9iRSrSSKrSSKrSSKrSSKrSSKrSSKrSSKJrJ r J r J r J r J r Jr SSKrSSKJs Jr SSKrSSKJr SSKJrJr SSKJr SSKJr SSKJrJrJrJ r SS K!J"r"J#r#J$r$ SS K%J&r& SS K'J(r(J)r)J*r* SS K+J,r, SS K-J.r.J/r/J0r0 SSK1J2r2 SSK3J4r5 SSK6J7r7 SSK8J9r9 SSK:J;r; \RxS:wal\RzR}S5 \RzRS5 \RzRS5 \RzRS5 \5R"5r4SrCSrDS/rE"SS\5rFSrG"SS\R"R5rI\RRS5rLS rM"S!S"\R"R5rNS#rO\RRS$5rPS8S%jrQ"S&S'\R"R5rR"S(S)\R"R5rS"S*S+\R"R5rT"S,S-\R"R5rU"S.S/\0\&5rVS0rWS1rX\)"S2\W5"S3S4\V55rY\.R"\7R\,RS59"S6S7\V55r]g)9zPyTorch ChatGLM model. N)AnyCallableDictListOptionalTupleUnion)nn)CrossEntropyLoss LayerNorm) skip_init)LogitsProcessor)GenerationConfigLogitsProcessorList ModelOutputStoppingCriteriaList)BaseModelOutputWithPast)BaseModelOutputWithPastAndCrossAttentionsCausalLMOutputWithPast)PreTrainedModel)add_code_sample_docstringsadd_start_docstrings%add_start_docstrings_to_model_forward)Models)MODELSModel TorchModel) OutputKeys)logger)Tasks) ChatGLMConfig)ChatGLMTokenizerdarwinFTzTHUDM/ChatGLM-6BChatGLM6BConfigzTHUDM/chatglm-6bcf\rSrSrS\R S\R S\R 4SjrSrg)InvalidScoreLogitsProcessor8 input_idsscoresreturnc[R"U5R5(d)[R"U5R5(aUR 5 SUS'U$)Ngj@).)torchisnananyisinfzero_)selfr)r*s m/var/www/html/land-doc-ocr/venv/lib/python3.13/site-packages/modelscope/models/nlp/chatglm/text_generation.py__call__$InvalidScoreLogitsProcessor.__call__:sH ;;v  " " $ $ F(;(?(?(A(A LLN F6N N) __name__ __module__ __qualname____firstlineno__r. LongTensor FloatTensorr5__static_attributes__r8r7r4r'r'8s0%"2"2**/4/@/@r7r'c SSKnSSKnSSKn[ RRU5n[RSU35 URRU5n/n/n UH]up[RSU SU 35 URRXj5n URU 5 U RU 5 M_ [X5GHupU RS5n [!SU 55(a)[RS SR#U 535 MWUn U HnUR$"S U5(aUR"S U5nOU/nUSS :Xd USS :Xa ['U S5n OZUSS:Xd USS:Xa ['U S5n O;USS:Xa ['U S5n O%USS:Xa ['U S5n O['XS5n [+U5S:dM[-US5nU Un M WSSS:Xa ['U S5n OUS :XaUR/U 5n U R0U R0:Xd"SU R0SU R0S35e[RSU 35 [6R8"U 5U lGM U$![a [R S5 ef=f![(a, [RS SR#U 535 GMf=f![2a1nU=R4U R0U R04- 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 zLoading TF weight z with shape /c3,# UH nUS;v M g7f))adam_vadam_mAdamWeightDecayOptimizerAdamWeightDecayOptimizer_1 global_stepNr8).0ns r4 0load_tf_weights_in_chatglm_6b.._s%a  sz Skipping z [A-Za-z]+_\d+z_(\d+)kernelgammaweight output_biasbetabiasoutput_weightssquad classifierr!i _embeddingszPointer shape z and array shape z mismatchedzInitialize PyTorch weight )renumpy tensorflow ImportErrorrerrorospathabspathinfotrainlist_variables load_variableappendzipsplitr0join fullmatchgetattrAttributeErrorlenint transposeshapeAssertionErrorargsr. from_numpydata)modelconfigtf_checkpoint_pathrWnptftf_path init_varsnamesarraysnamermarraypointerm_name scope_namesnumes r4load_tf_weights_in_chatglm_6brBs ggoo01G KK8 BC''0I E F  (l5'BC&&w5 T e ! 5) zz#    KK)CHHTN#34 5 F||,f55 hhy&9 %h 1~)[^w-F!'84Q=0KNf4L!'62Q#33!'84Q7*!'<8%g1~>G;1$+a.)!