T^(hddlmZddlmZmZmZmZddlZddlmZddl m Z ddl m Z m Z mZmZddlmZdd lmZdd lmZdd lmZmZmZmZmZdd lmZmZdd lm Z m!Z!ddl"m#Z#ddl$m%Z%m&Z&m'Z'm(Z(m)Z)m*Z*m+Z+ddl,m-Z-ddl.m/Z/e*j`e1Z2dZ3dZ4GddejjZ6GddejjZ7dZ8dBdZ9dejtde;dejtfdZ< dCdejjd ejtd!ejtd"ejtd#eejtd$e=d%e=fd&Z>Gd'd(ejjZ?Gd)d*ejjZ@Gd+d,ejjZAd-ZBe'd.eBGd/d0e!ZCd1ZDe'd.eBGd2d3eCZEGd4d5ee%ZFGd6d7eCeZGe'd8eBGd9d:eCZHe'd;eBGd<d=eCZIe'd>eBGd?d@eCZJgdAZKy)D)partial)CallableOptionalTupleUnionN)nn)ACT2FN)Cache DynamicCacheSlidingWindowCache StaticCache)GenerationMixin)AttentionMaskConverter)FlashAttentionKwargs)BaseModelOutputWithPastCausalLMOutputWithPastQuestionAnsweringModelOutput SequenceClassifierOutputWithPastTokenClassifierOutput)ROPE_INIT_FUNCTIONSdynamic_rope_update)ALL_ATTENTION_FUNCTIONSPreTrainedModel)Unpack) LossKwargsadd_code_sample_docstringsadd_start_docstrings%add_start_docstrings_to_model_forwardcan_return_tupleloggingreplace_return_docstrings)deprecate_kwarg) Qwen3Configz Qwen/Qwen3-8Br%c,eZdZdfd ZdZdZxZS) Qwen3RMSNormct|tjt j ||_||_y)z; Qwen3RMSNorm is equivalent to T5LayerNorm N)super__init__r Parametertorchonesweightvariance_epsilon)self hidden_sizeeps __class__s f/var/www/skyplay_api_hub/venv/lib/python3.12/site-packages/transformers/models/qwen3/modeling_qwen3.pyr*zQwen3RMSNorm.__init__?s1 ll5::k#:; #c"|j}|jtj}|j dj dd}|tj ||jzz}|j|j|zS)NT)keepdim) dtypetor,float32powmeanrsqrtr/r.)r0 hidden_states input_dtypevariances r4forwardzQwen3RMSNorm.forwardGsy#)) %((7  $$Q',,R,>% Ht?T?T4T(UU {{]--k:::r5c^t|jjd|jS)Nz, eps=)tupler.shaper/r0s r4 extra_reprzQwen3RMSNorm.extra_reprNs* ))*+6$2G2G1HIIr5)gư>)__name__ __module__ __qualname__r*rCrH __classcell__r3s@r4r'r'>s$;Jr5r'c$eZdZfdZdZxZS)Qwen3MLPct|||_|j|_|j|_t j |j|jd|_t j |j|jd|_t j |j|jd|_ t|j|_ yNFbias) r)r*configr1intermediate_sizerLinear gate_projup_proj down_projr hidden_actact_fnr0rTr3s r4r*zQwen3MLP.__init__Ss  !--!'!9!94#3#3T5K5KRWXyy!1!143I3IPUV 4#9#94;K;KRWXV../ r5c|j|j|j||j|z}|SN)rYr[rWrX)r0xrYs r4rCzQwen3MLP.forward]s6NN4;;t~~a/@#ADLLQRO#ST r5)rIrJrKr*rCrLrMs@r4rOrORs 0r5rOc|dd|jddzf}|d|jddzdf}tj| |fdS)z*Rotates half the hidden dims of the input..Nr8r7dim)rFr,cat)r_x1x2s r4 rotate_halfrfbsZ 3"!''"+"" " #B 3 q " " #B 99rc2YB ''r5c|j|}|j|}||zt||zz}||zt||zz}||fS)aApplies Rotary Position Embedding to the query and key tensors. Args: q (`torch.Tensor`): The query tensor. k (`torch.Tensor`): The key tensor. cos (`torch.Tensor`): The cosine part of the rotary embedding. sin (`torch.Tensor`): The sine part of the rotary embedding. position_ids (`torch.Tensor`, *optional*): Deprecated and unused. unsqueeze_dim (`int`, *optional*, defaults to 1): The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note that cos[position_ids] and sin[position_ids] have the shape [batch_size, seq_len, head_dim]. Then, if q and k have the shape [batch_size, heads, seq_len, head_dim], then setting unsqueeze_dim=1 makes cos[position_ids] and sin[position_ids] broadcastable to the shapes of q and k. Similarly, if q and k have the shape [batch_size, seq_len, heads, head_dim], then set unsqueeze_dim=2. Returns: `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding. ) unsqueezerf)qkcossin position_ids unsqueeze_dimq_embedk_embeds r4apply_rotary_pos_embrqisY( -- &C -- &C3w;q>C/0G3w;q>C/0G G r5r@n_repreturnc|j\}}}}|dk(r|S|dddddddddfj|||||}|j|||z||S)z This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch, num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim) r$N)rFexpandreshape)r@rrbatchnum_key_value_headsslenhead_dims r4 repeat_kvr{so 2?1D1D.E h z!!Qa"23::5BUW\^bdlmM  (;e(CT8 TTr5modulequerykeyvalueattention_maskscalingdropoutc Tt||j}t||j} tj||j dd|z} |#|ddddddd|j df} | | z} t jj| dtjj|j} t jj| ||j} tj| | } | j ddj} | | fS)Nr7r r8)rbr:)ptrainingr$)r{num_key_value_groupsr,matmul transposerFr functionalsoftmaxr<r;r:rr contiguous) r|r}r~rrrrkwargs key_states value_states attn_weights causal_mask attn_outputs r4eager_attention_forwardrs3 ; ;$>&B\B\$\!}}d* !'!9!9ii    : :T]] JQWQfQf  ii    : :T]] JQWQfQf  ii    : :T]] JQWQfQf  ii  & & 68J8JQWQfQf  #4==f6I6IJ "4==f6I6IJ $33 KK * * %5t<H$++"?"??"&D @r5r@position_embeddingsrpast_key_valuecache_positionrrsc |jdd}g|d|j}|j|j|j |j dd} |j |j|j |j dd} |j|j |j dd} |\} } t| | | | \} } |'| | |d}|j| | |j|\} } t}|jjdk7r^|jjdk(r(|jddrt j#d nt$|jj}||| | | |f|j&sd n |j(|j*|j,d |\}}|j.g|dj1}|j3|}||fS) Nr8r$r7)rlrkreagersdpaoutput_attentionsFz`torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to eager attention. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.)rrr)rFrzrrviewrrrrrqupdaterrrT_attn_implementationgetlogger warning_oncerrrrrrvrr)r0r@rrrrr input_shape hidden_shape query_statesrrrkrl cache_kwargsattention_interfacerrs r4rCzQwen3Attention.forwards$))#2. 88b8$--8 {{4;;}#=#B#B<#PQ[[\]_`a [[]!;!@!@!NOYYZ[]^_ {{=166|DNNqRST &S#7 jRUWZ#[ j  %#&snUL'5'<'d##L '>dkk>^>^&_#$7      % $}}C$2H2HLL.. %  % ! \*k));;;;FFH kk+. L((r5)NN)rIrJrK__doc__r%intr*r,Tensorrrr LongTensorrrrCrLrMs@r4rrsG'{'s'J+/59 0)||0)#5<<#=>0)!. 0) ! 0) !!1!12 0)-.0) u||Xell3XeELL>Q5RR S0)r5rcpeZdZdedeffd Z ddejdeejdeejdee dee d ee d eejd ee ejejfd e ed e ejee ejejfffdZxZS)Qwen3DecoderLayerrTrct||j|_t|||_t ||_t|j|j|_ t|j|j|_ |jr4|jdk7r$tjd|jdyyy)N)rTrrflash_attention_2z=Sliding Window Attention is enabled but not implemented for `z)`; unexpected results may be encountered.)r)r*r1r self_attnrOmlpr'rinput_layernormpost_attention_layernormrrrrrs r4r*zQwen3DecoderLayer.__init__s !--'vKF#+F,>,>FDWDWX(4V5G5GVM`M`(a%  ! !f&A&AEX&X   OPVPkPkOlm99 'Y !r5r@rrmrr use_cacherrrrsc |} |j|}|jd||||||||d| \}} | |z}|} |j|}|j|}| |z}|f} |r| | fz } | S)N)r@rrmrrrrr)rrrr) r0r@rrmrrrrrrresidualself_attn_weightsoutputss r4rCzQwen3DecoderLayer.forwards!,,]; ,:4>> , ')%)/) 3 ,  , ( (!=0 !55mD /  =0  "  )+ +Gr5)NNNFFNN)rIrJrKr%rr*r,rrrr boolrrr FloatTensorrCrLrMs@r4rrs  { s $2637*.,1$)59KO(||(!.(u//0 ( ! ( $D> (D>(!!1!12(&eELL%,,,F&GH(-.( u  (51B1BEDUDU1U+V"WW X(r5rc^eZdZddeffd Zej edZxZ S)Qwen3RotaryEmbeddingrTct|t|drG|j;|jj d|jj d|_nd|_|j |_|j |_||_ t|j |_ |j|j|\}|_ |jd|d|j|_y)N rope_scaling rope_typetypedefaultinv_freqF) persistent)r)r*hasattrrrrmax_position_embeddingsmax_seq_len_cachedoriginal_max_seq_lenrTr rope_init_fnattention_scalingregister_bufferroriginal_inv_freq)r0rTdevicerr3s r4r*zQwen3RotaryEmbedding.