# UH*nUS;aTUO[TU5R5v M, g7f))r r!N)getattrto_tuple).0kselfs `/var/www/html/shao/venv/lib/python3.13/site-packages/transformers/models/aimv2/modeling_aimv2.py 'Aimv2Output.to_tuple..Hs<   LLDGRYZ^`aRbRkRkRm m s14)tuplekeysr)s`r*r&Aimv2Output.to_tupleGs# YY[   )__name__ __module__ __qualname____firstlineno____doc__rrtorch FloatTensor__annotations__rrrrr rr!r-rr&__static_attributes__r2r1r*rr)s&)-D(5$$ %,48hu001837OXe//07/3K%++,304L(5,,-448186:3: %* r1rRMSNormc8^\rSrSrSU4SjjrSrSrSrU=r$) Aimv2RMSNormNc>[TU]5 [R"[R "U55UlX lg)z+ Aimv2RMSNorm is equivalent to T5LayerNorm N)super__init__r Parameterr8onesweightvariance_epsilon)r) hidden_sizeeps __class__s r*rBAimv2RMSNorm.__init__Ps/ ll5::k#:; #r1cURnUR[R5nUR S5R SSS9nU[R "X0R-5-nURURU5-$)NT)keepdim) dtypetor8float32powmeanrsqrtrFrE)r) hidden_states input_dtypevariances r*forwardAimv2RMSNorm.forwardXsw#)) %((7  $$Q',,R,>% H?T?T4T(UU {{]--k:::r1c^[URR5SUR3$)Nz, eps=)r-rEshaperFr/s r* extra_reprAimv2RMSNorm.extra_repr_s* ))*+6$2G2G1HIIr1)rFrE)gư>) r3r4r5r6rBrXr\r; __classcell__rIs@r*r>r>Ns$;JJr1r>c.^\rSrSrU4SjrSrSrU=r$)Aimv2MLPcc>[TU]5 XlURUlURUl[ R "URURURS9Ul[ R "URURURS9Ul [ R "URURURS9Ul [URUl g)Nbias)rArBconfigrGintermediate_sizerLinearmlp_bias gate_projup_proj down_projr hidden_actact_fnr)rfrIs r*rBAimv2MLP.__init__ds  !--!'!9!94#3#3T5K5KRXRaRabyy!1!143I3IPVP_P_` 4#9#94;K;KRXRaRabV../ r1cURURURU55URU5-5nU$N)rlrnrjrk)r)xrls r*rXAimv2MLP.forwardns6NN4;;t~~a/@#ADLLQRO#ST r1)rnrfrlrjrGrgrk)r3r4r5r6rBrXr;r^r_s@r*raracs0r1rac^\rSrSrS\4U4Sjjr\SSS\R4S\R4Sjj5r S \RS\R4S jr S r U=r $) Aimv2VisionEmbeddingssrfcB>[TU]5 XlURUl[R "UR URURURS9Ul[URUR5Ul URUR-S-nURR(d%[R"X!R5UlUR!S["R$"U5R'S5SS9 g)N) kernel_sizestriderL position_idsrrMF persistent)rArBrf patch_sizerConv2d num_channelsrG patch_embedr> rms_norm_epsrms_norm image_size is_native Embeddingposition_embeddingregister_bufferr8arangeexpand)r)rf num_patchesrIs r*rBAimv2VisionEmbeddings.__init__ts   ++99   !3!3ARAR[a[l[l %V%7%79L9LM ((F,=,==!C {{$$&(ll;@R@R&SD # ^U\\+-F-M-Mg-Vchir1g@cpur"c[R"[U5XTS9n[R"[U5XTS9n[R"XgSS9upvUS-n[R"XUS9U- n SX9-- n UR 5SU SSS24-n UR 5SU SSS24-n [R "U R 5U R5U R 5U R5/SS9SSS2SS24$) NrOdevicexy)indexingg?).Nrdim)r8rintmeshgridflattenconcatsincos) heightwidth embed_dim temperaturerrOgrid_wgrid_hpos_dimomegaout_hout_ws r*"build_2d_sincos_position_embedding8Aimv2VisionEmbeddings.build_2d_sincos_position_embeddingsc%jEc&kFFq. W&AGK{)* +eD!Gn< +eD!Gn<||UYY[%))+uyy{EIIKPVWXY]_`bcYcddr1 pixel_valuescUR5u p#nURU5RS5RSS5nUR U5nUR R (aTURX0R-X@R-UR RURURS9nOURUR5nXV-nU$)NrLr)rrrO)sizerr transposerrfrrrrGrrOrr{)r)r_rrrU pos_embeds r*rXAimv2VisionEmbeddings.forwards*//11e((6>>qAKKAqQ  m4 ;; ??