U yh&@sddlZddlmZddlmZmZmZddlmZddl m Z m Z m Z ddl mZmZmZdddgZe d ejjjd ZGd d d ejjjZGd ddeejZGdddeejZGdddeejZdS)N)_single_pair_triple) _FusedModule)TupleTypeVarUnion) _size_1_t _size_2_t _size_3_tConv1dConv2dConv3dMOD)boundc@s|eZdZeZd eeeedfeedfeedfeedfeeedfeeedd ddZ ddZ e dd d Z d d Z dS)_ConvNdN.) in_channels out_channels kernel_sizestridepaddingdilation transposedoutput_paddinggroupsbias padding_modereturnc CsX| |d}tjjjj|||||||||| | | f || s@td| |_| j|d|_dS)N)devicedtypez'qconfig must be provided for QAT module)factory_kwargs) nnmodulesconvr__init__AssertionErrorqconfigweightweight_fake_quant)selfrrrrrrrrrrrr&rrr r*N/var/www/html/venv/lib/python3.8/site-packages/torch/ao/nn/qat/modules/conv.pyr$s   z_ConvNd.__init__cCs||||j|jSNZ _conv_forwardr(r'rr)inputr*r*r+forward,sz_ConvNd.forwardFc Cst||jks(td|jd|jjt|ds:td|jsHtdtt|tr^|d}|j}||j|j |j |j |j |j |j|jdk |j|d }|j|_|j|_|S) zCreate a qat module from a float module Args: `mod`: a float module, either produced by torch.ao.quantization utilities or directly from user zqat.z.from_float only works for r&z,Input float module must have qconfig definedz,Input float module must have a valid qconfigrN)rrrrrrr&)type _FLOAT_MODULEr%__name__hasattrr& issubclassrrrrrrrrrrr')clsmoduse_precomputed_fake_quantr&Zqat_convr*r*r+ from_float/s4z_ConvNd.from_floatc Cst|}||j|j|j|j|j|j|j|j dk |j }t j |j|_|j dk rjt j |j |_ t|tr|g}t|dst|}|||j|}||j|S|SdS)z This works for both single qat conv, and the qat conv - relu modules to convert the qat module to a floating point module N_FLOAT_RELU_MODULE)r1_FLOAT_CONV_MODULErrrrrrrrrtorchr! Parameterr'detachr5rr4r%r:appendr2trainZtraining)r)r6r#r"ZreluZfusedr*r*r+to_floatJs0      z_ConvNd.to_float)NNN)F)r3 __module__ __qualname__rr2intrboolstrr$r0 staticmethodr9rAr*r*r*r+rs,       rc s`eZdZdZejZejZd eee e e e e fe ee e dd fdd Z edfd d ZZS)r aZ A Conv1d module attached with FakeQuantize modules for weight, used for quantization aware training. We adopt the same interface as :class:`~torch.nn.Conv1d` Similar to :class:`~torch.nn.Conv2d`, with FakeQuantize modules initialized to default. Attributes: weight_fake_quant: fake quant module for weight rTzerosN rrrrrrrrrrc s^t|} t|}t|tr|nt|}t|}tj||| |||dtd||| | | | ddSNFr) rrrrrrrrr&rr)r isinstancerFsuperr$r)rrrrrrrrrr&rrZ kernel_size_Zstride_Zpadding_Z dilation_ __class__r*r+r$xs( zConv1d.__init__Fcstj|||dSN)r8rMr9r6r7r8rOr*r+r9szConv1d.from_float) rHrrHrHTrINNN)F)r3rBrC__doc__r!r r2r;rDr rrFrEr$ classmethodr9 __classcell__r*r*rOr+r hs4  !c sheZdZdZejZejZdeee e e e e fe ee e dd fdd Z d d Zedfd d ZZS)r a A Conv2d module attached with FakeQuantize modules for weight, used for quantization aware training. We adopt the same interface as `torch.nn.Conv2d`, please see https://pytorch.org/docs/stable/nn.html?highlight=conv2d#torch.nn.Conv2d for documentation. Similar to `torch.nn.Conv2d`, with FakeQuantize modules initialized to default. Attributes: weight_fake_quant: fake quant module for weight rHrTrINrJc s^t|} t|}t|tr|nt|}t|}tj||| |||dtd||| | | | ddSrK)rrLrFrMr$rNrOr*r+r$s( zConv2d.__init__cCs||||j|jSr,r-r.r*r*r+r0szConv2d.forwardFcstj|||dSrQrRrSrOr*r+r9szConv2d.from_float) rHrrHrHTrINNN)F)r3rBrCrTr!r r2r;rDr rrFrEr$r0rUr9rVr*r*rOr+r s6 !c sheZdZdZejZejZdeee e e e e fe ee e dd fdd Z d d Zedfd d ZZS)ra A Conv3d module attached with FakeQuantize modules for weight, used for quantization aware training. We adopt the same interface as `torch.nn.Conv3d`, please see https://pytorch.org/docs/stable/nn.html?highlight=conv3d#torch.nn.Conv3d for documentation. Similar to `torch.nn.Conv3d`, with FakeQuantize modules initialized to default. Attributes: weight_fake_quant: fake quant module for weight rHrTrINrJc s^t|} t|}t|tr|nt|}t|}tj||| |||dtd||| | | | ddSrK)rrLrFrMr$rNrOr*r+r$s( zConv3d.__init__cCs||||j|jSr,r-r.r*r*r+r0 szConv3d.forwardFcstj|||dSrQrRrSrOr*r+r9 szConv3d.from_float) rHrrHrHTrINNN)F)r3rBrCrTr!rr2r;rDr rrFrEr$r0rUr9rVr*r*rOr+rs6 !)r<Ztorch.nnr!Ztorch.nn.modules.utilsrrrZtorch.ao.nn.intrinsicrtypingrrrZtorch.nn.common_typesr r r __all__r"r#rrr r rr*r*r*r+s  W5: