Ë
    ÜÓ^j9[  ã                   ó°  — d dl mZ d dlmZ d dlZd dlmZ ddlmZ ddlm	Z	m
Z
 ddlmZ dd	lmZmZmZ dd
lmZmZ ddlmZ ddlmZmZmZmZ ddlmZmZ ddlmZm Z  ddl!m"Z"m#Z# ddl$m%Z% ddl&m'Z'm(Z(m)Z) ddl*m+Z+m,Z, ddl-m.Z. ddl/m0Z0  ed«       G d„ dejb                  «      «       Z2 G d„ dejb                  «      Z3 G d„ dejb                  «      Z4d„ Z5 ed«      d>d„«       Z6d ejn                  d!e8d"ejn                  fd#„Z9	 d?d$ejb                  d%ejn                  d&ejn                  d'ejn                  d(ejn                  dz  d)e:d*e:d+e%e'   fd,„Z; ee6«       G d-„ d.ejb                  «      «       Z< G d/„ d0e«      Z=e( G d1„ d2e#«      «       Z>e( G d3„ d4e>«      «       Z?e( G d5„ d6e>e«      «       Z@ G d7„ d8ee>«      ZA G d9„ d:ee>«      ZB G d;„ d<ee>«      ZCg d=¢ZDy)@é    )ÚCallable)ÚOptionalN)Únné   )ÚACT2FN)ÚCacheÚDynamicCache)ÚGenerationMixin)Úuse_kernel_forward_from_hubÚuse_kernel_func_from_hubÚuse_kernelized_func)Úcreate_causal_maskÚ!create_sliding_window_causal_mask)ÚFlashAttentionKwargs)ÚGenericForQuestionAnsweringÚ GenericForSequenceClassificationÚGenericForTokenClassificationÚGradientCheckpointingLayer)ÚBaseModelOutputWithPastÚCausalLMOutputWithPast)ÚROPE_INIT_FUNCTIONSÚdynamic_rope_update)ÚALL_ATTENTION_FUNCTIONSÚPreTrainedModel)ÚUnpack)ÚTransformersKwargsÚauto_docstringÚcan_return_tuple)Úmaybe_autocastÚmerge_with_config_defaults)Úcapture_outputsé   )ÚQwen3ConfigÚRMSNormc                   óh   ‡ — e Zd Zddeddfˆ fd„Zdej                  dej                  fd„Zd„ Zˆ xZ	S )	ÚQwen3RMSNormÚepsÚreturnNc                 óŠ   •— t         ‰| �  «        t        j                  t	        j
                  |«      «      | _        || _        y)z;
        Qwen3RMSNorm is equivalent to T5LayerNorm
        N)ÚsuperÚ__init__r   Ú	ParameterÚtorchÚonesÚweightÚvariance_epsilon)ÚselfÚhidden_sizer'   Ú	__class__s      €ús/var/www/ramen.bs-engineer-server.com/venv/lib/python3.12/site-packages/transformers/models/qwen3/modeling_qwen3.pyr+   zQwen3RMSNorm.__init__3   s1   ø€ ô 	‰ÑÔÜ—l‘l¤5§:¡:¨kÓ#:Ó;ˆŒØ #ˆÕó    Úhidden_statesc                 ó"  — |j                   }|j                  t        j                  «      }|j	                  d«      j                  dd¬«      }|t        j                  || j                  z   «      z  }| j                  |j                  |«      z  S )Né   éÿÿÿÿT)Úkeepdim)	ÚdtypeÚtor-   Úfloat32ÚpowÚmeanÚrsqrtr0   r/   )r1   r6   Úinput_dtypeÚvariances       r4   ÚforwardzQwen3RMSNorm.forward;   sy   € Ø#×)Ñ)ˆØ%×(Ñ(¬¯©Ó7ˆØ ×$Ñ$ QÓ'×,Ñ,¨R¸Ð,Ó>ˆØ%¬¯©°H¸t×?TÑ?TÑ4TÓ(UÑUˆØ�{‰{˜]×-Ñ-¨kÓ:Ñ:Ð:r5   c                 ó^   — t        | j                  j                  «      › d| j                  › �S )Nz, eps=)Útupler/   Úshaper0   )r1   s    r4   Ú
extra_reprzQwen3RMSNorm.extra_reprB   s*   € Ü˜Ÿ™×)Ñ)Ó*Ð+¨6°$×2GÑ2GÐ1HÐIÐIr5   )g�íµ ÷Æ°>)