//)(++11$++#)) @I//0A0ABI%1 r1)rfrrrr)r3r4r5r6rrB staticmethodr8rQTensorrrXr;r^r_s@r*rvrvssb j0 j!$'%u}}e ee ELLU\\r1rvc ^\rSrSrS\4U4SjjrS S\\RS\\RS\\RS\R4Sjjr S r U=r $) Aimv2TextEmbeddingsrfcN>[TU]5 URn[R"UR U5Ul[R"URU5UlURS[R"UR5RS5SS9 g)Nr{r|Fr}) rArBrGrr vocab_sizetoken_embeddingmax_position_embeddingsrrr8rr)r)rfrrIs r*rBAimv2TextEmbeddings.__init__s && !||F,=,=yI"$,,v/M/My"Y  ELL)G)GHOOPWXej  r1 input_idsr{ inputs_embedsr"c<UbURSOURSnURRRSnXE:a[SUSU35eUcURSS2SU24nUcUR U5nURU5nX6-nU$)NrMrzRSequence length must be less than max_position_embeddings (got `sequence length`: z and max_position_embeddings: )r[rrE ValueErrorr{r)r)rr{r seq_lengthmax_position_embeddingposition_embeddings embeddingss r*rXAimv2TextEmbeddings.forwards -6,AY__R(}GZGZ[]G^ !%!8!8!?!?!E!Ea!H  .d,<=S[TU]5 XlURUlUR UlURUR -UlURUR -UR:wa&[SURSUR S35eURS-Ul URUl SUl [R"URURURS9Ul[R"URURURS9Ul[R"URURURS9Ul[R"URURURS9Ulg)Nz;embed_dim must be divisible by num_heads (got `embed_dim`: z and `num_heads`: z).gFrd)rArBrfrGrnum_attention_heads num_headshead_dimrscaleattention_dropoutr is_causalrrhqkv_biask_projv_projq_projout_projros r*rBAimv2Attention.__init__s0  ++33$..8 ==4>> )T^^ ;MdnnM]^NN#2' ]]D( // iiV__U iiV__U iiV__U  $..$..vW r1rUrr"c 2URupEnURU5nURU5nURU5n UR XEUR UR 5RSS5nUR XEUR UR 5RSS5nU R XEUR UR 5RSS5n [n URRS:wa[URRn U "UUUU UURURUR(dSO URS9upU R!XEU5R#5n UR%U 5n X4$)z#Input shape: Batch x Time x ChannelrrLeager)rrr)r[rrrviewrrrrrf_attn_implementationrrrrrreshaperr) r)rUrr batch_sizerrqueriesr.valuesattention_interfacerrs r*rXAimv2Attention.forwardsS-:,?,?) ++m,{{=)]+,,zt~~t}}U__`acdeyyOYYZ[]^_ZT^^T]]S]]^_abc(? ;; + +w 6"9$++:Z:Z"[ $7     nnJJ#}}C$,, % ! "))*)LWWY mmK0 ((r1) rfrrrrrrrrrrrr) r3r4r5r6r7rBr8rrr-rXr;r^r_s@r*rrs[GX,26$)||$)!.$) u||Xell33 4 $)$)r1rc ^\rSrSrS\4U4SjjrS S\RS\\RS\\ S\ \R\R44Sjjr S r U=r $) Aimv2EncoderLayeri#rfc>[TU]5 [U5Ul[ U5Ul[ URUR5Ul [ URUR5Ul grr) rArBr attentionraffnr>rGr rms_norm1 rms_norm2ros r*rBAimv2EncoderLayer.__init__$sZ '/F#%f&8&8&:M:MN%f&8&8&:M:MNr1rUroutput_attentionsr"cURU5nURXBS9upVX-nURU5nURU5nX-nU(aX4$US4$)N)rUr)rrrr)r)rUrrnorm_hidden_statesrr mlp_outputs r*rXAimv2EncoderLayer.forward+sk "^^M:$(NNASN$s! %3 !^^M:XX01 %2 0A ,\ W[G\\r1)rrrrNF)r3r4r5r6rrBr8rrboolr-rXr;r^r_s@r*rr#ssO0O26,1 ]||]!.]