fTh*LSrSSKrSSKrSSKrSSKJrJrJrJr SSK r SSK r SSK J r SSK J r SSKJrJrJr SSKJr SS KJr SS KJrJrJr SS KJr SS KJrJrJr SS K J!r!J"r"J#r# SSK$J%r% \"RL"\'5r(Sr)"SS\ RT5r+"SS\ RT5r,"SS\ RT5r-"SS\ RT5r.\!"SS\55r/\!"SS\/55r0\!"SS9"SS \/\55r1\!"S!S9"S"S#\/55r2/S$Qr3g)%zPyTorch OpenAI ImageGPT model.N)AnyOptionalTupleUnion)nn)autocast)BCEWithLogitsLossCrossEntropyLossMSELoss)ACT2FN)GenerationMixin))BaseModelOutputWithPastAndCrossAttentions!CausalLMOutputWithCrossAttentions SequenceClassifierOutputWithPast)PreTrainedModel)Conv1D find_pruneable_heads_and_indicesprune_conv1d_layer)auto_docstringlogging torch_float)ImageGPTConfigcf SSKnSSKn[ R RU5n[RSRU55 URRU5n/n/nUHtup[RSRX55 URRXY5n URU 5 URU R55 Mv [Xx5GHupU SSn U R!S5n [#SU 55(d U S S ;a5[RS RSR%U 555 MqUn U S S ;a ['U S 5n U GHgn UR)SU 5(aUR!SU 5nOU /nUSS:Xd USS:Xa ['U S5n OUSS:Xa ['U S5n OUSS:Xd USS:Xa['XS5n ['U S5n OUSS;a['U S5n ['U S5n O[+U 5S:Xa-U SS:Xa$USS:Xa['XS5n ['U S5n ORUSS:Xa['U S5n ['U S5n O0USS:Xa['U S5n ['U S5n O['XS5n [+U5S :dGMU[-US5nXn GMj [+U 5S:a U SS:XdU S S:XdU S S:Xd U S S:XaOU R.U R.:Xde[RS!RU 55 U S S":Xad[4R6"U R9UR:UR:55R<U R>SS2SUR:24'GMU S S#:Xaq[4R6"U R9UR:UR:55R<U R>SS2UR:S UR:-24'GMSU S S$:Xag[4R6"U R9UR:UR:55R<U R>SS2S UR:-S24'GM[+U 5S:XaTU SS:XaKU S S:XaB[4R6"U R9UR:UR:55U lGM&U S S:Xa[4R6"U 5U lGMMU S S:Xa9[4R6"U 5U R>SUR@S- 2SS24'GMU S S:Xa&[4R6"U 5U R>S 'GM[4R6"U 5U lGM U$![a [R S5 ef=f![0a1nU=R2U R.U R.4- sleSnAff=f)%z( Load tf checkpoints in a pytorch model rNzLoading a TensorFlow model in PyTorch, requires TensorFlow to be installed. Please see https://www.tensorflow.org/install/ for installation instructions.z(Converting TensorFlow checkpoint from {}z"Loading TF weight {} with shape {}/c3,# UH nUS;v M g7f))adam_vadam_mAdamWeightDecayOptimizerAdamWeightDecayOptimizer_1 global_stepN).0ns f/var/www/auris/envauris/lib/python3.13/site-packages/transformers/models/imagegpt/modeling_imagegpt.py .load_tf_weights_in_imagegpt..Qs  n ns)_stepz Skipping {})wtet transformerz [A-Za-z]+\d+z(\d+)wgweightbbiaswpewte)q_projk_projv_projc_attnr rattnc_projr,lm_headsoszInitialize PyTorch weight {}r5r6r7)!re tensorflow ImportErrorloggererrorospathabspathinfoformattrainlist_variables load_variableappendsqueezezipsplitanyjoingetattr fullmatchlenintshapeAssertionErrorargstorch from_numpyreshapen_embdTdata vocab_size)modelconfigimagegpt_checkpoint_pathr>tftf_path init_varsnamesarraysnamerUarraypointerm_name scope_namesnumes r'load_tf_weights_in_imagegptrn0sf ggoo67G KK:AA'JK''0I E F  8??LM&&w5 T emmo& ! 