§ |”j‡ãó¼—UddlmZddlZddlZddlmZmZddlZddl m Z ddlmZm Z mZmZddlmZerddlZed Zd ed<gZGd„d ¦«Z ddd„ZdS)é)ÚannotationsN)Ú TYPE_CHECKINGÚLiteral)Ú TypeAliasé)ÚEagerParamBaseÚdefault_main_programÚin_dygraph_modeÚuse_pir_apié)Úlazy_init_helper©ÚsigmoidÚlinearÚconv1dÚconv2dÚconv3dÚconv1d_transposeÚconv_transpose1dÚconv2d_transposeÚconv_transpose2dÚconv3d_transposeÚconv_transpose3dÚtanhÚreluÚ leaky_reluÚselurÚ _NonLinearitycóN—eZdZdZdd„Z ddd „Z ddd„Z ddd „Zdd„Zdd„Z dS)ÚInitializeraBase class for parameter initializers Defines the common interface of parameter initializers. They add operations to the init program that are used to initialize parameter. Users should not use this class directly, but need to use one of its implementations. ÚreturnÚNonecó—dS©N©)Úselfs úq/lsinfo/ai/hellotax_ai/data_center/backend/venv/lib/python3.11/site-packages/paddle/nn/initializer/initializer.pyÚ__init__zInitializer.__init__As€ØˆóNÚparamú paddle.TensorÚblockúpaddle.pir.Block | Nonecó€—t¦«js| ||¦«S| ||¦«Sr$)r ÚstateÚforwardÚ _lazy_init©r&r*r,s r'Ú__call__zInitializer.__call__Ds=€õ Ñ!Ô!Ô'ð .Ø—<’< uÑ-Ô-Ð-àŠ˜u eÑ,Ô,Ð,r)úpaddle.Tensor | Nonecó—t‚)z;Add corresponding initialization operations to the network.)ÚNotImplementedErrorr2s r'r0zInitializer.forwardLs €õ"Ð!r)cóÄ—t¦«sJ‚dd„}| tj|j¦«¦«tj||j¦«|_|S)z+ Apply lazy initialization r*r+r,r-có¼—t¦«r|}n| d|¬¦«}t¦«5|||¦«ddd¦«dS#1swxYwYdS)NT)r,)rÚ_to_static_varr )r0r*r,Únew_vars r'Úinit_op_creatorz/Initializer._lazy_init..init_op_creatorZs²€õ‰}Œ}ð BØà×.Ò.¨t¸5Ð.ÑAÔAå!Ñ#Ô#ð (ð (Ø˜ Ñ'Ô'Ð'ð (ð (ð (ñ (ô (ð (ð (ð (ð (ð (ð (ð (øøøð (ð (ð (ð (ð (ð (s· AÁAÁA©r*r+r,r-)r Ú set_init_funcÚ functoolsÚpartialr0Ú_init_op_creator)r&r*r,r;s r'r1zInitializer._lazy_initRsp€õÑ Ô Ð Ð Ð ð (ð (ð (ð (ð ×ÒIÔ-¨d¬lÑ;Ô;Ñ<Ô<Ð<Ý!*Ô!2Ø˜Tœ\ñ" ô" ˆÔðˆr)úpaddle.pir.BlockcóJ—|€ t¦« ¦«}|Sr$)r Úglobal_block)r&r,s r'Ú_check_blockzInitializer._check_blockms#€Øˆ=Ý(Ñ*Ô*×7Ò7Ñ9Ô9ˆEàˆr)Úvarútuple[int, int]có”—t|t¦«r| ¦«r|jn|j}|rt|¦«dkrdx}}ntt|¦«dkr|dx}}nVt|¦«dkr|d}|d}n2t j|dd…¦«}|d|z}|d|z}||fS)aèCompute the fan_in and the fan_out for layers This method computes the fan_in and the fan_out for neural network layers, if not specified. It is not possible to perfectly estimate fan_in and fan_out. This method will estimate it correctly for matrix multiply and convolutions. Args: var: variable for