קg!$ddlmZmZddlmZmZddlmZmZm Z m Z m Z m Z ddl Z ddl mZddlmZddlmZddlmZgd ZGd d eZed d GddeZed d GddeZ e e eefefZde_ed d GddeZed d GddeZeGddZGddeZdS))ABCabstractmethod) dataclassfield)CallableDictListOptionalTupleUnionN)Tensor)ObserverOrFakeQuantize)"_ObserverOrFakeQuantizeConstructor)Node) QuantizerQuantizationSpecBaseQuantizationSpecFixedQParamsQuantizationSpec EdgeOrNodeSharedQuantizationSpecDerivedQuantizationSpecQuantizationAnnotationceZdZdZdS)rzBase class for different types of quantization specs that allows users to specify how to quantize a Tensor (input/output of a Node) in the model N)__name__ __module__ __qualname____doc__e/var/www/html/ai-engine/env/lib/python3.11/site-packages/torch/ao/quantization/quantizer/quantizer.pyrrsrrT)eqfrozenceZdZUdZejed<eed<dZe e ed<dZ e e ed<dZ e ej ed<dZ e e ed<d Zeed <d ZdS) rzQuantization spec for common operators that allows user to specify how to quantize a Tensor, this includes dtype, quant_min, quant_max etc. dtypeobserver_or_fake_quant_ctrN quant_min quant_maxqschemech_axisF is_dynamicc|j7|j0|j|jkr td|jd|jd|j|jdkrtddSdS)Nz quant_min z must be <= quant_max .rzCh_axis is < 0.)r&r' ValueErrorr))selfs r __post_init__zQuantizationSpec.__post_init__1s N &*//TT^TT4>TTT  < # q(8(8.// / $ #(8(8r)rrrrtorchr$__annotations__rr&r intr'r(r)r*boolr/rrr rrs ; !CBBB#Ix}####Ix}###'+GXem $+++!GXc]!!!J00000rrceZdZUejed<eed<eed<dZe eed<dZ e eed<dZ e ej ed<dZ e ed <dS) rr$scale zero_pointNr&r'r(Fr*)rrrr0r$r1floatr2r&r r'r(r*r3rrr rrCs ; LLLOOO#Ix}####Ix}###'+GXem $+++Jrrz)torch.ao.quantization.quantizer.quantizerceZdZUdZeed<dS)rzg Quantization spec for the Tensors whose quantization parameters are shared with other Tensors edge_or_nodeN)rrrrrr1rrr rrXs* rrceZdZUdZeeed<eeege e e ffed<e j ed<dZ eeed<dZeeed<dZee jed<dZeeed <d Zeed <dS) rz^Quantization spec for the Tensors whose quantization parameters are derived from other Tensors derived_fromderive_qparams_fnr$Nr&r'r(r)Fr*)rrrrr rr1rrr r r0r$r&r r2r'r(r)r*r3rrr rrbshhz""""&ffn@U UVVVV ;#Ix}####Ix}###'+GXem $+++!GXc]!!!JrrceZdZUdZeeZeee e fe d<dZ e e e d<dZ ee d<dZee d <dS) rzHow are input arguemnt or output should be quantized, expressed as QuantizationSpec, this corresponds to how a Tensor in the operator Graph is observed (PTQ) or fake quantized (QAT) )default_factoryinput_qspec_mapN output_qspecTallow_implicit_sharingF _annotated)rrrrrdictr?rrr rr1r@rAr3rBrrr rrps CH%CCCOT$)= >>? 48L(/0777 $(D'''JrrceZdZdejjdejjfdZedejjdejjfdZedejjddfdZ dS)rmodelreturnc|S)aAllows for user defined transforms to run before annotating the graph. This allows quantizer to allow quantizing part of the model that are otherwise not quantizable. For example quantizer can a) decompose a compound operator like scaled dot product attention, into bmm and softmax if quantizer knows how to quantize bmm/softmax but not sdpa or b) transform scalars to tensor to allow quantizing scalares. Note: this is an optional method rr.rEs r transform_for_annotationz"Quantizer.transform_for_annotations  rcdSNrrHs r annotatezQuantizer.annotate rNcdSrKrrHs r validatezQuantizer.validaterMr) rrrr0fx GraphModulerIrrLrOrrr rrs X)         eh2 ux7K   ^  eh2 t   ^   rr)abcrr dataclassesrrtypingrrr r r r r0r torch.ao.quantizationrtorch.ao.quantization.qconfigrtorch.fxr__all__rrrrrrrrrrrr rYs########((((((((???????????????? 888888LLLLLL   3  d4    0 0 0 0 0+ 0 0!  0F d4   #7!  5t$d* + C  d4   1!  d4        2  !    0          r