$D> ] u||U\\) * ]]r1rc ^\rSrSrSrS\4U4Sjjr\S S\\ RS\\ S\\ S\ 4S jj5r S rU=r$) Aimv2Encoderi<z Transformer encoder consisting of `config.num_hidden_layers` self attention layers. Each layer is a [`Aimv2EncoderLayer`]. Args: config: Aimv2Config rfc>[TU]5 Xl[R"[ UR 5Vs/sHn[U5PM sn5UlSUl gs snfr) rArBrfr ModuleListrangenum_hidden_layersrlayersgradient_checkpointing)r)rfrrIs r*rBAimv2Encoder.__init__EsS  mmfNfNfHg$hHg1%6v%>Hg$hi &+#%isA%rroutput_hidden_statesr"cFUbUOURRnUbUOURRnU(aSOSnU(aSOSnUnURH-nU(aXW4-nU"UUUS9n U SnU(dM%XiS4-nM/ U(aXW4-n[ UUUS9$)a Args: inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`): Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert `input_ids` indices into associated vectors than the model's internal embedding lookup matrix. attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*): Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`: - 1 for tokens that are **not masked**, - 0 for tokens that are **masked**. [What are attention masks?](../glossary#attention-mask) output_attentions (`bool`, *optional*): Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned tensors for more detail. output_hidden_states (`bool`, *optional*): Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for more detail. return_dict (`bool`, *optional*): Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple. Nr2)rrr)last_hidden_staterU attentions)rfrr r r ) r)rrrr encoder_statesall_attentionsrU encoder_layer layer_outputss r*rXAimv2Encoder.forwardLs<2C1N-TXT_T_TqTq$8$D $++JjJj  40d% ![[M#!/2B!B)"3M *!,M  !/3C2E!E) +.>>N+(%  r1)rfr r r)r3r4r5r6r7rrBrrr8rrr rXr;r^r_s@r*rr<sl,{,26,0/3 < !.< $D> < 'tn <  < < r1rcj^\rSrSrS\4U4SjjrS\RS\R4SjrSr U=r $)Aimv2AttentionPoolingHeadirfc>[TU]5 URUlURUl[ R "URURURS9Ul[ R "URURURS9Ul [ R"[R"SSUR55Ul [ R "URURSS9Ulg)NrdrT)rArBrGrrrrhrrrrCr8zeros cls_token output_projros r*rB"Aimv2AttentionPoolingHead.__init__s !--33ii 0 0$2B2BY ii 0 0$2B2BY ekk!Q8H8H&IJ99T%5%5t7G7GdSr1rUr"cURup#nURRUSS5nURU5R X#UR X@R -5nUR U5R X#UR X@R -5nUR USUR X@R -5nURSSSS5nURSSSS5nURSSSS5n[R"XU5n U RSS5R USU5n U RSS9n URU 5n U $)NrMrrrLrr) r[rrrrrrpermuteFscaled_dot_product_attentionrrSr) r)rUrseq_len hidden_dimrrrrroutputs r*rX!Aimv2AttentionPoolingHead.forwards8*7*=*=' ZNN))*b"= kk-(00dnnV`drdrVrs M*22:XbftftXtu!!*a~~A]^kk!Q1% aAq) aAq)44UG !++Aq199*aT !&&1&- !!