__init__<s  6> *v/B/B/N#0044[&BUBUBYBYZ`BabDN&DN"("@"@$*$B$B! /?+/+<+>@GG HZHZ[\H]_acdehhijiqiqr ,QaZ 8 > > @'1!((--'E!((--[`J`ahhmmfk ^^ U C 5&,,.1F1L1L1NNYYZ[]^_E))UEN3C'')d444C'')d444C  5 vvAGGv$cff177f&;;;  5 5s BF%%F.r^) rIrJrKr%r*r,no_gradrrCrLrMs@r4rr;s3/{/"U]]_ < std) rTinitializer_rangerrrVr.datanormal_rSzero_ Embedding padding_idx)r0r|rs r4 _init_weightsz"Qwen3PreTrainedModel._init_weightsskk++ fbii ( MM   & &CS & 9{{&   &&('  - MM   & &CS & 9!!- ""6#5#56<<>..r5N)rIrJrKr% config_classbase_model_prefixsupports_gradient_checkpointing_no_split_modules_skip_keys_device_placement_supports_flash_attn_2_supports_sdpa_supports_flex_attn_supports_cache_class_supports_quantized_cache_supports_static_cache_supports_attention_backendrrr5r4rrnsU L&*#,-#4"5!N  $!"& ?r5ra$ Args: input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`): Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide it. Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. [What are input IDs?](../glossary#input-ids) 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) Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. If `past_key_values` is used, optionally only the last `input_ids` have to be input (see `past_key_values`). If you want to change padding behavior, you should read [`modeling_opt._prepare_decoder_attention_mask`] and modify to your needs. See diagram 1 in [the paper](https://arxiv.org/abs/1910.13461) for more information on the default strategy. - 1 indicates the head is **not masked**, - 0 indicates the head is **masked**. position_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*): Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0, config.n_positions - 1]`. [What are position IDs?](../glossary#position-ids) past_key_values (`Cache`, *optional*): Pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention blocks) that can be used to speed up sequential decoding. This typically consists in the `past_key_values` returned by the model at a previous stage of decoding, when `use_cache=True` or `config.use_cache=True`. It is a [`~cache_utils.Cache`] instance. For more details, see our [kv cache guide](https://huggingface.co/docs/transformers/en/kv_cache). If `past_key_values` are used, the user can optionally input only the last `input_ids` (those that don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of all `input_ids` of shape `(batch_size, sequence_length)`. inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, 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. use_cache (`bool`, *optional*): If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding (see `past_key_values`). 