which fan_in and fan_out have to be computed. Returns: tuple of two integers (fan_in, fan_out). rréN)Ú isinstancerÚis_distÚ_local_shapeÚshapeÚlenÚnpÚprod)r&rErLÚfan_inÚfan_outÚreceptive_field_sizes r'Ú _compute_fanszInitializer._compute_fansssá€õ"˜3¥Ñ/Ô/ð Ø47·K²K±M´Mð ˆCÔÐà”ð ð ð 6˜E™ œ aš˜Ø Ð ˆFWWÝ ‰ZŒZ˜1Š_ˆ_Ø$ QœxÐ'ˆFWWÝ ‰ZŒZ˜1Š_ˆ_à˜1”XˆFØ˜A”hˆGˆGõ$&¤7¨5°°°¬9Ñ#5Ô#5Ð Ø˜1”XÐ 4Ñ4ˆFØ˜A”hÐ!5Ñ5ˆGà˜Ð Ð r))r!r"r$r<)r*r+r,r-r!r4)r,r-r!rA)rEr+r!rF) Ú__name__Ú __module__Ú__qualname__Ú__doc__r(r3r0r1rDrSr%r)r'r r 8sº€€€€€ððð ð ð ð ðFJð-ð-ð-ð-ð-ðFJð"ð"ð"ð"ð"ðFJðððððð6ðððð%!ð%!ð%!ð%!ð%!ð%!r)r ÚnonlinearityÚstrr*úbool | float | Noner!ÚfloatcóR—|€d}n3t|tttf¦«sJ‚t|¦«}ddddddddddddt jd¦«t jdd|dzzz¦«ddœ}|| ¦«vr||Std |›d ¦«‚)a® Get the recommended ``gain`` value of some nonlinearity function. ``gain`` value can be used in some ``paddle.nn.initializer`` api to adjust the initialization value. Args: nonlinearity(str): name of nonlinearity activation function. If it is a linear function, such as: `linear/conv1d/conv2d/conv3d/conv1d_transpose/conv2d_transpose/conv3d_transpose` , 1.0 will be returned. param(bool|int|float|None, optional): optional parameter for some nonlinearity function. Now, it only applies to 'leaky_relu'. Default: None, it will be calculated as 0.01 in the formula. Returns: A float value, which is the recommended gain for this nonlinearity function. Examples: .. code-block:: python >>> import paddle >>> gain = paddle.nn.initializer.calculate_gain('tanh') >>> print(gain) 1.6666666666666667 >>> # 5.0 / 3 >>> gain = paddle.nn.initializer.calculate_gain('leaky_relu', param=1.0) >>> print(gain) 1.0 >>> # math.sqrt(2.0 / (1+param^2)) >>> initializer = paddle.nn.initializer.Orthogonal(gain) Ng{®Gáz„?rg«ªªªªªú?g@rHgè?rznonlinearity function z is not supported now.)rIÚboolÚintr[ÚmathÚsqrtÚkeysÚ ValueError)rXr*Úrecommended_gains r'Úcalculate_gainrd›s×€ð@ €}Øˆˆå˜%¥$UÐ!3Ñ4Ô4Ð4Ð4Ð4Ýe‘”ˆàØØØØØØØØØØØÝ” ˜#‘”Ý”i q¨5°!©8¡|Ñ 4Ñ5Ô5ØððÐð"Ð'×,Ò,Ñ.Ô.Ð.Ð.Ø Ô-Ð-åØI \ÐIÐIÐIñ ô ð r)r$)rXrYr*rZr!r[)Ú __future__rr>r_ÚtypingrrÚnumpyrNÚtyping_extensionsrÚbase.frameworkrr r rÚ lazy_initr ÚpaddlerÚ__annotations__Ú__all__r rdr%r)r'úrnsKðð#Ð"Ð"Ð"Ð"Ð"Ð"àÐÐÐØ€€€Ø)Ð)Ð)Ð)Ð)Ð)Ð)Ð)àÐÐÐØ'Ð'Ð'Ð'Ð'Ð'ððððððððððððð(Ð'Ð'Ð'Ð'Ð'àðØ€M€M€Mà&ð ô €Mðððñð$ €ð`!ð`!ð`!ð`!ð`!ñ`!ô`!ð`!ðH59ð; ð; ð; ð; ð; ð; ð; r)