5) ABxzz#     "X " KK ,,SXXd^< =  88 #g}5GF||OV44 hhx8 %h 1~$ A#(=!'84Q3&!'62Q5(KNe,C!'q>:!'84Q#AA!'84!'84TaDGv$5+a.H:T!'q>:!'84Q6)!'95!'84Q5(!'51!'84!'q>:;1$+a.)!,;> t9q=T!W.$r(f2DRTYHY]abd]ein]n  }} 333  299$?@ 8x /4/?/? fmm]c]j]j@k/l/n/nGLLOfmmO+ , "X !AFAQAQ fmmV]];Ba LLFMMA ,=== >"X !383C3CEMMRXR_R_agananDo3p3r3rGLLA -// 0 Y!^Q6 1d1g6I ++EMM&--,WXGL "X  ++E2GL "X 7<7G7G7NGLL06,,q00!3 4 "X $//6GLL  ++E2GLY*\ LC   Q   P" 7==%++66 s#WW5!W25 X0?,X++X0cx^\rSrSrS S\\S\4U4SjjjrS\RS\R4Sjr Sr U=r $) ImageGPTLayerNorm hidden_sizeepsc>[TU]5 X l[R"[ R "U55UlgN)super__init__rsr ParameterrXTensorr0)selfrrrs __class__s r'rwImageGPTLayerNorm.__init__s, ll5<< #<= tensorreturnc U[R"[R"[R"U5SSS9UR-5- nXR -nU$)Nr*T)axiskeepdim)rXsqrtmeansquarersr0)rzr~s r'forwardImageGPTLayerNorm.forwardsI%**UZZ V0D2W[%\_c_g_g%ghh++% r})rsr0)gh㈵>) __name__ __module__ __qualname____firstlineno__rrTfloatrwrXryr__static_attributes__ __classcell__r{s@r'rprps?>E#J>U>> ellu||r}rpc8^\rSrSrSS\\S\\4U4SjjjrSrSSjr SSjr Sr S r SS \ RS \\S \\ RS \\ RS\\ RS\\ RS\\S\\S\4SjjrSrU=r$)ImageGPTAttentionis_cross_attention layer_idxc >[TU]5 URnURS[R "[R "XD4[RS95RSSXD5SS9 URS[R"S5SS9 URUl URUl URUR-UlURUlURUR-UR:wa&[!SURS URS 35eUR"UlX lUR&UlX0lUR*UlUR$(aN[-S UR-UR5Ul[-URUR5UlO([-S UR-UR5Ul[-URUR5Ul[4R6"UR85Ul[4R6"UR<5Ul[A5Ul!g) Nr2dtyperF) persistent masked_biasgz=`embed_dim` must be divisible by num_heads (got `embed_dim`: z and `num_heads`: z).r=r )"rvrwmax_position_embeddingsregister_bufferrXtrilonesboolviewr~rr embed_dimnum_attention_heads num_headshead_dim split_size ValueErrorscale_attn_weightsrscale_attn_by_inverse_layer_idxrreorder_and_upcast_attnrr8q_attnr:rDropout attn_pdrop attn_dropout resid_pdrop resid_dropoutset pruned_heads)rzr`rr max_positionsr{s r'rwImageGPTAttention.__init__s 66    JJuzz="@ S T Y Y1m    ]ELL,>5Q++33$..8 .. ==4>> )T^^ ;OPTP^P^O_`NN#2'  #)";";"406/U/U,"'-'E'E$  " " T^^!3T^^DDK @DK T^^!3T^^DDKT^^T^^< JJv'8'89ZZ(:(:;Er}c8[U5S:Xag[XRURUR5up[ R "X"UR-USUR--/5n[URUSS9Ul [URUSS9Ul URUR-UR[U5- -UlUR[U5- UlURRU5Ulg)Nrr=rdim) rSrrrrrXcatrrr8r:union)rzheadsindex index_attns r' prune_headsImageGPTAttention.prune_headss u:? 7~~t}}^b^o^op YYt'>T__I\@]^_ )jaH (eC  ??dnn<RUV[R\A\]#e*4 --33E:r}cT[R"XRSS55nUR(a U[ UR S5S-5- nUR (aU[URS-5- nUR(dUR S5UR S5pURSS2SS2X- U2SU24n [R"UR5Rn [R"XRURS9n [R "XU 5nUbXd-n["R$"SS9"U5nUR'UR5nUR)U5nUbXe-n[R"Xc5n X4$)Nr*?rrdevicer)rXmatmul transposerrsizerrrrr2finforminr~rwhererSoftmaxtyper) rzquerykeyvalueattention_mask head_mask attn_weights query_length key_length causal_mask mask_value attn_outputs