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. cache_position (`torch.LongTensor` of shape `(sequence_length)`, *optional*): Indices depicting the position of the input sequence tokens in the sequence. Contrarily to `position_ids`, this tensor is not affected by padding. It is used to update the cache in the correct position and to infer the complete sequence length. ceZdZdZdeffd ZdZdZee e  dde e jde e jde e jd e ed e e j d e ed e ed e ede e jdeedefdZ dde jde jde jd ed ef dZede jdedede j2de j4de jdeded efdZxZS) Qwen3Modelz Transformer decoder consisting of *config.num_hidden_layers* layers. Each layer is a [`Qwen3DecoderLayer`] Args: config: Qwen3Config rTc  t|||j|_|j|_t j |j|j|j|_t jt|jDcgc]}t||c}|_ t|j|j|_t#||_d|_|j)ycc}w)Nr)rTF)r)r* pad_token_idr vocab_sizerrr1 embed_tokens ModuleListrangenum_hidden_layersrlayersr'rnormr rotary_embgradient_checkpointing post_initrs r4r*zQwen3Model.__init__s  !.. ++LL):):F?ab b   --i8M 0*nO  !CRC^==?de "\\ "2]5H5H5K"KTaThThN  )33A6L.. M>?L] & #oom\J#7BD0d![[)H4;;+H+HI 6M#!m%55!**t}} $ A AM22H6GH! #%"' ! !.! !#.!-#2&7'#1(; !( ! *!,M =#3"55A 6D -0  -!1 1 &+/8Od+%   r5 input_tensorc |jjdk(rS|H|F|dddfjj|j dk7}|r t d|d|vr|Sy||j nd}t|t}t|t} |jjdk(r?|s=| s;|s9tj||||jj|jry|j|j} } t!j"| j$} |j&d} | s|r|j)}n1t|t j*r|j&dn|| zdz}|j-|| || | ||j&d|j| }|jjdk(r2|0|jj.d vr|stj0|| }|S) Nrr8rzYou are attempting to perform batched generation with padding_side='right' this may lead to unexpected behaviour for Flash Attention version of Qwen3. Make sure to call `tokenizer.padding_side = 'left'` before tokenizing the input. rr)rpast_key_values_lengthr is_trainingr$)sequence_length target_lengthr:rr batch_sizerTr)cudaxpu)rTrsumitemsizer%r&rrr r_ignore_causal_mask_sdparrr:rr,finfominrFget_max_cache_shaper5_prepare_4d_causal_attention_mask_with_cache_positionr_unmask_unattended)r0rr0rrris_padding_rightr+using_static_cacheusing_sliding_window_cacher:r min_dtyper4r5rs r4r(zQwen3Model._update_causal_mask^s ;; + +/B B)o.I#1!R%#8#<#<#>#C#C#EIZIZI\]^I_#_ #$a )c^.C%% @O?Z?99;`a'E%/AS%T" KK , , 6'+E%%>>*'7#{{99 MM $**L,?,?vKK&** &,,Q/ %);+??AM nell;$$R(%7!; PP +')#))!,;;+Q   KK , , 6*%%**o=% 1CCKQZ[Kr5r4r5r:rr6c p||jdk(r|} | Stj|j} tj||f| ||} tj |||j ddkD} |j]t|tr||kDrHtj |||j dd|jz k} | j| | | z} | ddddddfj|ddd} || j} |jd|kDr |ddd|f}|jd} | ddddddd| f|ddddddfj| jz}|dk(}| ddddddd| fj!|| | ddddddd| f<| S)aT Creates a causal 4D mask of shape `(batch_size, 1, query_length, key_value_length)` from a 2D mask of shape `(batch_size, key_value_length)`, or if the input `attention_mask` is already 4D, do nothing. Args: attention_mask (`torch.Tensor`): A 2D attention mask of shape `(batch_size, key_value_length)` or a 4D attention mask of shape `(batch_size, 1, query_length, key_value_length)`. sequence_length (`int`): The sequence length being processed. target_length (`int`): The target length: when generating with static cache, the mask should be as long as the static cache, to account for the 0 padding, the part of the cache that is not filled yet. dtype (`torch.dtype`): The dtype to use for the 4D attention mask. device (`torch.device`): The device to place the 4D attention mask on. cache_position (`torch.Tensor`): Indices depicting the position of the input sequence tokens in the sequence. batch_size (`torch.Tensor`): Batch size. config (`Qwen3Config`): The model's configuration class past_key_values (`Cache`): The cache class that is being used currently to generate N) fill_valuer:rr"r8r$r)rbr,r=r>fullr'rvrrr bitwise_or_ruclonerFr;r masked_fill)rr4r5r:rrr6rTrrrEdiagonal_attend_masksliding_attend_mask mask_length padding_masks r4r@z@Qwen3Model._prepare_4d_causal_attention_mask_with_cache_positionsH  %.