fThnSSKJrJrJrJr SSKrSSKJr SSKJ r SSK J r SSK J r Jr SSKJrJr SSKJr SS KJrJrJrJr S S KJr \R6"\5r"S S \R<5r"SS\R<5r "SS\R<5r!S2S\R<S\RDS\RDS\RDS\\RDS\#S\#4Sjjr$Sr%S\RDS\&S\RD4Sjr'S\RDS \RDS!\RDS"\RDS\\RD\RD44 S#jr("S$S%\R<5r)"S&S'\R<5r*"S(S)\R<5r+"S*S+\R<5r,\"S,S-\55r-\"S.S/9"S0S1\-55r.S-S1/r/g)3)CallableOptionalTupleUnionN)ACT2FN)FlashAttentionKwargs)BaseModelOutputBaseModelOutputWithPooling)ALL_ATTENTION_FUNCTIONSPreTrainedModel)Unpack)auto_docstringcan_return_tuplelogging torch_int)MLCDVisionConfigcb^\rSrSrU4SjrS\R S\R 4SjrSrU=r $)MLCDMLP&c>[TU]5 Xl[URUl[ R"URUR5Ul [ R"URUR5Ul gN) super__init__configr hidden_act activation_fnnnLinear hidden_sizeintermediate_sizefc1fc2selfr __class__s ^/var/www/auris/envauris/lib/python3.13/site-packages/transformers/models/mlcd/modeling_mlcd.pyrMLCDMLP.__init__'sb  #F$5$5699V//1I1IJ99V55v7I7IJ hidden_statesreturnclURU5nURU5nURU5nU$r)r#rr$)r&r+s r(forwardMLCDMLP.forward.s4/ **=9 / r*)rrr#r$) __name__ __module__ __qualname____firstlineno__rtorchTensorr.__static_attributes__ __classcell__r's@r(rr&s)KU\\ellr*rcf^\rSrSrS S\S\SS4U4SjjjrS\S\S\R4S jr S r U=r $) MLCDRotaryEmbedding5dimthetar,Nc >[TU]5 SU[R"SUS[RS9U- -- nUR SUSS9 g)N?rdtypeinv_freqF persistent)rrr4arangefloatregister_buffer)r&r<r=rCr's r(rMLCDRotaryEmbedding.__init__6sK %ELLC%++$NQT$TUV ZeDr*num_patches_heightnum_patches_widthc~[R"XRRS9R S5R SU5n[R"X RRS9R S5R US5n[R "UR5UR5/SS9n[X5n[R"X`RRURRS9n[R"XpR5nXRS5n U $)aE Calculate the Rotary Position Embedding (RoPE) for MLCDVisionModel based on the grid size. Args: num_patches_height (int): Number of patches in the height dimension. num_patches_width (int): Number of patches in the width dimension. Returns: torch.Tensor: Rotary positional embeddings for the given grid size. )devicerrr<)rMrB) r4rFrCrM unsqueezeexpandstackflattenmaxrBouter) r&rJrKhpos_idswpos_idspos_ids max_grid_sizeseqrotary_pos_emb_fullrotary_pos_embs r(r.MLCDRotaryEmbedding.forward;s LL+MM4H4H I S STU V ] ]^`bs t  LL*==3G3G H R RST U \ \]oqs t  ++x//183C3C3EFBO.B ll=1E1ET]]M`M`a#kk#}}=-5==a@r*)g@) r0r1r2r3intrGrr4r5r.r6r7r8s@r(r:r:5sJECEEDEE ##%,,r*r:c^\rSrSrS\4U4SjjrS\RS\S\S\R4Sjr S \RS\R4S jr S r U=r $) MLCDVisionEmbeddings\rcv>[TU]5 XlURUlUR UlUR Ul[R"[R"UR55Ul [R"URURUR UR SS9UlUR UR -S-UlURS-UlUR#S[R$"UR 5R'S5SS9 g)NF) in_channels out_channels kernel_sizestridebiasr@r position_ids)rrNrD)rrrr! embed_dim image_size patch_sizer Parameterr4randnclass_embeddingConv2d num_channelspatch_embedding num_patches num_positionsrHrFrQr%s r(rMLCDVisionEmbeddings.__init__]s  ++ ++ ++!||EKK,GH!yy++??  !OOt>1D!--1 ^U\\$:L:L-M-T-TU\-]jopr* embeddingsheightwidthr,cURSS- nURRRS5nURSS- n[R R 5(d%XF:Xa X#:XaURUR5$USS2SS24nUSS2SS24nURSn X R-n X0R-n [US-5n URSXU 5nURSSSS5n[RRUX4SS S 9nURSSSS5RSSU 5n[R "Xx4SS 9$) a This method allows to interpolate the pre-trained position encodings, to be able to use the model on higher resolution images. This method is also adapted to support torch.jit tracing. Adapted from: - https://github.com/facebookresearch/dino/blob/de9ee3df6cf39fac952ab558447af1fa1365362a/vision_transformer.py#L174-L194, and - https://github.com/facebookresearch/dinov2/blob/e1277af2ba9496fbadf7aec6eba56e8d882d1e35/dinov2/models/vision_transformer.py#L179-L211 rrNrNg?rr@bicubicF)sizemode align_cornersrO)shapeposition_embeddingweightrPr4jit is_tracingrirlrreshapepermuter functional interpolateviewcat) r&rvrwrxrsrrtclass_pos_embedpatch_pos_embedr< new_height new_widthsqrt_num_positionss r(interpolate_pos_encoding-MLCDVisionEmbeddings.interpolate_pos_encodingrsi!&&q)A- !44;;EEaH*003a7 yy##%%+*F6?**4+<+<= =,QU3,QU3r". __, &}c'9:)11!5G]`a)11!Q1=--33 ( 4 *11!Q1=BB1b#Nyy/;CCr* pixel_valuescHURSnURRRnURUR US95nUR S5R SS5nURRUSS5n[R"XT/SS9nU$)NrrAr@rrNrO) r~rrrrBtorS transposerorQr4r)r&r batch_size target_dtype patch_embeds class_embedsrvs r(r.MLCDVisionEmbeddings.forwards!''* ++2288 ++LOO,O,OP #++A.88A> ++22:q"E YY ;C r*)rorrjrkrsrtrrrl)r0r1r2r3rrr4r5r_r FloatTensorr.r6r7r8s@r(rara\sgq/q*'D5<<'D'DUX'D]b]i]i'DR E$5$5 %,,  r*ramodulequerykeyvalueattention_maskscalingdropoutc @[X R5n[X0R5n [R"XR SS55U-n Ub"USS2SS2SS2SUR S24n X-n [ RRU S[RS9RUR5n [ RRXURS9n [R"X5n U R SS5R5n X4$)Nr@rrN)r<rB)ptrainingr) repeat_kvnum_key_value_groupsr4matmulrr~rrsoftmaxfloat32rrBrr contiguous) rrrrrrrkwargs key_states value_states attn_weights causal_mask attn_outputs r(eager_attention_forwardrs3 ; ;