r'_attnImageGPTAttention._attns`||E==R+@A  " "'+ejjn6K*LLL  / /'%0B*CCL&&',zz"~sxx|*))Aq**Cj*PR]S]R]$]^K\%7%78<sAKKM#2&))DD i(~~aAq))r}cURSSSS5R5nUR5SSX#-4-nURU5$)zC Merges attn_head_size dim and num_attn_heads dim into hidden_size rr=rr Nr)r contiguousrrrs r' _merge_headsImageGPTAttention._merge_headsFsM1a+668KKM#2&)*D)FF {{9%%r} hidden_states layer_pastrrencoder_hidden_statesencoder_attention_mask use_cacheoutput_attentionsrc Ub[[US5(d [S5eURU5n URU5R UR SS9upUnO,URU5R UR SS9upn UR XRUR5n UR XRUR5n UR XRUR5n Ub0Uup[R"X4SS9n [R"X4SS9n USLaX4nOSnUR(aURXXU5unnOURXXU5unnURXRUR5nURU5nUR!U5nX4nU(aUU4- nU$)NrzIf class is used as cross attention, the weights `q_attn` have to be defined. Please make sure to instantiate class with `ImageGPTAttention(..., is_cross_attention=True)`.r=rrT)hasattrrrr8rNrrrrrXrrrrrr:r)rzrrrrrrrrrrrpast_key past_valuepresentrroutputss r'rImageGPTAttention.forwardNs ! ,4** t KK .E%:;AA$//WXAYJC3N $ M : @ @VW @ X E!!%G^^T]]C!!%G  !#- H))XO4CIIz1r:E  lGG  ' '(,(G(GTYkt(u %K(, 5uV_(` %K'' ^^T]]S kk+. ((5 (    &Gr})rr8r:rrrrrrrrrrrr)FN)NNNNNNNFF)rrrrrrrTrwrrrrrrXrytuplerrrrs@r'rrs)"8D>)"V^_bVc)")"V ;$)L2)h*&&*15,08<9=$),13||3TN3!. 3 ELL) 3 ( 5 3!) 63D>3$D>3 33r}rcb^\rSrSrU4SjrS\R S\R 4SjrSrU=r $) ImageGPTMLPic>[TU]5 URn[X5Ul[X15Ul[ URUl[R"UR5Ul gru) rvrwrrrc_fcr:r activation_functionactrrrdropout)rzintermediate_sizer`rr{s r'rwImageGPTMLP.__init__sZ && ,8 Y: &445zz&"4"45 r}rrcURU5nURU5nURU5nURU5nU$ru)rr r:r )rzrs r'rImageGPTMLP.forwards@ -0 /  M2  ]3 r})r rr:r ) rrrrrwrXryrrrrs@r'rrs(6U\\ellr}rc^\rSrSrSU4SjjrSS\R S\\S\\R S\\R S\\R S\\R S \\S \\S \ 4S jjr S r U=r $) ImageGPTBlockic>[TU]5 URnURb UROSU-n[ X1R S9Ul[XS9Ul[ X1R S9Ul UR(a([USUS9Ul [ X1R S9Ul [XA5Ulg)NrsrT)rr)rvrwrrn_innerrplayer_norm_epsilonln_1rr9ln_2add_cross_attentioncrossattention ln_cross_attnrmlp)rzr`rrr inner_dimr{s r'rwImageGPTBlock.__init__s (( &,nn&@FNNa+o %k7P7PQ %fB %k7P7PQ  % %"3Ft_h"iD !2;D]D]!^D y1r}rrrrrrrrrc Un URU5nURUUUUUUS9n U Sn U SSn X-nUbW[US5(d[SUS35eUn UR U5nUR UUUUUUS9n U Sn X-nXSS-n Un UR U5nURU5nX-nU4U(aU -n U $U SS-n U $) N)rrrrrrrrz'If `encoder_hidden_states` are passed, z` has to be instantiated with cross-attention layers by setting `config.add_cross_attention=True`)rrrrrr=)rr9rrrrrr)rzrrrrrrrrresidual attn_outputsrrcross_attn_outputsfeed_forward_hidden_statess r'rImageGPTBlock.forwardsG! -0 yy !)/ ! #1o qr"#. ,4!122 =dVDZZ%H ..}=M!%!4!4-#&;'="3 "5" -Q/K$2M12 66G  -0 %)XXm%<" =  "gLAH Lr})r9rrrrrrur) rrrrrwrXryrrrrrrrs@r'rrs 2$&*15,08<9=$),18||8TN8!. 8 ELL) 8 ( 5 8!) 