*<*<*>!*C(K:7 E*..I** -0Ye\bK$)<< f#MP^PfPfgiklPm#m $$0"/3EF/\iJi*/,,}V*T&..r158M8MM+')445HI / /K%dD!Q&67>>z1bRTUK))//1 !''+m;%3A~ ~4E%FN,2226 *1aL[L+@ANSTVZ\`bcScDdDgDg&&E   ,q0 5@Aq,;,AV5W5c5c )6 Aq!\k\12r5 NNNNNNNNN)F)rIrJrKrr%r*rrr rQWEN3_INPUTS_DOCSTRINGrr,rrr rrrrrrCr( staticmethodrr:rr@rLrMs@r4r r s { !"*+AB151537+/59$(,0/359i E,,-i !.i u//0 i "% i   1 12 i D>i $D>i 'tni !!1!12i $$89i  !i Ci b#( Q QllQ Q  Q  QfB BBB{{ B  B  BBBBBr5r c eZdZy)KwargsForCausalLMN)rIrJrKrr5r4rUrUsr5rUceZdZdgZddiZddgdgfiZfdZdZdZd Z d Z d Z d Z e ed ddeeeee ddeej,deej.deej,deedeej2deej,deedeedeedeej,deeej.fdeedefdZxZ S) Qwen3ForCausalLMzlm_head.weightlm_head colwise_repr@logitsct||t||_|j|_t j |j|jd|_|jyrQ) r)r*r rr rrVr1rXrr\s r4r*zQwen3ForCausalLM.__init__sU  '  ++yy!3!3V5F5FUS  r5c.|jjSr^rrrGs r4rz%Qwen3ForCausalLM.get_input_embeddingszz&&&r5c&||j_yr^r]rs r4rz%Qwen3ForCausalLM.set_input_embeddings "' r5c|jSr^rXrGs r4get_output_embeddingsz&Qwen3ForCausalLM.get_output_embeddingss ||r5c||_yr^rb)r0new_embeddingss r4set_output_embeddingsz&Qwen3ForCausalLM.set_output_embeddingss % r5c||_yr^r)r0decoders r4 set_decoderzQwen3ForCausalLM.set_decoders  r5c|jSr^rhrGs r4 get_decoderzQwen3ForCausalLM.get_decoders zzr5num_logits_to_keepz4.50logits_to_keep)versionnew_name) output_typerrrrmrrlabelsrrrrrrsc ||n|jj}| | n|jj} |jd|||||||| | d | } | j}t | t r t| dn| }|j|dd|ddf}d}|*|jd|||jjd| }t||| j| j| jS)a labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*): Labels for computing the masked language modeling loss. Indices should either be in `[0, ..., config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`. logits_to_keep (`int` or `torch.Tensor`, *optional*): If an `int`, compute logits for the last `logits_to_keep` tokens. If `0`, calculate logits for all `input_ids` (special case). Only last token logits are needed for generation, and calculating them only for that token can save memory, which becomes pretty significant for long sequences or large vocabulary size. If a `torch.Tensor`, must be 1D corresponding to the indices to keep in the sequence length dimension. This is useful when using packed tensor format (single dimension for batch and sequence length). Returns: Example: ```python >>> from transformers import AutoTokenizer, Qwen3ForCausalLM >>> model = Qwen3ForCausalLM.from_pretrained("Qwen/Qwen3-8B") >>> tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-8B") >>> prompt = "Hey, are you conscious? Can you talk to me?" >>> inputs = tokenizer(prompt, return_tensors="pt") >>> # Generate >>> generate_ids = model.generate(inputs.input_ids, max_length=30) >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0] "Hey, are you conscious? Can you talk to me?