+. r1)rrGrrrr) r3r4r5r6rrBr8rrXr;r^r_s@r*rrs2 T0 TU\\ellr1rcT^\rSrSr%Sr\\S'SrSr/SQr Sr Sr Sr U4Sjr SrU=r$) Aimv2PreTrainedModeliz An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained models. The model is only intended for inference and doesn't support finetuning. rfaimv2T)rrrvrc>[TU]U5 [US5(ad[UR[ R 5(a:URRR[R"S55 gg[U[5(a9URRRSURRS9 gg)N logit_scaleg$I$I,@r)rSstd)rA _init_weightshasattr isinstancer(rrCdatafill_mathlogrrnormal_rfinitializer_range)r)rrIs r*r*"Aimv2PreTrainedModel._init_weightss f% 6= ) )&,,bll;;""''--dhhx.@A<  9 : :    ! ! ) )s 8U8U ) V;r1r2)r3r4r5r6r7rr:base_model_prefixsupports_gradient_checkpointing_no_split_modules_supports_sdpa_supports_flash_attn_supports_flex_attnr*r;r^r_s@r*r%r%sC &*# NWWr1r%zL The Vision model from AIMv2 without any head or projection on top. ) custom_introc ^\rSrSr%\\S'SrS\4U4SjjrS\R4Sjr \ \ S S\ \RS\ \S \ \S\4S jj55rS rU=r$) Aimv2VisionModelirfrc>>[TU]U5 Xl[U5Ul[ U5Ul[URUR5Ul URUl UR(a[U5Ul UR5 grr)rArBrfrvrrencoderr>rGrruse_headrhead post_initros r*rBAimv2VisionModel.__init__so   /7#F+ $V%7%79L9LM  ==1&9DI r1r"c.URR$rr)rrr/s r*get_input_embeddings%Aimv2VisionModel.get_input_embeddingss***r1rrr cfUbUOURRnUbUOURRnURU5nUR UUUS9nUSnUR U5nUR (aURU5OSn[UUURURS9$)ar Examples: ```python >>> from PIL import Image >>> import requests >>> from transformers import AutoProcessor, Siglip2VisionModel >>> model = Aimv2VisionModel.from_pretrained("apple/aimv2-large-patch14-native") >>> processor = AutoProcessor.from_pretrained("apple/aimv2-large-patch14-native") >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg" >>> image = Image.open(requests.get(url, stream=True).raw) >>> inputs = processor(images=image, return_tensors="pt") >>> outputs = model(**inputs) >>> last_hidden_state = outputs.last_hidden_state >>> pooled_output = outputs.pooler_output # pooled features ```N)rrr rr pooler_outputrUr) rfrr rr>rr?r@rrUr) r)rrrr rUencoder_outputsrrHs r*rXAimv2VisionModel.forwards:2C1N-TXT_T_TqTq$8$D $++JjJj  5 ,,'/!5'  ,A. MM*;<8<  "344 )/')77&11   r1)rfrr>r@rr?r)r3r4r5r6rr:main_input_namerBrModulerDrrrr8rrrrXr;r^r_s@r*r<r<s $O 0 +bii+26,0/3 2 !.2 $D> 2 'tn 2 $ 2 2 r1r<zJ The text model from AIMv2 without any head or projection on top. c ^\rSrSrSrS\4U4SjjrS\R4Sjr Sr \ \ S S\ \RS \ \S \ \S\4S jj55rS rU=r$)Aimv2TextModelirrfc>[TU]U5 Xl[U5Ul[ U5Ul[URUR5Ul URUl UR5 grr) rArBrfrrrr>r>rGrr eos_token_idrAros r*rBAimv2TextModel.