o gQ@srddlmZddlZddlmZddlmZddlmZddl Z ddl m Z eGdd d Z Gd d d eZ dS) ) annotationsN)MutableMapping) dataclass)Any) QuantTypec@sheZdZUdZded<dZded<dZded<dZded <dd d Ze   ddddZ dddZ dS) QuantTypeInfozB The quantization type information for a tensor override. r quant_typeN bool | None symmetric reduce_rangez int | NoneaxisotherobjectcCsft|tr1|j|jko0|jdup|jdup|j|jko0|jdup*|jdup*|j|jko0|j|jkStSN) isinstancerr r r r NotImplemented)selfrrf/var/www/visachat/venv/lib/python3.10/site-packages/onnxruntime/quantization/tensor_quant_overrides.py__eq__s   zQuantTypeInfo.__eq__raw_dictdict[str, Any] default_qtypeQuantType | Nonedefault_symmetricdefault_reduce_rangereturncCs,t|d||d||d||dSNr r r r )rget)rrrrrrrload_from_dict's    zQuantTypeInfo.load_from_dictcCsN|j|d<|jdur|j|d<|jdur|j|d<|jdur%|j|d<dSdSr)r r r r )rrrrr save_to_dict5s      zQuantTypeInfo.save_to_dict)rr)NNN) rrrrrr rr rr)rr) __name__ __module__ __qualname____doc____annotations__r r r r staticmethodr r!rrrrrs       rcseZdZdZdUddZdVd d ZdVd d ZdVddZ dWdXddZ dWdYddZ dZddZ d[ddZ d\d"d#Z d]d(d)Z  *d^d_d/d0Z dWd`d7d8Z  dadbdd?Zddd@dAZdedBdCZdfdGdHZdgdIdJZdhdKdLZdMdNZdOdPZdcdQdRZdcfdSdT ZZS)iTensorQuantOverridesHelperzZ Utility wrapper over the tensor quantization overrides passed via extra_options. raw_overridesdict[str, list[dict[str, Any]]]cCs||_d|_hd|_dS)N>rmaxrminr r ) overrides quant_typeskeys_unsupported_with_scale_zp)rr)rrr__init__Dsz#TensorQuantOverridesHelper.__init__ tensor_namestrrboolcCs|j|}|o d|dvSNr rr-rrr1overrides_listrrrhas_per_tensor_overridesI z3TensorQuantOverridesHelper.has_per_tensor_overridescCs|j|}|o d|dvSr4r5r6rrrhas_per_channel_overridesMr9z4TensorQuantOverridesHelper.has_per_channel_overridescCs(|j|}|od|dvod|dvS)Nscaler zero_pointr5r6rrroverrides_scale_zpQs z-TensorQuantOverridesHelper.overrides_scale_zpN default_valdict[str, Any] | NonecCsP|dur|gnd}|j||}|r d|dvr td|d|r&|dSdS)Nr rExpected tensor 'zL' to use per-tensor quantization overrides, but found per-channel overrides.r-r ValueError)rr1r>default_list_valr7rrrget_per_tensor_overridesUs z3TensorQuantOverridesHelper.get_per_tensor_overrideslist[dict[str, Any]] | NonecCs6|j||}|s dSd|dvrtd|d|S)Nr rr@zE' to have per-channel quantization overrides (axis value is missing).rA)rr1r>r7rrrget_per_channel_overridesds  z4TensorQuantOverridesHelper.get_per_channel_overridesset[QuantType]cCs~|jdur|jSt|_|jr<|jD]'}|D]"}d|vr&|j|dd|vr:d|dvr:|j|ddqq|jS)Nr convert)r.setr-valuesadd)rquant_overrides_listquant_overridesrrrget_quant_typesus z*TensorQuantOverridesHelper.get_quant_typesrMrtuple[bool, str | None]cCst|ts dd|dfS||v}|d}|r|j|d|v}d|v}|r*|r.|r0|s0dS|rG|jt|} | rGddd | d fSd |vrT|sTdd |fSd |vr|r\dSd|d vrjdd|dfSd |d vrwdd |fS|d d} |dur|n|} | | krdd|dfSd|d v} d|d v} | r| r| r| sdd|dfS| r|jt|d } | rddd | d|dfS|j| dS)NF#Tensor quantization overrides for '' are not in a dictr r;r<FzNMust provide both 'scale' and 'zero_point' if one of the overrides is providedTensor override option(s) [, +] are invalid with 'scale' and 'zero_point'r zMOption 'reduce_range' is only supported for initializers, not for activation rH)Fz.Cannot use 'convert' override