\nI'm not conscious, but I can talk to you." ```N) rrrmrrrrrr)rZrrr lossrZrr@r$r)rTrrrr#rrslicerX loss_functionr rrr@r$)r0rrrmrrrrrrrrrnrrr@ slice_indicesrZrus r4rCzQwen3ForCausalLM.forwardsd2C1N-TXT_T_TqTq$8$D $++JjJj  ,64:: , )%+'/!5) ,  ,  11 8B>SV8W~ot4]k mA}a,?@A  %4%%pVFt{{OeOepiopD%#33!//))   r5) NNNNNNNNNNr)!rIrJrK_tied_weights_keys_tp_plan_pp_planr*rrrcrfrjrlr r#rrRr"r_CONFIG_FOR_DOCrr,rrr rrrrrrUrCrLrMs@r4rWrWs*+=)H_-z:;H'(&)6DTU*+AB+AP_`151537+/59-1$(,0/35934P E,,-P !.P u//0 P "% P   1 12 P ))*P D>P $D>P 'tnP !!1!12P c5<</0P *+P  P aCVP r5rWa The Qwen3 Model transformer with a sequence classification head on top (linear layer). [`Qwen3ForSequenceClassification`] uses the last token in order to do the classification, as other causal models (e.g. GPT-2) do. Since it does classification on the last token, it requires to know the position of the last token. If a `pad_token_id` is defined in the configuration, it finds the last token that is not a padding token in each row. If no `pad_token_id` is defined, it simply takes the last value in each row of the batch. Since it cannot guess the padding tokens when `inputs_embeds` are passed instead of `input_ids`, it does the same (take the last value in each row of the batch). c(eZdZfdZdZdZeee dde e jde e jde e jde e de e jd e e jd e ed e ed e ed efdZxZS)Qwen3ForSequenceClassificationct|||j|_t||_t j |j|jd|_|jyrQ) r)r* num_labelsr rrrVr1scorerr\s r4r*z'Qwen3ForSequenceClassification.__init__sS   ++' YYv114??O  r5c.|jjSr^r]rGs r4rz3Qwen3ForSequenceClassification.get_input_embeddingsr^r5c&||j_yr^r]rs r4rz3Qwen3ForSequenceClassification.set_input_embeddingsr`r5rrrmrrrrrrrrsc |j|||||||| } | j} |j| } ||jd} n|jd} |jj | dk7r t d|jj d}n|||jj k7j| jtj}tj|jd| jtj}||zjd}n.d}tj|jj d| tj| | j |f}d}||j#| |||j }t%||| j&| j(| j* S)  labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*): Labels for computing the sequence classification/regression loss. Indices should be in `[0, ..., config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If `config.num_labels > 1` a classification loss is computed (Cross-Entropy). rrmrrrrrNrr$z=Cannot handle batch sizes > 1 if no padding token is defined.r8)rr:z will not detect padding tokens in `inputs_embeds`. Results may be unexpected if using padding tokens in conjunction with `inputs_embeds.`r")rZrr pooled_logitsrTrt)rr#rrFrTr r%r;rr,int32r'argmaxrrr3rIrwrrr@r$)r0rrrmrrrrrrrtransformer_outputsr@rZr6last_non_pad_token non_pad_mask token_indicesrrus r4rCz&Qwen3ForSequenceClassification.forwards*8!F!Fr!J !#    >>**+,ZZ  u||Jv}}MOaab   %%VFR_hlhshs%tD/ /??-;;*55   r5rQ)rIrJrKr*rrr rrRrr,rrr rrrrCrLrMs@r4r~r~qs '(*+AB151537+/59-1$(,0/3A E,,-A !.A u//0 A "% A   1 12 A ))*A D>A $D>A 'tnA  *A CA r5r~z The Qwen3 Model transformer with a token classification head on top (a linear layer on top of the hidden-states output) e.g. for Named-Entity-Recognition (NER) tasks. cDeZdZfdZdZdZeeee e e e  dde ejde ej de ejde ed e ej$d e ejd e ed e ed e ede fdZxZS)Qwen3ForTokenClassificationct|||j|_t||_t |dd |j }nt |dd |j}nd}tj||_ tj|j|j|_ |jy)Nclassifier_dropouthidden_dropoutg?)r)r*rr rrrrrDropoutrrVr1rr)r0rTrr3s r4r*z$Qwen3ForTokenClassification.