#,+, #$<= (gx0G x LL'E  , Y .? }KX Y,  078''. [*\ LA   Q   Z&KK)CHHTN+; <=  FFw}}ekk2 2F  s5 J% K .K? L= ,L88L=cN^\rSrSrSrU4SjrS\R4SjrSr U=r $) PrefixEncoderz The torch.nn model to encode the prefix Input shape: (batch-size, prefix-length) Output shape: (batch-size, prefix-length, 2*layers*hidden) c >[TU]5 URUlUR(a[RR UR UR5Ul[RR[RRURUR5[RR5[RRURURUR-S-55Ul g[RR UR URUR-S-5Ulg)NrU)super__init__prefix_projectionr.r Embedding pre_seq_len hidden_size embedding SequentialLinearTanh num_layerstransr3rs __class__s r4rPrefixEncoder.__init__s !'!9!9  ! !"XX//0B0B060B0BDDN,, 2 2F4F4FG  2 2 & 1 1F4F4F F JLMDJ #XX//""F$5$58J8J$JQ$NPDNr7prefixcUR(a$URU5nURU5nU$URU5nU$N)rrr)r3r prefix_tokenspast_key_valuess r4forwardPrefixEncoder.forwardsE  ! ! NN62M"jj7O#nnV4Or7)rrr) r9r:r;r<__doc__rr.Tensorrr? __classcell__rs@r4rrs# P ellr7rc^SU-S[R"SU-SSU-U---5--$)zOpenAI's gelu implementation.??gQ63E?gHm?)r.tanhxs r4 gelu_implrsF 7 ejj+a/1 q 0023 3 44r7c[U5$r)rrs r4gelurs Q<r7cb^\rSrSrS\R S4U4SjjrSrS SjrU4Sjr Sr U=r $) RotaryEmbedding'Fcl>[TU]5 SU[R"SUS5R 5U- -- nUR 5nX@lU(a,[RRU5Ul SUl O'URSU5 SUl SUl SUl X0lg)NrrrUinv_freq)rrr.arangefloathalf learnabler Parameterrmax_seq_len_cachedregister_buffer cos_cached sin_cached precision)r3dimbaserrrrs r4rRotaryEmbedding.__init__s  QQ 7 = = ?# EFG==?" !HH..x8DM&*D #  X 6&*D #"DO"DO"r7cgrr8)r3 state_dictrlocal_metadatastrict missing_keysunexpected_keys error_msgss r4_load_from_state_dict%RotaryEmbedding._load_from_state_dicts r7cjUcURUnURbX0R:Ga_UR(aSOUUl[R"X1R UR RS9n[R"SX@R 5n[R"XU4SS9RUR 5nUR[R:XaUR5nUR5SS2SSS24nUR5SS2SSS24nUR[R:Xa UR5nUR5nUR(aXx4$XxsUlUlUR SU2S4UR"SU2S44$)N)devicedtypezi,j->ijr.)rmrrr.rrrreinsumcattorbfloat16rcossinrr) r3rseq_dimseq_lentfreqsembrrs r4rRotaryEmbedding.forwardsT ?ggg&G  " " *111.2nnd'D #  0C0CEALLA}}=E))UN366qxx@C~~/iik1dA:.J1dA:.J~~/'002 '002 ~~!--/9 ,DOT_xx}-txx}/MMMr7c>URbU"UR5UlURbU"UR5Ul[TU] U5$r)rrr_apply)r3fnrs r4rRotaryEmbedding._applysF ?? & 1DO ?? & 1DOw~b!!r7)rrrrrr)r!N) r9r:r;r<r.rrrrrr?rrs@r4rrs(!&%** # N2""r7rcUSSURSS-24USURSS-S24p![R"U*U4URS- S9$)N.rrUr!r)rmr.rndim)rx1x2s r4 rotate_halfrse s%QWWR[A%%% &#qwwr{a/?/@*@(A 99 b GGaK r7c [R"XBRS55RS5[R"XCRS55RS5p2X-[ U5U--X-[ U5U--pX4$)Nr!rU)Frsqueeze unsqueezer)qkrr position_ids r4apply_rotary_pos_emb_indexrsx{{; A7AA!D KQ0::1= G A, -A0q 4Kr7c Ub5USUSp[R"X4SS9n[R"X4SS9nURuppU (aX#4nOSn[US-5nU(aU[R "U5U-- nUR S5UR S5UR S5UR S54nURUSUSUS-S5nURUSUSUS-S5n[R"SSSURURS9n[R"UURSS5URSS5RSS5SS S 9nUR"U6nUR(a2UURlURUUR55nOqUS:HR!5(dUR#US 5 URnUR5nUU-n[$R&"USS9nUR)U5nUR S5UR S5UR S5UR S54nURUR S5USUS-S5nURUSUS-USS5n[R*"UURSS55nUR"U6nUR-SSSS5R5nUR 5SS U4-nUR"U6nUUU4nU$) Nrr!rrUrrrgr)rPalpha)r.rrmrmathsqrtsizeviewzerosrrbaddbmmrlscale_mask_softmaxscale contiguousall masked_fill_rsoftmaxtypebmmpermute)r3 query_layer key_layer value_layerattention_maskhidden_size_per_partitionlayer_id layer_pastscaling_attention_score use_cachepast_key past_valuerbnhrpresentquery_key_layer_scaling_coeff output_size matmul_resultattention_scoresattention_probsr context_layernew_context_layer_shapeoutputss r4 attention_fnrsz)!