for initializersz'convert' options (tensor 'z') must specify a 'quant_type'zC'convert' quant_type must differ from original quant_type (tensor 'z')zXMust provide both 'scale' and 'zero_point' if one of the overrides is provided (tensor 'z5] are invalid with 'scale' and 'zero_point' (tensor 'TN) rdictrr.rKr/ intersectionrIjoin)r initializersdefault_activation_qtyper1rMis_initializerr has_scalehas_zero_pointkeysconvert_quant_typeoriginal_quant_typeconvert_has_scaleconvert_has_zero_pointrrr_is_valid_per_tensorsj             z/TensorQuantOverridesHelper._is_valid_per_tensorrLlist[dict[str, Any]]c Cs^||v}|sdd|dfS|dd}|dur!dd|dfSt||j}t|}|}|dkr6||7}|dks@|t|krMdd|d t|d fSt|d krpt|||krpdd |d |d||dt|d fSd|dvr~dd|dfS|dd} | r|j| |dd} |dd} d|dv} d|dv} | o| }| r| r| r| sdS|r|jt|d}|rddd |dfSd|dv}d|dv}|o|}|r|r|s|rdSt |d dD]\}}t |t sdd|d |d!fSd|vrdd|dfSd|vr%| |dkr%d"Sd|vr;||dkr;||dkr;d#Sd|vrJ| |dkrJd$Sd|vrY| |dkrYd%Sd|voad|v}|ru|sudd&|d'|dfS|r|jt|}|rddd |dfSd|vod|v}|r|sdd(|d'|dfSqd)S)*NFTensor 'z6' has per-channel overrides, but is not an initializerrr z!Per-channel overrides for tensor z< is missing an 'axis' value in the first channel dictionary.z0Axis override value is out-of-bounds for tensor z (rank )rz1Incorrect number of channel overrides for tensor z (axis z ), expected z , but found .rHz8Cannot use 'convert' override for initializers, such as r r r r;r<rRrSrTrUr,r+)Fz6Must provide both 'rmin' and 'rmax' if one is providedz'Tensor quantization overrides at index z for 'rQ)FzTChannel quantization types for tensor '{tensor_name}' do not match at index {index}.)FzHChannel axis for tensor '{tensor_name}' does not match at index {index}.)FzSChannel symmetric value for tensor '{tensor_name}' does not match at index {index}.)FzVChannel reduce_range value for tensor '{tensor_name}' does not match at index {index}.z]Per-channel overrides that specify scale/zero_point must do so for all channels, but tensor 'z' is missing them at index zVPer-channel overrides that specify rmin/rmax must do so for all channels, but tensor 'rV) rlistdimslenr.rKr/rXrIrY enumeraterrW)rrZr1rLr\r weight_shape weight_rank norm_axisr r r r]r^ has_scale_zpr_has_rminhas_rmax has_rmin_rmaxindexrMchan_has_scale_zpchan_has_rmin_rmaxrrr_is_valid_per_channels          &  z0TensorQuantOverridesHelper._is_valid_per_channelrZdict[str, onnx.TensorProto]activation_namesset[str]cCst|_|jrr|jD]e\}}||vr"||vr"dd|dfSt|ts1dd|dfS|s4q t|dtsEdd|dfS|dsJq |dd }t|d kpZ|du}|rf| |||S| ||||dSd S) NFrfz5' in TensorQuantOverrides is not present in the modelrPz' are not in a listrz.Tensor quantization overrides at index 0 for 'rQr rrV) rIr.r-itemsrrirWrrkrwrd)rrZryr[r1rLr is_per_channelrrris_validis*  z#TensorQuantOverridesHelper.is_validTnew_valschannelslist[int] | None overwritec Cs|sdS|dur t|nd}|j|}d}|s<|rr?rr?)r1r2r>rErrE)rrG)r1r2rMrrrO)r1r2rLrerrO)rZrxryrzrrO)NT) r1r2r~rrrrr3rr3)rr2rrrr rr)NN) rr2rr2rrrr rr rr)rr2)rr3)rr*)rr2rr)rr2rr)rr2)r"r#r$r%r0r8r:r=rDrFrNrdrwr}rrrrrrrrrrrrr __classcell__rrrrr(?sB        U * '  '      r() __future__rrcollections.abcr dataclassesrtypingronnx quant_utilsrrr(rrrrs     ,