68D>8$D>8 88r}rcH^\rSrSr\r\rSrSr Sr S/r U4Sjr Sr SrU=r$) ImageGPTPreTrainedModelir- input_idsTrc&>[TU]"U0UD6 gru)rvrw)rzinputskwargsr{s r'rw ImageGPTPreTrainedModel.__init__s &+F+r}c [U[R[45(ajURR R SURRS9 URb$URR R5 O[U[R5(awURR R SURRS9 URb1URR URR5 O:[U[5(a%URR RS5 UR5Hqup#SU;dM SU;dMUR R SURR[ R""SURR$-5- S9 Ms g)zInitialize the weights.g)rstdNrr:r0r=) isinstancerLinearrr0r]normal_r`initializer_ranger2zero_ Embedding padding_idxrpfill_named_parametersmathrn_layer)rzmodulergps r' _init_weights%ImageGPTPreTrainedModel._init_weightssS fryy&1 2 2 MM   & &CT[[5R5R & S{{&   &&(  - - MM   & &CT[[5R5R & S!!- ""6#5#56<<>  1 2 2 MM   $ $S )..0GD4H$4Cdkk.K.KdiiXY\`\g\g\o\oXoNp.pr1r}r$)rrrrr config_classrnload_tf_weightsbase_model_prefixmain_input_namesupports_gradient_checkpointing_no_split_modulesrwr;rrrs@r'r&r&s8!L1O%!O&*#(),ssr}r&c"^\rSrSrS\4U4SjjrSrSrSr\ SS\ \ RS\ \ \ \ RS \ \ RS \ \ RS \ \ RS \ \ RS \ \ RS\ \ RS\ \ RS\ \S\ \S\ \S\ \S\S\\ \44Sjj5rSrU=r$) ImageGPTModelir`c v>[TU]U5 URUl[R "UR UR5Ul[R "URUR5Ul [R"UR5Ul [R"[UR5Vs/sH n[!XS9PM sn5Ul[%URUR&S9UlSUlSUlSUlUR15 gs snf)NrrF)rvrwrrrrr3r^r4rr3r embd_pdropdrop ModuleListrangenum_hidden_layersrhrprln_fmodel_parallel device_mapgradient_checkpointing post_init)rzr`ir{s r'rwImageGPTModel.__init__ s  ++<< 1 14>>B<< > >OJJv001 ERXRjRjLklLkq f BLklm%dnn&:S:ST $&+#  msD6cUR$rur4rzs r'get_input_embeddings"ImageGPTModel.get_input_embeddingss xxr}cXlgrurTrznew_embeddingss r'set_input_embeddings"ImageGPTModel.set_input_embeddingss!r}cUR5H-up#URURRU5 M/ g)zf Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer} N)itemsrKr9r)rzheads_to_prunelayerrs r' _prune_headsImageGPTModel._prune_heads"s5+002LE FF5M   * *5 13r}r'past_key_valuesrtoken_type_ids position_idsr inputs_embedsrrrroutput_hidden_states return_dictr*rc ^$SU;a:[R"S[5 Ub [S5eUR S5nU bU OUR R n U bU OUR Rn U bU OUR Rn U bU OUR Rn UbUb [S5eUbFURX5 UR5nURSUS5nURSnO1Ub#UR5SSnURSnO [S5eUb URO URnUbURSUS5nUc%Sn[S/[!UR"5-5nOUSSRS 5nUc<[$R&"UUSU-[$R(US 9nUR+S5nUbyUS::a [S 5eURUS5nUSS2SSSS24nUR-UR.S 9nS U- [$R0"UR.5R2-nUR R4(aEUbBUR5unnnUU4nU c[$R6"UUS9n UR9U 5n OSn UR;X`R R<5nUcUR?U5nURAU5nUUR-UR5-m$UbUR?U5nT$U-m$URCT$5m$UT$RS54-nURD(a/URF(aU (a[HRKS5 Sn U (aSOSnU (aSOSnU (aUR R4(aSOSnU (aSOSn[M[OUR"U55GHununn URP(a[$RRRUT$R5 U b[U$4SjU 55n UbUR-T$R5n[WU[$RX5(aUR-T$R5nU (aUT$4-nURD(a8URF(a'UR[UR\T$SUUUUU U U 5 n!OU"T$U UUUUU U U S9n!U!Sm$U SLa UU!S4-nU (a?UU!U (aSOS4-nUR R4(aUU!U (aSOS4-nURP(dGMUR^Ra5HOun"n#UU#S:XdMS[cU"5-URd:wdM/T$R-S[cU"S-5-5m$MQ GM URgT$5m$T$R"U6m$U (aUT$4-nU (d[ST$UUUU455$[iT$UUUUS9$)aI input_ids (`torch.LongTensor` of shape `(batch_size, input_ids_length)`): `input_ids_length` = `sequence_length` if `past_key_values` is `None` else `past_key_values[0][0].shape[-2]` (`sequence_length` of input past key value states). Indices of input sequence tokens in the vocabulary. If `past_key_values` is used, only `input_ids` that do not have their past calculated should be passed as `input_ids`. Indices can be obtained using [`AutoImageProcessor`]. See [`ImageGPTImageProcessor.__call__`] for details. Examples: ```python >>> from transformers import AutoImageProcessor, ImageGPTModel >>> from PIL import Image >>> import requests >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg" >>> image = Image.open(requests.get(url, stream=True).raw) >>> image_processor = AutoImageProcessor.from_pretrained("openai/imagegpt-small") >>> model = ImageGPTModel.from_pretrained("openai/imagegpt-small") >>> inputs = image_processor(images=image, return_tensors="pt") >>> outputs = model(**inputs) >>> last_hidden_states = outputs.last_hidden_state ``` pixel_values`The `pixel_values` argument is deprecated and will be removed in v4.47, use `input_ids` instead.N_You cannot pass both `pixel_values` and `input_ids`. Please make sure to only pass `input_ids`.zDYou cannot specify both input_ids and inputs_embeds at the same timer*rz5You have to specify either input_ids or inputs_embedsrrz$batch_size has to be defined and > 0rr)rzZ`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`...Fr$c3X># UHoRTR5v M! g7fru)tor)r% past_staters r'r((ImageGPTModel.forward..s#&h]gz}}]5I5I'J'J]gs'*)rrrrrrrTrr=r zcuda:c30# UH nUcMUv M g7frur$)r%vs r'r(rpspAps )last_hidden_statercr attentionscross_attentions)5warningswarn FutureWarningrpopr`rrgruse_return_dict%warn_if_padding_and_no_attention_maskrrrUrrrSrKrXarangelong unsqueezernrrrrrinvert_attention_mask get_head_maskr8r4r3rGrOtrainingrA warning_once enumeraterMrMcuda set_devicer.ry_gradient_checkpointing_func__call__rNr^str last_devicerLr)%rzr'rcrrdrerrfrrrrrgrhr* input_shape batch_sizer past_lengthencoder_batch_sizeencoder_sequence_lengthrencoder_hidden_shapeposition_embedstoken_type_embeds output_shapepresentsall_self_attentionsall_cross_attentionsall_hidden_statesrQblockrrrrrrs% @r'rImageGPTModel.forward)s+^ V # MMr  $ u >2I1B1N-TXT_T_TqTq$8$D $++JjJj "+!6IDKKcd d  "  6 6y Q#..*K!r;r?;I"+J  &',,.s3K&,,Q/JTU U%.%:!! @T@T  %+00[_EN  "K#TFS[$89O)!,Q/44R8K   << [_{5RZ_ZdZdmstL'11!4L  %Q !GHH+00R@N ,AtT1,<=N,..TZZ.