}jm*IIx3; ii 9qA #,//G*$)(Q,$7!! IIk "%B BD ##A& (8(8(;##A& q(9;K"";q>#.q>KN#BBHK{1~{1~ A/N!#IKK   !! MMMa#Aq!++Aq1  M%));7 (E%112B2@2K2K2MO!#((**  ) ).( C &&+113+.KK))$4"=)..u5##A& (8(8(;##A& (8(8(;=K""[^k!nKN+J+6q>2?OIIo{/D/DQ/JKM"&& 4M"))!Q15@@BM,0023B7!;%%!&&(?@Mg7G Nr7c ^\rSrSrSS\R S4U4Sjjr\S5rSSjr SS\RS\RS \ \ \R\R4S \ S \ 4 S jjrS rU=r$) SelfAttentioniuNTcT>[[U] 5 X0lXlXlX lX lXpl[U(aURUR S--OURUR -S[RSS9Ul SUl Uc X-UlOX@lX R-Ul[![R"R$USUR-UUS9Ul[![R"R$URUUUS9Ulg)NrUrF)rrrrrQr)rrrrrrnum_attention_heads!num_attention_heads_per_partitionposition_encoding_2drr.r rotary_embrhidden_size_per_attention_headinner_hidden_sizer r rquery_key_valuedense) r3rrrrrQ params_dtyperrs r4rSelfAttention.__init__ws mT+- &)4 1D.$8!).B     % % ) +    8 8 8jj  #' ) 12=2TD /2P /!47Z7Z!Z ) HHOO  && &   HHOO  " "    r7c*URUS5 U$)Nr)r)rrs r4attention_mask_func!SelfAttention.attention_mask_funcs%%nh?r7cUR5S- nUR5UU-n[R"XUS9nU(a[ SU55$U$)zSplit a tensor along its last dimension. Arguments: tensor: input tensor. num_partitions: number of partitions to split the tensor contiguous_split_chunks: If True, make each chunk contiguous in memory. r!rc3@# UHoR5v M g7fr)r)rHchunks r4rJ.sE))++s)rrr.retuple)r3tensornum_partitionscontiguous_split_chunkslast_dim last_dim_size tensor_lists r4split_tensor_along_last_dim)SelfAttention.split_tensor_along_last_dimsS::A kk&XF "EEE Er7 hidden_statesrrroutput_attentionsc >URU5nUR5SSURSUR-4-n UR"U 6nUR US5upn UR (Ga U RSU RS- S9upU RSU RS- S9unnURXR5S-S9unnUSS2SSS24RSS5R5USS2SSS24RSS5R5nn[XUUU5up[UUUUU5unn[R"X/U RS- S9n [R"UU/URS- S9n OEURSS5nURXR5S-S9unn[XUUU5up[!UU U U UUR"UUUS 9 unnnUR%U5nUU4nU(aUU4- nU$) Y hidden_states: [seq_len, batch, hidden_size] attention_mask: [(1, 1), seq_len, seq_len] NrrrUr!r)rr) r3rrrrrrrr)rrrrrr0rr'rrmaxrlrrr.concatrrr)r3r2 position_idsrrrrr3mixed_raw_layernew_tensor_shaperrrq1q2k1k2rrblock_position_idsrr routputrs r4rSelfAttention.forwardsT ..}=+//1#26  2 2 33 3:  *..0@A88!L    $ $ $ &&q{/?/?!/C&EFB__QY^^a-?_AFBr3C3C3E3IJHC/;Aq!G/D/N/NqRS/T/_/_/aQ1W%//15@@B-L/S,OFB/BS0BDFB,,xbggkCK b"XBGGaKAI'11!Q7L%5%5%7!%;'=HC&@S,&@ "K3?##)&*&D&D! 3!/ wM*7#  * *Gr7) rrrrrrrrrrrrFNFF)r9r:r;r<r.rr staticmethodr#r0rrrboolrr?rrs@r4rrus 15#kk&*1 f  =B8CG"'D||D D U5<<#=>? DD DDr7rc.^\rSrSrU4SjrSrSrU=r$)GEGLUi cL>[TU]5 [RUlgr)rrrr activation_fn)r3rs r4rGEGLU.__init__s VVr7cfURSURS- S9up#X RU5-$)NrUr!r)r'rrI)r3rrrs r4r GEGLU.forwards2!-&&r***r7)rI)r9r:r;r<rrr?rrs@r4rGrG s$++r7rGcP^\rSrSrSSS\\R 4U4SjjrSrSr U=r $)GLUiNTcb>[[U] 5 X0lXPlXlUcSU-nX l[[RRUR UR UUS9Ul [[RRUR UR UUS9Ul g)Nr) rrNrractivation_funcrrr r.r r dense_h_to_4h dense_4h_to_h)r3rrrrQrQr rs r4r GLU.__init__s c4!# .'  $ !K !2& HHOO     " "  ' HHOO  " "     r7clURU5nURU5nURU5nU$)z. hidden_states: [seq_len, batch, hidden_size] )rRrQrS)r3r2intermediate_parallelr@s r4r GLU.forward:s= !