+ jT^^RIt^RIHt^RIHt^RIHt^RIt^RIH uH t ^RIH t ^RI H t ^RIHt^RIHt^R IHt^R IHtHt^R IHtHtHt^R IHtHt^R IH t ^RI!H"t"H#t#H$t$^RI%H&t&^RI'H(t(H)t)^RI*H+t+H,t,^RI-H.t.H/t/H0t0H1t1H2t2^RI3H4t4H5t5^RI6H7t7H8t8^RI9H:t:^RI;Ht>^RI?H@t@^RIAHBtB^RICHDtD^RIEHFtFHGtG^RIHHItIHJtJ^RIKHLtL^RIMHNtNRR ltO!R!R"]+R#R$7tP]0!R%R&]44tQ!R'R(]>4tR!R)R*] P4tT!R+R,] P4tU!R-R.]B4tV!R/R0] P4tW!R1R2] P4tX]1!R3R4]J44tY!R5R6]L4tZ!R7R8] 4t[!R9R:]I4t\]0!R;R<7!R=R>]\44t]]0!R?R<7!R@RA]\44t^]0!RBR<7]!RCRD]"444t_!RERF] P4t`!RGRH] P4ta]0!RIR<7!RJRK]\44tb.RLOtcR#)MN)Callable) dataclass) Annotated)nn) functional)initialization)TorchvisionBackend) BatchFeature)group_images_by_shapereorder_images) ImageInputPILImageResamplingSizeDict)create_bidirectional_mask(create_bidirectional_sliding_window_mask)GradientCheckpointingLayer)BaseModelOutputBaseModelOutputWithPoolingMaskedLMOutput)ROPE_INIT_FUNCTIONS)ALL_ATTENTION_FUNCTIONSPreTrainedModel) ImagesKwargsUnpack) TensorTypeTransformersKwargsauto_docstringno_inherit_decoratortorch_compilable_check)IMAGENET_STANDARD_MEANIMAGENET_STANDARD_STD)can_return_tuplemaybe_autocast)capture_outputs) positive_int) Gemma4RMSNorm)apply_rotary_pos_emb)LagunaRotaryEmbedding)convert_image_to_patches)eager_attention_forward repeat_kv)MuseGlimmerPreTrainedModelMuseGlimmerTextAttention) NemotronMLP NeoMMEConfigc V^8dQhR\R\R\R,R\R,R\\\3,/#)heightwidthmax_sideNsizereturn)intrtuple)formats"j/home/luca/screen-checking/.venv/lib/python3.14/site-packages/transformers/models/neomme/modular_neomme.py __annotate__r<9sG&)&)3&)s&)cDj&)PX[_P_&)dijmorjrds&)cVe VPMRpVe VPMRpVe$W,V8dW@V,, R,MRpVe\Wb\W4, 4pVe \WeW,, R,4pVR8XdW3#\^\ W,44p\^\ W,44pVRJ;'d\Wx4V8;'gVRJ;'d Wx,V8p V 'd\^\ P !W,44p\^\ P !W,44pVe;Wx,V8d.Wx8d\^WX,4pWx3#\^WW,4pWx3#VeW,V8dWx,V8dWx3#\^\ P!W,44p \^\ P!W,44p VRJ;'d\W4V8;'gVRJ;'d W,V8p V 'gW3#Wx3#)zHCompute integer height and width from the configured image size targets.Ng??) min_pixels max_pixelsminmaxroundmathfloorceil) r3r4r5r6r@rAscaleresized_height resized_width exceeds_cap ceiled_height ceiled_widths &&&& r;get_resize_output_sizerN9s$($4$J$($4$J6@6LQWQ_blQlZE> *s 2ruEEc&&889E&.9cAB |}E&.12N5/0M4'YYC,NQY,Y$NN>#AJ#NQ 6> :;Atzz%-89  !n&Dz&Q.!$Q (C!D,,!$Az'C D ,,&.:"=  )Z 7!0 0Atyy89 1dii 67 t+[[M0PS[0[  d " P P}'Cj'P  . .  ((r=c0a]tRt^btoRtV3RltRtVtR#)NeoMMEImageProcessorKwargsz patch_size (`int`, *optional*, defaults to `self.patch_size`): Side, in pixels, of one patch token. The image is padded to a whole multiple of it. max_side (`int`, *optional*): Longest-side cap in pixels. Unset means no longest-side resize. c<V^8dQh/S[S[S[!43,;R&S[S[R,S[!43,;R&#)r2 patch_sizeNr5)rr8r%)r: __classdict__s"r;r<'NeoMMEImageProcessorKwargs.__annotate__bs;#|~-..d LN233r=N)__name__ __module__ __qualname____firstlineno____doc____annotate_func____static_attributes____classdictcell__rSs@r;rPrPbsr=rPF)totalcaa]tRt^ntoRt]t]Pt ] t ] t RtRtRtRtRt^ tRtRtRR.tV3RlV3Rllt]V3RlV3R ll4tV3R lV3R lltV3R lR ltV3RlRltV3RlRltRV3RlRlltRtVt V;t!