@N!N2ekk$**6M6Q6QQN ;; * */D/P=R=W=W=Y :  7$68O#P %-).4HQW)X&%)%?%?@V%W "%) " &&y++2E2EF   HHY/M((<0%(:(:=;O;O(PP  % $ 8 ),==M -0 "m&8&8&<%>>  & &4==##p" "2$5b4%64;;;Z;Zr`d"6BD&/DFFO0L&M "A"z"" %%m&:&:;)!&&h]g&h!hJ!-%3%6%6}7K7K%LNi66 ) ]-A-A BI#$58H$H!**t}};;NN!"aL)*%  !)#1'l*?+A'&7 $AJMD #wqzm3 &9W)QYZ=[<]&]#;;22+?7PY1_`CaBc+c(""" OO113DAqAbEzgA&6$:J:J&J(5(8(83q1u:9M(N 4e'Nl -0 %**L9  1]4D D '3DFY[op  9+$+*1   r}) rNrGrrOrKrLrMr3r4) NNNNNNNNNNNNN)rrrrrrwrVr[rarrrXryrrrrrrrrrs@r'rDrDsv~&"2-1@D1515/3,0048<9=$(,0/3&*d ELL)d "%ell(;"<=d !. d !. d u||, d ELL)d   -d  ( 5d !) 6d D>d $D>d 'tnd d^d d u?? @!d d r}rDz The ImageGPT Model transformer with a language modeling head on top (linear layer with weights tied to the input embeddings). ) custom_introc$\^\rSrSrS/rS\4U4SjjrSrSr\ SS\ \ RS\ \ \ \ RS \ \ RS \ \ RS \ \ RS \ \ RS \ \ RS\ \ RS\ \ RS\ \ RS\ \S\ \S\ \S\ \S\S\\ \44 Sjj5r\S\ \ \ RS\ RS\ \ \ R4Sj5rSrU=r$)ImageGPTForCausalImageModelingizlm_head.weightr`c>[TU]U5 [U5Ul[R "UR URS- SS9UlSUl SUl UR5 g)NrFr2) rvrwrDr-rr/r[r^r;rMrNrPrzr`r{s r'rw'ImageGPTForCausalImageModeling.__init__s[  (0yy0A0AA0EER $ r}cUR$rur;rUs r'get_output_embeddings4ImageGPTForCausalImageModeling.get_output_embeddings%s ||r}cXlgrurrYs r'set_output_embeddings4ImageGPTForCausalImageModeling.set_output_embeddings(s% r}r'rcrrdrerrfrrlabelsrrrgrhr*rc SU;a:[R"S[5 Ub [S5eUR S5nUbUOUR R nURUUUUUUUUU U U U US9 nUSnURU5nSnU brUSSS2SS24R5nU SS S24R5n[5nU"URSURS55URS55nU(dU4US S-nUbU4U-$U$[UUURURUR UR"S 9$) a input_ids (`torch.LongTensor` of shape `(batch_size, input_ids_length)`): `input_ids_length` = `sequence_length` if `past_key_values` is `None` else `past_key_values[0][0].shape[-2]` (`sequence_length` of input past key value states). Indices of input sequence tokens in the vocabulary. If `past_key_values` is used, only `input_ids` that do not have their past calculated should be passed as `input_ids`. Indices can be obtained using [`AutoImageProcessor`]. See [`ImageGPTImageProcessor.__call__`] for details. labels (`torch.LongTensor` of shape `(batch_size, input_ids_length)`, *optional*): Labels for language modeling. Note that the labels **are shifted** inside the model, i.e. you can set `labels = input_ids` Indices are selected in `[-100, 0, ..., config.vocab_size]` All labels set to `-100` are ignored (masked), the loss is only computed for labels in `[0, ..., config.vocab_size]` Examples: ```python >>> from transformers import AutoImageProcessor, ImageGPTForCausalImageModeling >>> import torch >>> import matplotlib.pyplot as plt >>> import numpy as np >>> image_processor = AutoImageProcessor.from_pretrained("openai/imagegpt-small") >>> model = ImageGPTForCausalImageModeling.from_pretrained("openai/imagegpt-small") >>> device = torch.device("cuda" if torch.cuda.is_available() else "cpu") >>> model.to(device) # doctest: +IGNORE_RESULT >>> # unconditional generation of 8 images >>> batch_size = 4 >>> context = torch.full((batch_size, 1), model.config.vocab_size - 1) # initialize with SOS token >>> context = context.to(device) >>> output = model.generate( ... input_ids=context, max_length=model.config.n_positions + 1, temperature=1.0, do_sample=True, top_k=40 ... ) >>> clusters = image_processor.clusters >>> height = image_processor.size["height"] >>> width = image_processor.size["width"] >>> samples = output[:, 1:].detach().cpu().numpy() >>> samples_img = [ ... np.reshape(np.rint(127.5 * (clusters[s] + 1.0)), [height, width, 3]).astype(np.uint8) for s in samples ... ] # convert color cluster tokens back to pixels >>> f, axes = plt.subplots(1, batch_size, dpi=300) >>> for img, ax in zip(samples_img, axes): # doctest: +IGNORE_RESULT ... ax.axis("off") ... ax.imshow(img) ```rjrkNrl) rcrrdrerrfrrrrrgrhr.r*r)losslogitsrcrrtru)rvrwrxrryr`rzr-r;rr rrrrcrrtru)rzr'rcrrdrerrfrrrrrrgrhr*transformer_outputsr lm_logitsr shift_logits shift_labelsloss_fctoutputs r'r&ImageGPTForCausalImageModeling.forward+sL V # MMr  $ u >2I%0%L!#qr'?557L')HL--b,2C2CB2GH,J[J[\^J_`D\$7$;;F)-)9TGf$ Ev E0/??-;;*550AA   r}beam_idxc.^[U4SjU55$)z This function is used to re-order the `past_key_values` cache if [`~PreTrainedModel.beam_search`] or [`~PreTrainedModel.beam_sample`] is called. This is required to match `past_key_values` with the correct beam_idx at every generation step. c3N># UHn[U4SjU55v M g7f)c3x># UH/oRSTRUR55v M1 g7f)rN) index_selectrnr)r%rors r'r(JImageGPTForCausalImageModeling._reorder_cache...s1j_iQ[))!X[[9J9J-KLL_is7:Nr)r%rrs r'r(@ImageGPTForCausalImageModeling._reorder_cache..s' -  j_ij j j-s"%r)rcrs `r'_reorder_cache-ImageGPTForCausalImageModeling._reorder_caches -   r})rNr;rMr-)NNNNNNNNNNNNNN)rrrr_tied_weights_keysrrwrrrrrXryrrrrrr staticmethodrrrrs@r'rrs++ ~ &-1@D1515/3,0048<9=)-$(,0/3&*{ ELL){ "%ell(;"<={ !. { !. { u||, { ELL){   -{  ( 5{ !) 6{ &{ D>{ $D>{ 'tn{ d^{ !{ " u77 8#{ { z  uU\\23  ?D||  uU\\" #    r}rz The ImageGPT Model transformer with an image classification head on top (linear layer). [`ImageGPTForImageClassification`] average-pools the hidden states in order to do the classification. c ^\rSrSrS\4U4Sjjr\SS\\RS\\ \ \RS\\RS\\RS\\RS \\RS \\RS \\RS \\ S \\ S\\ S\\ S\ S\ \ \44Sjj5rSrU=r$)ImageGPTForImageClassificationir`c>[TU]U5 URUl[U5Ul[ R "URURSS9UlUR5 g)NFr) rvrw num_labelsrDr-rr/r[scorerPrs r'rw'ImageGPTForImageClassification.