__init__$s_   -f5#F+ $V%7%79L9LM "// r1r"c.URR$rrrrr/s r*rD#Aimv2TextModel.get_input_embeddings/s...r1c$XRlgrrrS)r)rs r*set_input_embeddings#Aimv2TextModel.set_input_embeddings2s */'r1rrr c UbUOURRnUbUOURRnURU5nURupgn[ R "U[ RURS9n U RS5RUS5n Ub[URUU UU SS9nURUUUUS9n U Sn URU 5n U [ R "U RSU RS9UR[ RU RS9UR :HR5R#SS94n [%U U U R&U R(S9$) NrrrM)rf input_embedsr{rcache_positionpast_key_values)rrrr )rrrG)rfrr rr[r8rlongr unsqueezerr r>rrPrrPargmaxrrUr)r)rrrr rUrr rrZr{rIr pooled_outputs r*rXAimv2TextModel.forward5s2C1N-TXT_T_TqTq$8$D $++JjJj  2 !.!4!4 QgUZZ H\H\]%//299*bI  %/{{*)-- $ N,,')/!5 ' ,A. MM*;<* LL*003rPrr)r3r4r5r6rKrrBrrLrDrVrrrr8rrrrXr;r^r_s@r*rNrNs "O  /bii/026,0/3 0 !.0 $D> 0 'tn 0 $ 0 0 r1rNtensorr"c[R"US5n[R"USSS9n[R"US5nU$)z This method is equivalent to tensor.norm(p=2, dim=-1, keepdim=True) and used to make model `executorch` exportable. See issue https://github.com/pytorch/executorch/issues/3566 rLrMT)rrNg?)r8rRsum)ra square_tensor sum_tensor normed_tensors r*_get_vector_normrgjs< IIfa(M=b$?JIIj#.M r1c^\rSrSr%\\S'/SQrS\4U4Sjjr\SS\ \ RS\ \ RS\ \ RS\ \ S \ \ S \ R4 S jj5r\SS \ \ RS\ \ S \ \ S \ S \ R4 Sjj5r\\SS\ \ R"S \ \ RS\ \ RS\ \ S \ \ S \4 Sjj55rSrU=r$) Aimv2Modeliurf)rrrvc>[TU]U5 URUlURRUlUR RUl[RUR5Ul [RUR 5Ul [R"UR URSS9Ul[R"URURSS9Ul[R""[$R&"UR(R*55Ul[.R0"UR25UlUR75 g)NFrd)rArBprojection_dim vision_configrGvision_embed_dim text_configtext_embed_dimr< _from_config vision_modelrN text_modelrrhvisual_projectiontext_projectionrCr8rarflogit_scale_init_valuer(r/r0max_logit_scalemax_log_logit_scalerAros r*rBAimv2Model.__init__zs  $33 & 4 4 @ @$00<<,99&:N:NO(55f6H6HI!#4+@+@$BUBU\a!b!yy)<)Q>QX]^<< T[[5W5W(XY#'88F,B,B#C  r1rrr{rr r"cUbUOURRnUbUOURRnURUUUUUS9nURnUR U5nU$)aG Returns: text_features (`torch.FloatTensor` of shape `(batch_size, output_dim`): The text embeddings obtained by applying the projection layer to the pooled output of [`Aimv2TextModel`]. Examples: ```python >>> from transformers import AutoTokenizer, Aimv2Model >>> model = Aimv2Model.from_pretrained("openai/aimv2-vit-base-patch32") >>> tokenizer = AutoTokenizer.from_pretrained("openai/aimv2-vit-base-patch32") >>> inputs = tokenizer(["a photo of a cat", "a photo of a dog"], padding=True, return_tensors="pt") >>> text_features = model.get_text_features(**inputs) ```)rrr{rr )rfrr rrrHrt) r)rrr{rr  text_outputsr_ text_featuress r*get_text_featuresAimv2Model.get_text_featuress42C1N-TXT_T_TqTq$8$D $++JjJj 48??)