% 2 2= A $ 4 45J K##$9: r7)rQrSrRrrr) r9r:r;r<rr.rrrr?rrs@r4rNrNs($(!%#kk  @  r7rNc ^\rSrSrSS\S\R SS4U4SjjrS S\RS\RS\ \ \R\R4S \ S \ 4 S jjr S r U=r$)GLMBlockiINTc >[[U] 5 X@lU"XS9UlXl[ UUUUUU UR S9UlU"XS9UlXl [UUUUU S9Ul g)Neps)rrQr r)rrQrr ) rrYrrinput_layernormrr attentionpost_attention_layernormrrNmlp) r3rrlayernorm_epsilonrrr layernormuse_biasr rrrs r4rGLMBlock.__init__Ks h&(!  )L$8!'   +I%!%!:!:<)2 )0%% /%  r7r2rrrr3c URU5nURUUUUUUUS9n U Sn U SSn SUR-S-n X-U -nURU5n UR U 5nX-U-nU(aU4U -n U $U4U SS-n U $)r5)rrrrr3rr!NrUr)r^r_rr`ra)r3r2r8rrrrr3attention_inputattention_outputsattention_outputrr mlp_input mlp_outputr@s r4rGLMBlock.forwardzs"..}=!NN  )!/+1-Q/#AB'T__$s*'/2BB 11-@ XXi( "Z/ j7*Gj712;.Gr7)r_r^rrarrr`rC)r9r:r;r<r r.rrrrrrErr?rrs@r4rYrYIs$(04$#kk&*- jCG"'2||2 2 U5<<#=>? 22 22r7rYc^\rSrSrSrSrSr\rSr S/r U4Sjr S\ R4S jrS rSS jrSS jr\U4S j5rSrU=r$)ChatGLMPreTrainedModelizz An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained models. FT transformerrYc b>[TU]"UR40UD6 [[U]U5 gr)rr name_or_pathr)r3rskwargsrs r4rChatGLMPreTrainedModel.__init__s* ,,77 eT#F+r7modulecg)zInitialize the weights.Nr8)r3rts r4 _init_weights$ChatGLMPreTrainedModel._init_weightssr7cURup4UVs/sH5oUR5RURR5PM7 nn[ R "X4U4US9nUR5 [U5HupSXxSS2SU 24'M URS5 US:R5nU$s snf)Nrr!r) rmtolistindexrs bos_token_idr.onestril_ enumerate unsqueeze_rE) r3r)r batch_size seq_lengthseqcontext_lengthsricontext_lengths r4 get_masks ChatGLMPreTrainedModel.get_maskss!* DM DMSJJL  t{{77 8I  ZZ$H+13!*?!; A45Na.0 1"<!!!$(3.446 sURSS5nURSS5 [[U]"SSU0UD6nX#lU$)zInstantiate the model. Args: kwargs: Input args. model_dir: The model dir used to load the checkpoint and the label information. Returns: The loaded model, which is initialized by transformers.PreTrainedModel.from_pretrained model_dirNcfgpretrained_model_name_or_pathr8)poprrfrom_pretrainedr)clsrrrrrrs r4 _instantiate#ChatGLMPreTrainedModel._instantiatesNJJ{D1  5$eS1?*3?7=?# r7r8rB)r9r:r;r<ris_parallelizablesupports_gradient_checkpointingr" config_classbase_model_prefix_no_split_modulesrr Modulervrrr classmethodrr?rrs@r4rnrnsb &*# L%# ,BII "H2r7rnaM This model is a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) sub-class. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior. Parameters: config ([`~ChatGLM6BConfig`]): Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the [`~PreTrainedModel.from_pretrained`] method to load the model weights. a: Args: input_ids (`torch.LongTensor` of shape `({0})`): Indices of input sequence tokens in the vocabulary. Indices can be obtained using [`ChatGLM6BTokenizer`]. See [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. [What are input IDs?](../glossary#input-ids) attention_mask (`torch.FloatTensor` of shape `({0})`, *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) token_type_ids (`torch.LongTensor` of shape `({0})`, *optional*): Segment