__init__sR   ++(0YYv}}dooEJ  r}r'rcrrdrerrfrrrrgrhr*rc  SU ;a:[R"S[5 Ub [S5eU R S5nU bU OUR R n URUUUUUUUU U U U S9 nUSnURSS9nURU5nSnUGbUR RcURS:XaS UR l OoURS:aNUR[R:XdUR[R:XaS UR l OS UR l UR RS :XaJ[!5nURS:Xa&U"UR#5UR#55nOU"UU5nOUR RS :Xa=[%5nU"UR'S UR5UR'S 55nO-UR RS :Xa[)5nU"UU5nU (dU4USS-nUbU4U-$U$[+UUUR,UR.UR0S 9$)a input_ids (`torch.LongTensor` of shape `(batch_size, input_ids_length)`): `input_ids_length` = `sequence_length` if `past_key_values` is `None` else `past_key_values[0][0].shape[-2]` (`sequence_length` of input past key value states). Indices of input sequence tokens in the vocabulary. If `past_key_values` is used, only `input_ids` that do not have their past calculated should be passed as `input_ids`. Indices can be obtained using [`AutoImageProcessor`]. See [`ImageGPTImageProcessor.__call__`] for details. labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*): Labels for computing the sequence classification/regression loss. Indices should be in `[0, ..., config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If `config.num_labels > 1` a classification loss is computed (Cross-Entropy). Examples: ```python >>> from transformers import AutoImageProcessor, ImageGPTForImageClassification >>> from PIL import Image >>> import requests >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg" >>> image = Image.open(requests.get(url, stream=True).raw) >>> image_processor = AutoImageProcessor.from_pretrained("openai/imagegpt-small") >>> model = ImageGPTForImageClassification.from_pretrained("openai/imagegpt-small") >>> inputs = image_processor(images=image, return_tensors="pt") >>> outputs = model(**inputs) >>> logits = outputs.logits ```rjrkNrl) rcrrdrerrfrrrgrhrrr regressionsingle_label_classificationmulti_label_classificationr*)rrrcrrt)rvrwrxrryr`rzr-rr problem_typerrrXr}rTr rLr rr rrcrrt)rzr'rcrrdrerrfrrrrgrhr*rrpooled_hidden_statesrrrrs r'r&ImageGPTForImageClassification.forwardsGd V # MMr  $ u >2I%0%s $D>s 'tns d^s s  u66 7s s r}r)rrrDr&rn)4__doc__r7rCrvtypingrrrrrXtorch.utils.checkpointrtorch.cuda.amprtorch.nnr r r activationsr generationrmodeling_outputsrrrmodeling_utilsr pytorch_utilsrrrutilsrrrconfiguration_imagegptr get_loggerrrArnModulerprrrr&rDrr__all__r$r}r'rsP% .. #AA!) .YY 3   H %iX  X Xv"))"HBIIHV#so#s#sLF +F F R ^ %