#)NeoMMEImageProcessora{ Constructs an image processor for NeoMME. The processor converts each image to RGB and splits it into row-major patches. It preserves the original size by default, but `max_side` or a pixel-area `size` can resize images. It pads the bottom and right edges to complete the final patch. A batch returns one concatenated patch tensor and one grid height and width pair per image. TN pixel_values image_grid_hwc0<V^8dQhRS[S[,/#)r2kwargs)rrP)r:rSs"r;r<!NeoMMEImageProcessor.__annotate__s##(B!C#r=c (<\SV`!R/VBR#)NrU)super__init__)selfre __class__s&,r;riNeoMMEImageProcessor.__init__s "6"r=c<<V^8dQhRS[RS[S[,RS[/#)r2imagesrer7)r rrPr )r:rSs"r;r<rfs)444v>X7Y4^j4r=c &<\SV`!V3/VB#)z Returns: [`BatchFeature`] with `pixel_values` of shape `(total_patches, 3 * patch_size ** 2)` and `image_grid_hw` of shape `(batch_size, 2)`. `pixel_values` concatenates patches from every image in the batch. )rh preprocess)rjrnrerks&&,r;rpNeoMMEImageProcessor.preprocesssw!&3F33r=c <V^8dQhRS[/#r2r7)r9)r:rSs"r;r<rfs==u=r=c <VR,pVe'\\V44RR08wd \R4hVR,fVPRR4\SV`!R/VB#)r6Nr@rAz4size must contain exactly min_pixels and max_pixels. do_resizerU)setdict ValueErrorpoprh_validate_preprocess_kwargs)rjrer6rks&, r;rz0NeoMMEImageProcessor._validate_preprocess_kwargss]f~  DJL,3O OST T &> ! JJ{D )w2vh(]%  *vo =L~C{{HPT{UUr=c B<V^8dQhRRRS[RS[RS[S[3,/#)r2r}r~rRr7)r8r9)r:rSs"r;r<rfs1...C.ER`begjRjLk.r=cVPR,VPR,rCV)V,)V)V,)reWR,V, pWb,V, pV'g V'd\P!V^^W.^R7pWV3#)r2)fillrr)rtvFpad) rjr}rRr3r4 grid_height grid_width pad_height pad_widths &&& r;_pad_to_patch_grid'NeoMMEImageProcessor._pad_to_patch_gridsw BR%+Gz$9":uf >R!%!7!7X`!a,: "5 )&<((>@TU/D^fv/w,,#%  %3%9%9%; !E>7;6M6Mn^h6i 3NK!77 LV_N/G~Wa.b $U +).z*+c..AA*  &&<&&>@TU &':]!;'   r=c,<V^8dQhRS[RS[RS[/#)r2r3r4r7)r8)r:rSs"r;r<rfs(DD#DcDZ]Dr=c T;'g/pVPRVP4pVPRVP4pVPVPRVP4R7R,pVPRVP 4'd\ WWV4wrV)V,)V)V,),#)z Return the number of image patches for one image, excluding row markers. The method applies the effective resize settings, then rounds each image dimension up to a whole patch. Values in `images_kwargs` override the processor settings. rRr5r6)r6ru)getrRr5_standardize_kwargsr6rurN)rjr3r4 images_kwargsrRr5r6s&&&& r;get_number_of_image_patches0NeoMMEImageProcessor.get_number_of_image_patchess&++ "&&|T__E  $$Z?'']->->vtyy-Q'RSYZ   [$.. 9 926(QMFJ&'eVz-A+BCCr=rUgp?N)"rVrWrXrYrZrP valid_kwargsrBILINEARrr rr!rdo_convert_rgbrurrrrRr5r6model_input_namesrirrprzrrrrr\r] __classcell__rkrSs@@r;rarans.L!**H'J%INIJNLJH D'9##444== V V..0 0 dDDDr=rac]tRt^tRtR#) NeoMMERMSNormrUNrVrWrXrYr\rUr=r;rrr=rcTaa]tRtRtoRtV3RlV3RlltV3RlRltRtVtV;t #)NeoMMEEmbeddingszZFactorized (ALBERT-style) token embeddings: `vocab_size -> embedding_rank -> hidden_size`.c <V^8dQhRS[/#r2configr/)r:rSs"r;r<NeoMMEEmbeddings.__annotate__see|er=c<\SV`4\P!VPVP 4Vn\P!VP VPRR7Vn R#FbiasN) rhrir Embedding vocab_sizeembedding_rankword_embeddingsLinear hidden_sizeembedding_projectionrjrrks&&r;riNeoMMEEmbeddings.__init__sO !||F,=,=v?T?TU$&IIf.C.CVEWEW^c$d!r=cN<V^8dQhRS[PRS[P/#)r2 input_idsr7)r LongTensorTensor)r:rSs"r;r<rs)JJ!1!1JellJr=cBVPVPV44#rrr)rjrs&&r;forwardNeoMMEEmbeddings.forwards(()=)=i)HIIr=r rVrWrXrYrZrirr\r]rrs@@r;rrs&dee JJJr=rcTaa]tRtRtoRtV3RlV3RlltV3RlRltRtVtV;t #)NeoMMEPatchEmbeddingsi z