token indices to indicate first and second portions of the inputs. Indices are selected in `[0, 1]`: - 0 corresponds to a *sentence A* token, - 1 corresponds to a *sentence B* token. [What are token type IDs?](../glossary#token-type-ids) position_ids (`torch.LongTensor` of shape `({0})`, *optional*): Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0, config.max_position_embeddings - 1]`. [What are position IDs?](../glossary#position-ids) head_mask (`torch.FloatTensor` of shape `(num_heads,)` or `(num_layers, num_heads)`, *optional*): Mask to nullify selected heads of the self-attention modules. Mask values selected in `[0, 1]`: - 1 indicates the head is **not masked**, - 0 indicates the head is **masked**. inputs_embeds (`torch.FloatTensor` of shape `({0}, hidden_size)`, *optional*): 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. 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. zdThe bare ChatGLM-6B Model transformer outputting raw hidden-states without any specific head on top.c^\rSrSrSrS\4U4SjjrSrS\R4Sjr \R4Sjr \ "\RS 55\"\\\S 9SS \\R*S \\R*S \\RS\\\\R\R4S4S\\R*S\\S\\S\\S\\S\\\RS4\44Sjj55rSrU=r$)riKa The model can behave as an encoder (with only self-attention) as well as a decoder, in which case a layer of cross-attention is added between the self-attention layers, following the architecture described in [Attention is all you need](https://arxiv.org/abs/1706.03762) by Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser and Illia Polosukhin. To behave as an decoder the model needs to be initialized with the `is_decoder` argument of the configuration set to `True`. To be used in a Seq2Seq model, the model needs to initialized with both `is_decoder` argument and `add_cross_attention` set to `True`; an `encoder_hidden_states` is then expected as an input to the forward pass. rsc>^[TT]U5 URTlURTl[R TlURTlURTlURTl URTl URTl TRTR-Tl URTl URTlURTl[![R"R$TRTRTR S9TlSTlU4Sjn[R"R+[-TR5Vs/sH o2"U5PM sn5Tl[1TRTRS9TlTRbTR55H nSUlM [R8"TR5R;5Tl[?U5Tl [R"RCS5Tl"ggs snf)N)num_embeddings embedding_dimrFc >[TRTRTRUTRTR [ STRTRS9 $)NT)rrrcrdr r) rYrrrbrrr r r)rr3s r4 get_layer(ChatGLMModel.__init__..get_layerwsX  ((&&"&"8"8/3..#!..%)%>%>  r7r\g?)#rrmax_sequence_lengthrr.rr r vocab_sizerrbrrrrrr r rword_embeddingsr ModuleListrangelayersr final_layernorm parameters requires_gradrrrrprefix_encoderDropoutdropout)r3rsrrparamrs` r4rChatGLMModel.__init___s  $*#=#= !--!JJ#)#=#=  ++ ++!'!9!9!'!9!9.2.>.>$BZBZ.Z+$*$?$?!!--!'!9!9( HH  ??**##  % ',# hh))16t1G H1GXYx 1G HJ  )   $"8"8 :    '*&+#+!&d.>.