__init__s   ++' 6/ 6 B!'!:!:  V-t 4 @!'!6!6 !$ zz"45 YYv1163D3DE  r5c.|jjSr^r]rGs r4rz0Qwen3ForTokenClassification.get_input_embeddingsr^r5c&||j_yr^r]rs r4rz0Qwen3ForTokenClassification.set_input_embeddingsr`r5) checkpointrqrrrrmrrrrrrrrsc |j|||||||| } | j} |j| } |j| } d} ||j | ||j } t | | | j| jS)rrN)rurZr@r$) rr#rrrwrTrr@r$)r0rrrmrrrrrrrrsequence_outputrZrus r4rCz#Qwen3ForTokenClassification.forwards4,0:: )%+'/!5,6 , "33,,7O,  %%ffdkkBD$!//))   r5rQ)rIrJrKr*rrr rrRr_CHECKPOINT_FOR_DOCrr|rr,rrr rrrCrLrMs@r4rrs '(*+AB&)$151537+/59-1$(,0/3* E,,-* !.* u//0 * "% *   1 12 * ))** D>* $D>* 'tn*  * C* r5rz The Qwen3 Model transformer with a span classification head on top for extractive question-answering tasks like SQuAD (a linear layer on top of the hidden-states output to compute `span start logits` and `span end logits`). c@eZdZdZfdZdZdZeee  dde e jde e jde e jde ed e e jd e e jd e e jd e ed e edefdZxZS)Qwen3ForQuestionAnswering transformerct||t||_t j |j d|_|jy)Nr7) r)r*r rrrVr1 qa_outputsrr\s r4r*z"Qwen3ForQuestionAnswering.__init__2sA  %f-))F$6$6: r5c.|jjSr^rrrGs r4rz.Qwen3ForQuestionAnswering.get_input_embeddings:s,,,r5c&||j_yr^rrs r4rz.Qwen3ForQuestionAnswering.set_input_embeddings=s(-%r5rrrmrrstart_positions end_positionsrrrsc |j||||||| } | j} |j| } | jdd\}}|j dj }|j dj }d}|||j ||||fi| }t|||| j| jS)a start_positions (`torch.LongTensor` of shape `(batch_size,)`, *optional*): Labels for position (index) of the start of the labelled span for computing the token classification loss. Positions are clamped to the length of the sequence (`sequence_length`). Position outside of the sequence are not taken into account for computing the loss. end_positions (`torch.LongTensor` of shape `(batch_size,)`, *optional*): Labels for position (index) of the end of the labelled span for computing the token classification loss. Positions are clamped to the length of the sequence (`sequence_length`). Position outside of the sequence are not taken into account for computing the loss. )rrmrrrrr$r8raN)ru start_logits end_logitsr@r$) rr#rsplitsqueezerrwrr@r$)r0rrrmrrrrrrrrrrZrrrus r4rCz!Qwen3ForQuestionAnswering.forward@s4,0+;+; )%+'/!5,<, "331#)<<r<#: j#++B/::< ''+668   &=+D%4%%lJQ^ibhiD+%!!//))   r5rQ)rIrJrKrr*rrr rrRrr,rrr rrrCrLrMs@r4rr(s &-.*+AB156:37+/596:48,0/33 E,,-3 !!2!233 u//0 3 "% 3   1 12 3 "%"2"233   0 013 $D>3 'tn3  &3 C3 r5r)rWrr rr~r)Nr$)r)L functoolsrtypingrrrrr,r activationsr cache_utilsr r r r generationrmodeling_attn_mask_utilsrmodeling_flash_attention_utilsrmodeling_outputsrrrrrmodeling_rope_utilsrrmodeling_utilsrrprocessing_utilsrutilsrrrrr r!r"utils.deprecationr#configuration_qwen3r% get_loggerrIrrr|Moduler'rOrfrqrrr{rrrrrQWEN3_START_DOCSTRINGrrRr rUrWr~rr__all__rr5r4rsv,33 !OO)>BLF&1,   H %%J299J( ryy  (6 UU\\ U# U%,, U&% II% <<% % << % U\\* %  %%4S)RYYS)l8 8v<299t +_t n S %9S S l H "6H H V F 4F F R r5