%/!5 4C4 %22 ,,]; r1rinterpolate_pos_encodingcUbUOURRnUbUOURRnURUUUUS9nURnUR U5nU$)a Returns: image_features (`torch.FloatTensor` of shape `(batch_size, output_dim`): The image embeddings obtained by applying the projection layer to the pooled output of [`Aimv2VisionModel`]. Examples: ```python >>> from PIL import Image >>> import requests >>> from transformers import AutoProcessor, Aimv2Model >>> model = Aimv2Model.from_pretrained("openai/aimv2-vit-base-patch32") >>> processor = AutoProcessor.from_pretrained("openai/aimv2-vit-base-patch32") >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg" >>> image = Image.open(requests.get(url, stream=True).raw) >>> inputs = processor(images=image, return_tensors="pt") >>> image_features = model.get_image_features(**inputs) ```)rrr r~)rfrr rqrHrs)r)rrr r~vision_outputsr_image_featuress r*get_image_featuresAimv2Model.get_image_featuress>2C1N-TXT_T_TqTq$8$D $++JjJj 6:5F5F%/!5%= 6G6 '44 // >r1c `UbUOURRnUbUOURRnURUUUS9nUR UUUUS9nUR nUR U5nUR n URU 5n U[U5- nU [U 5- n URRSUR5R5RU R5n X-UR5-n U R5n [!U U U UUUS9$)a Examples: ```python >>> from PIL import Image >>> import requests >>> from transformers import AutoProcessor, Aimv2Model >>> model = Aimv2Model.from_pretrained("apple/aimv2-large-patch14-224-lit") >>> processor = AutoProcessor.from_pretrained("apple/aimv2-large-patch14-224-lit") >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg" >>> image = Image.open(requests.get(url, stream=True).raw) >>> inputs = processor( ... text=["a photo of a cat", "a photo of a dog"], images=image, return_tensors="pt", padding=True ... ) >>> outputs = model(**inputs) >>> logits_per_image = outputs.logits_per_image # this is the image-text similarity score >>> probs = logits_per_image.softmax(dim=1) # we can take the softmax to get the label probabilities ```)rrr )rrrr r)rrrrr r!)rfrr rqrrrHrsrtrgr(clamprwexprPrtr) r)rrrrr rrzrrr(rrs r*rXAimv2Model.forwardsSB2C1N-TXT_T_TqTq$8$D $++JjJj 6:5F5F%/!56G6  48??)/!5 4C4 &33 --l; "00 **;7 $&6|&DD !$4[$AA &&,,S$2J2JKOOQTTU`UgUgh &4 8HH*,,.-+#%* .   r1) r(rwrkrorrrtrmrqrs)NNNNN)NNNF)r3r4r5r6rr:r6rBrrr8rrr9r|rrrrrXr;r^r_s@r*ririus ]{$-115/3,0/3 )ELL))!.)u||, ) $D> ) 'tn )   ))V59,0/3). -u001-$D>-'tn - #' -    --^154815,0/3 F E,,-F u001F !. F $D> F 'tn F  F F r1ri)r<rir%rN)r)5r/ dataclassesrtypingrrrr8torch.nn.functionalrrr activationsr integrationsr masking_utilsr modeling_layersr modeling_outputsr rmodeling_utilsrrutilsrrrconfiguration_aimv2rrrrrLr>rarvrrfloatrrrrrr%r<rNrgri__all__r2r1r*rs. !** !7/9KFBBPP   +     FY'J299J(J( ryy  1BII1h%"))%^% II% <<% % << % U\\* %  %%.:)RYY:)z]2]2M 299M ` DW?WW8 I +I  I X F )F  F RU\\ellz %z z z Wr1