>!?!D!D!FD "/"7D  88++C0DL ( Is3IcUR$rrr3s r4get_input_embeddings!ChatGLMModel.get_input_embeddingss###r7new_embeddingscXlgrrr3rs r4set_input_embeddings!ChatGLMModel.set_input_embeddingss-r7cURRS5RUS5RU5nUR U5R U5nUR XRURS-URURUR-5nURU5nUR/SQ5RS5nU$)NrrrU)rUr!rrrP)rrexpandrrrrrrrrrrre)r3rrrrrs r4 get_promptChatGLMModel.get_prompts**44Q7>>z?ACCE2f: --m<AA%H).. (($//A*=  $ $    8 8 8: ,,7)11/BHHKr7zbatch_size, sequence_length) checkpoint output_typerr)r8rr. inputs_embedsrr3output_hidden_states return_dictr+c nUbUOURRnUbUOURRnUbUOURRnU bU OURRn UR (aUR (a U(aSnUbUb [S5eUbURSSupO!UbURSSupn O [S5eUcURU5nUGc URb3URURSURURS9nO"[S/[UR 5-5nUcUR#XRS9nUcURR$URR&pX;aUOU nX;aSOSnUVs/sH"nUR)5R+U5PM$ nnUR-UUURUS 9nURbxUbu[.R0"U S UR3S 5UR5R5UR5nUS :R75n[.R8"UU4S S9nUR;SS 5nU(aSOSnU(aSOSnU(aSOSnUc/[.R<"S S URS9R75nOUR5UR5n[?UR 5HunnU(aUU4-nUUnUR (aTUR (aC[.R@RBRCUUX#[.RD"U5UXg5nO U"UUU[.RD"U5UUUS9nUSnU(a UUS 4-nU(dMUUU(aSOS 4-nM URGU5nU(aUU4-nU (d[SUUUU455$[IUUUUS9$s snf)NFzDYou cannot specify both input_ids and inputs_embeds at the same timerUz5You have to specify either input_ids or inputs_embedsr)rrrryT)rrrr!rrrrr8)r8rrrrr3c30# UH nUcMUv M g7frr8)rHvs r4rJ'ChatGLMModel.forward..2s  %q%s )last_hidden_staterr2 attentions)%rsr3rruse_return_dictrtraining ValueErrorrmrrrrrr)rjrr mask_token_idgmask_token_idrzr{rr.r}rrrErrlrrutilsrr*rr)r3r)r8rrrrr3rrrr_MASKgMASK mask_token use_gmaskrrprefix_attention_maskr2presentsall_self_attentionsall_hidden_statesrlayerr layer_rets r4rChatGLMModel.forwardsL*2C1N-TXT_T_TqTq$8$D KK , , "+!6IDKKV  "%.__Ra%8 "J  &(5(;(;BQ(? %JAGI I   00;M  "+"&//(q1$++'--#2#/ #(T[[1A(A"B%!%&6&6"0"8#"kk779S9Se&+&8Ud $)$6DE ?H">GsCJJL&&z2i" $44#1$++# 5 %    'N,F$)JJAy~~b1  %""$"^%:%:"; "&;S%@$F$F$H !"YY(=~'N+,.N&//15 "2$5b4"6BD  !"[[Ai6F6FGLLNN,..y/?/?@N!$++.HAu#$58I$I!(+J**t}}!KK22===,LLOZO "!!-#1"\\!_)'&79 &aLM#y|&66  &99a!4=8'8#5/<,,];  1]4E E  x):H $D>H 'tnH d^H  uU\\3&')@@ AH KH r7r) module_namecF^\rSrSrS\4U4SjjrSrSrS,S\S\ \ \ 4S\ S \ S \ \ \ 44 S jjr S-S \RS \\R"S\\R"S\\R"S\\R"S \4 SjjrS.S \\R"S\\R"S\\R"S\\\R*S\\R"S\\R"S\\ S\\ S\\ S\\ 4Sjjr\S \\\R"\R"4S4S\RS \\\R"\R"4S44Sj5rSr\R4"5S/S\ S\\\ \ 4S\4S jj5r\R4"5S0S\ S\\\ \ 4S\4S!jj5r\R4"5S-S"\\S#\\ S$\\!S%\\"\\R"/\\44S&jj5r#S1S'\4S(jjr$S)\ S \ 4S*jr%S+r&U=r'$)2ChatGLMForConditionalGenerationi>rsc>[TU]U5 URUlURUl[ U5Ul[ [RURURS[RS9Ul XlSUlURR (a$UR#URR SS9 [$R&"UR(5Ulg)NFrT) empty_init)rrrrrror r rrrr.rlm_headrs quantizedquantization_bitquantizer#rrq tokenizerrs r4r(ChatGLMForConditionalGeneration.__init__As  $*#=#= $*$?$?!'/ II      **    ;; ' ' MM$++664M H*99&:M:MNr7cUR$rrrs r4get_output_embeddings5ChatGLMForConditionalGeneration.get_output_embeddings^s ||r7cXlgrrrs r4set_output_embeddings5ChatGLMForConditionalGeneration.set_output_embeddingsas% r7r model_kwargsis_encoder_decoderstandardize_cache_formatr+c URXS9US'SU;aUSnUbUR[R:Xar[R"UUR /UR SSQSP75/SS9nUSS2SS2SS24R5nSUS '[R"XV/S S9US'S U;aIUS nUS SS24R5nUSS2SSS24==S- ss'[R"Xx/SS9US 'U$) N)rrrrr!rrF).rrUr8.)_extract_past_from_model_outputrr.rErnew_onesrmclone) r3rrrrrnew_attention_maskr8new_position_ids r4#_update_model_kwargs_for_generationCChatGLMForConditionalGeneration._update_model_kwargs_for_generationds=+/*N*N +O+H &' | +)*:;N)n.B.Bejj.P!