an item in a node list r)ranger) elem_name elem_listelem_idxrs r-get_elem_indexrs4H 1c)n % <9 $H& Or1cF[RRSX/U5$)z Helper function to create a Mul node. parameter inputs: list of input names. parameter output: output name. parameter name: name of the node. return: Mul node in NodeProto format. Mul)onnxhelper make_node)inputsoutputr*s r- get_mul_noders ;; $ ??r1filename identifierreturncURRURU-5RUR5$)zk Helper function to generate a identifiable filepath by concatenating the given identifier as a suffix. )parentjoinpathstem with_suffixsuffix)rrs r-generate_identified_filenamer%s1 ?? # #HMMJ$> ? K KHOO \\r1c SSKnSSKnSSKJs Js Jn SSKJs Js Jn [R"SRU55 [SS5nURURU55 SSS5 URS5n/n[UR!55HnXn [#[%['U S5['U S555n UR)U5n UR)U 5n UR+U5 UR-Xk5 UR/Xl5 UR1U5n UR3U 5 M UR5U[7U55 UHn UR9U 5 M UR;[7U55nUR=U5 UR?Xn5 URAU5nURCU5 URE5n[SS 5nURU5 SSS5 [S S5n[UR!55H\nXn US -[#[%['U S5['U S555-nURU5 URS 5 M^ SSS5 g!,(df  GNC=f!,(df  N=f!,(df  g=f) z9 Helper function to write calibration table to files. rNzcalibration cache: {}zcalibration.jsonwirzcalibration.flatbufferswbzcalibration.cache  ))json flatbuffers5onnxruntime.quantization.CalTableFlatBuffers.TrtTable quantizationCalTableFlatBuffersTrtTable5onnxruntime.quantization.CalTableFlatBuffers.KeyValueKeyValuelogginginfor\openwritedumpsBuildersortedkeysstrrhr~ CreateString KeyValueStartKeyValueAddKeyKeyValueAddValue KeyValueEndappendTrtTableStartDictVectorrPrependUOffsetTRelative EndVector TrtTableStartTrtTableAddDict TrtTableEndFinishOutputGetRootAsTrtTable DictLengthrDictKeyValue)calibration_cacherrrrfilebuilderkey_value_listkeyvaluesrflat_key flat_value key_value main_dict cal_tablebufdict_lenrrs r-write_calibration_tabler+sX LLLL LL(//0ABC  # &$ 4::/01 '!!$'GN',,./"'CF1IF1I78'',))%0 w'2!!'6((1 i(0 $$Wc..AB# '' 2$!!#n"56I 7# W0$$W-I NN9 .. C ' .$ 3 /  !3 '4+0023C%*Ec CCaM3uQx= ABBA JJqM JJt  4 ( '[ ' &@ / . ( 's%!J/K,A:K/ J> K K cSSKnUS:HRUR5nUS:gRUR5nUR5nURU- nU(dgU[ U5-[ U5- nUS:d SXVU4-5eURUR5nXU-U*U--- nUS:*R5S:XdeU$)ajGiven a discrete distribution (may have not been normalized to 1), smooth it by replacing zeros with eps multiplied by a scaling factor and taking the corresponding amount off the non-zero values. Ref: http://web.engr.illinois.edu/~hanj/cs412/bk3/KL-divergence.pdf https://github.com//apache/incubator-mxnet/blob/master/python/mxnet/contrib/quantization.py rNrg?z"n_zeros=%d, n_nonzeros=%d, eps1=%f)rirlrmsumsizer) pepsnpis_zeros is_nonzerosn_zeros n_nonzeroseps1hists r-smooth_distributionr#ksQrzz*H6//"**-KllnG'!J  w % "3 3D #:Y;wTX>YYY: 88BJJ D(Nte{2 22D AI??  !! ! Kr1)NN)F)g-C6?)&rrirenumrrr|pathlibr __producer__ __version__ onnx_domain ms_domain type_to_namer%rCrMrWr}rrvrrgrzrrrrrrrrrrrrrr#r9r1r-r+s@ &       . t      $ V!4  5;;w#7 " 94 ?8&''"#J0@]4]S]T] >@r1