&""++,Jn.B.B2A.F,J,JK,01 "2 &4Aq"#I%>%D%D%F".3"7+16#8a2A -. \ )'7L*384::A+_?>??>:]N(n)DE?>?$jj(// $1 2;2 !$)<<I  Irlabelsrr3rrc ^UbUOURRnU bU OURRn URUUUUUUUU U S9 n U Sn UR U 5R SSS5R 5n SnUbU R[R5n U SSS2SS24R 5nUSSS24R 5n[SS9nURUR5nU"URSURS55URS55nU RU R5n URU R5nU (dU 4U SS-nUbU4U-$U$[UU U R U R"U R$S 9$) N) r)r8rrrrr3rrrr!rU.ri) ignore_index)losslogitsrr2r)rsrrrorrrrr.float32r rrrrrrr2r)r3r)r8rrrrrr3rrtransformer_outputsr2 lm_logitsr shift_logits shift_labelsloss_fctr@s r4r'ChatGLMForConditionalGeneration.forwards"+!6IDKKL!#qr'?557L'T:H'??<+>+>?L!!"l&7&7&;<!!"%'D" ]%8%89I77=../D]%8%<# UHgnUSRSTRUSR55USRSTRUSR554v Mi g7f)rr!N) index_selectrr)rHrrs r4rJAChatGLMForConditionalGeneration._reorder_cache..sl"!j qM & &q(++jm6J6J*K L qM & &q(++jm6J6J*K L !sA/A2)r))r rs `r4_reorder_cache.ChatGLMForConditionalGeneration._reorder_cache s!"!"" "r7cUR5nURSS5nSS/SS/SS/S S /S S //nUHKn[R"S US-SUS-U5n[R"SUS-SUS-U5nMM U$)Nu[[训练时间]]u2023年,u，!u！:u：;u；z\?u？z([\u4e00-\u9fff])%srz\1%sr!z%s([\u4e00-\u9fff])z%s\1)stripreplacerWsub)r3responsepunktsitems r4process_response0ChatGLMForConditionalGeneration.process_responses>>###$6 B %L %L %L %L 5M  Dvv4tAw>%Q/;Hvv4tAw>%Q/;H r7queryhistory max_lengthc  *Uc/nU c [5n U R[55 UUUUUU S.U En U(dUn OLSn [U5Hun upU SR XU5- n M U SR [ U5U5- n U"U /SS9nUR UR5nUR"S 0UDU D6nUR5S[ USS5SnURU5nURU5nX2U4/-nX4$) N)r2 num_beams do_sampletop_p temperaturelogits_processor[Round {}] 问：{} 答：{} [Round {}] 问：{} 答：ptreturn_tensorsrr)r8) rrcr'rrrjrrgeneraterzdecoder.)r3rr0r1r2r4r5r6r7r8rr gen_kwargspromptr old_queryr+inputsrs r4_chat%ChatGLMForConditionalGeneration._chat,sJ ?G  #24  ; =>$""& 0   FF,5g,>((I<CC(,,-? 4;;CL%P PFF8D94;;'--7&7J7.."1%c&*=a*@&A&BC##G,((2X.//  r7c +F# Uc/nUc [5nUR[55 UUUUUS.U En U(dUn OLSn [U5Hun upU SR XU5- n M U SR [ U5U5- n U"U /SS9nUR UR5nUR"S 0UDU D6HWnUR5S[ USS5SnURU5nURU5nX2U4/-nUU4v MY g7f) N)r2r5r6r7r8r9r:r;r<r=rr)r8) rrcr'rrrjrrstream_generaterzr@r.)r3rr0r1r2r5r6r7r8rrrArBrrCr+rDr new_historys r4 stream_chat+ChatGLMForConditionalGeneration.stream_chatWsT ?G  #24  ; =>$"& 0    FF,5g,>((I<CC(,,-? 4;;CL%P PFF8D94;;'++CfC CGnn&q)#f[.A!.D*E*FGG ''0H,,X6H!X%6$77KK' ' DsDD!generation_configr8stopping_criteriaprefix_allowed_tokens_fnc +^# URSURSpUc URn[R"U5nUR"S0UD6n UR UR p[U [5(aU /n URS5SL=(a URSLn U (a7URc*[R"SURS3[5 O]URbPURU-Ul U (d5[RSURSURS3[5 XR:aHUR R"(aS OS n [R%S U S US URS35 UbUO ['5nUbUO [)5nUR+UUUUUS9nUR-UUS9nUR/U5n UR1URS5R3S5nSnUR4"U40U D6nU"S0UDSSSS.D6nUR6SS2SSS24nU"UU5nU "UU5n[8R:R=USS9nUR>(a%[@RB"USS9RES5mO[@RF"USS9m[@RH"UTSS2S4/SS9nURKUU UR R"S9n URM[OU4SjU 55RQ55nURS5S:Xd U"X5(agUv GMP7f)Nrrr2zUsing `max_length`'s default (z) to control the generation length. This behaviour is deprecated and will be removed from the config in v5 of Transformers -- we recommend using `max_new_tokens` to control the maximum length of the generation.zBoth `max_new_tokens` (=z) and `max_length`(=z) seem to have been set. `max_new_tokens` will take precedence. Please refer to the documentation for more information. (https://huggingface.co/docs/transformers/main/en/main_classes/text_generation)decoder_input_idsr)zInput length of z is z, but `max_length` is set to zX. This can lead to unexpected behavior. You should consider increasing `max_new_tokens`.)rLinput_ids_seq_lengthencoder_input_idsrNr8)rLrMr!TF)rr3rr) num_samples)rc3.># UH nTU:gv M g7frr8)rHr next_tokenss r4rJBChatGLMForConditionalGeneration.stream_generate..s<|![A%|sr8)*rmrLcopydeepcopyupdater| eos_token_idrrkgetr2max_new_tokenswarningswarn UserWarningrrsrwarningrr_get_logits_processor_get_stopping_criteria_get_logits_warpernewfill_rrr functionalrr5r. multinomialrargmaxrrmulsumrr6)r3r)rLr8rMrNrrrrQrrZhas_default_max_lengthinput_ids_string logits_warperunfinished_sequencesr* model_inputsrnext_token_logitsnext_token_scoresprobsrUs @r4rH/ChatGLMForConditionalGeneration.stream_generates#,//!"4ioo 7   $ $ 6 6  MM*;<(//9&9 +88:K:X:X< lC ( ((>L!' "!"""O&7&B&B$&N  !&7&F&F&N MM01B1M1M0NOee    - - 9+<+K+KNb+b  () ./@/O/O.PPd(3345ff  #?#? ?6:kk6T6T2Ze  NN"#3"4D9M8NO%001200 1 0@/K+QdR 1B1N-ThU  55/!5'%=- 6 !77//81//0AB (}}Y__Q-?@FFqI==+)+L "'%* G!(q"ax 8 !1> KK, -K'+ $#   dE/9E=CE r7inputc USnUSnSU;aUSnOSnSU;aUSnOSnSU;aUSnOSnS U;aUS nOS n[U5[R:XaUR5nUR UR UUUUUUS 9up[ RS 5 [RU[RU0$) Ntextr1r2r7ffffff?r4r!r5T)r2r7r4r5zGeneration finished.) rr.rrzrErrr_rRESPONSEHISTORY) r3rxrzr1r2r7r4r5r+s r4chat$ChatGLMForConditionalGeneration.chatsV} " 5 |,JJ E ! .KK % k*II % k*II =ELL (nn&G JJ NN  !#'!  *+##Xz/A/A7KKr7)rsrrrrrro)FF)NNNN) NNNNNNNNNN)Nr{r!Tffffff?r|N)Nr{Trr|NrB)(r9r:r;r<r"rrrrrstrrrErr.r=rrdictrrr>rrDr!r.no_gradrrkrErJrrrrrHrrr?rrs@r4rr>sO}O:&$)). !!38n!! ! #' ! c3h !L,06:5937 D''D5<<(D&ell3 D %U\\2 D #5<<0 DDP-1/315>B04)-$(,0/3&*; ELL); u||,; !. ; "%(9(9":; ;   - ; &; D>; $D>; 'tn; d^; z"E%,, 45"$)$4$4" uU\\5<</0#5 6"" " ]]_04 $#(!(!E#s(O,(! (!(!T ]]_6:&*" $%)&(&("%S/2&(!$ &(&(P ]]_9=:><@BFl$$45l##67 l $$89 l #+8S%,,4G48I5>,?#@ ll\S$#L$#L4#L#Lr7r)NTF)^rrWrr\rWsysr]typingrrrrrrr r.torch.nn.functionalr rfrtorch.utils.checkpointtorch.nnr r torch.nn.utilsr &transformers.generation.logits_processrtransformers.generation.utilsrrrrtransformers.modeling_outputsrrrtransformers.modeling_utilsrtransformers.utilsrrrmodelscope.metainformodelscope.modelsrrrmodelscope.outputsrmodelscope.utilsrloggingmodelscope.utils.constantr configurationr" tokenizationr#platform_C_jit_set_profiling_mode_jit_set_profiling_executor_jit_override_can_fuse_on_cpu_jit_override_can_fuse_on_gpu get_loggerrr(CHATGLM_6B_PRETRAINED_MODEL_ARCHIVE_LISTr'rrrjitscriptrrrrrrrrGrNrYrnCHATGLM_6B_START_DOCSTRINGrrregister_moduler chatglm_6brr8r7r4rsk DDD 0$BAA8GG'77).+(*<<8 HH$$U+ HH((/ HH**40 HH**40    (#,( /GTEHHOO@443"ehhoo3"l" slTEHHOOTn +EHHOO +.%((//.bcuxxcL\Z\~ /djl )l l ^ 0A0ABeL&<eLCeLr7