Ú__name__Ú
__module__Ú__qualname__Úfloatr+   r-   ÚTensorrC   rG   Ú__classcell__©r3   s   @r4   r&   r&   1   s7   ø„ ñ$¨ð $¸$õ $ð; U§\¡\ð ;°e·l±ló ;öJr5   r&   c                   ó$   ‡ — e Zd Zˆ fd„Zd„ Zˆ xZS )ÚQwen3MLPc                 óÊ  •— t         ‰| �  «        || _        |j                  | _        |j                  | _        t        j                  | j                  | j                  d¬«      | _        t        j                  | j                  | j                  d¬«      | _        t        j                  | j                  | j                  d¬«      | _	        t        |j                     | _        y ©NF©Úbias)r*   r+   Úconfigr2   Úintermediate_sizer   ÚLinearÚ	gate_projÚup_projÚ	down_projr   Ú
hidden_actÚact_fn©r1   rU   r3   s     €r4   r+   zQwen3MLP.__init__G   s¦   ø€ Ü‰ÑÔØˆŒØ!×-Ñ-ˆÔØ!'×!9Ñ!9ˆÔÜŸ™ 4×#3Ñ#3°T×5KÑ5KÐRWÔXˆŒÜ—y‘y ×!1Ñ!1°4×3IÑ3IÐPUÔVˆŒÜŸ™ 4×#9Ñ#9¸4×;KÑ;KÐRWÔXˆŒÜ˜V×.Ñ.Ñ/ˆ�r5   c                 óˆ   — | j                  | j                  | j                  |«      «      | j                  |«      z  «      }|S ©N)rZ   r\   rX   rY   )r1   ÚxrZ   s      r4   rC   zQwen3MLP.forwardQ   s6   € Ø—N‘N 4§;¡;¨t¯~©~¸aÓ/@Ó#AÀDÇLÁLÐQRÃOÑ#SÓTˆ	ØÐr5   )rH   rI   rJ   r+   rC   rM   rN   s   @r4   rP   rP   F   s   ø„ ô0ör5   rP   c                   óÄ   ‡ — e Zd ZU ej                  ed<   ddefˆ fd„Ze	 	 	 ddedz  de	d   de
dz  ded	ef   fd
„«       Z ej                  «       ed„ «       «       Zˆ xZS )ÚQwen3RotaryEmbeddingÚinv_freqNrU   c                 ó®  •— t         ‰| �  «        |j                  | _        |j                  | _        || _        | j
                  j                  d   | _        | j                  }| j                  dk7  rt        | j                     } || j
                  |«      \  }| _
        | j                  d|d¬«       | j                  d|j                  «       d¬«       y )NÚ	rope_typeÚdefaultrc   F)Ú
persistentÚoriginal_inv_freq)r*   r+   Úmax_position_embeddingsÚmax_seq_len_cachedÚoriginal_max_seq_lenrU   Úrope_parametersre   Úcompute_default_rope_parametersr   Úattention_scalingÚregister_bufferÚclone)r1   rU   ÚdeviceÚrope_init_fnrc   r3   s        €r4   r+   zQwen3RotaryEmbedding.__init__Y   s¶   ø€ Ü‰ÑÔØ"(×"@Ñ"@ˆÔØ$*×$BÑ$BˆÔ!àˆŒàŸ™×4Ñ4°[ÑAˆŒØ!%×!EÑ!EˆØ�>‰>˜YÒ&Ü.¨t¯~©~Ñ>ˆLÙ+7¸¿¹ÀVÓ+LÑ(ˆ�$Ô(à×Ñ˜Z¨¸eÐÔDØ×ÑÐ0°(·.±.Ó2BÈuÐÕUr5   rq   ztorch.deviceÚseq_lenr(   ztorch.Tensorc                 ó  — | j                   d   }t        | dd«      xs | j                  | j                  z  }d}d|t	        j
                  d|dt        j                  ¬«      j                  |t        j                  ¬«      |z  z  z  }||fS )	a¨  
        Computes the inverse frequencies according to the original RoPE implementation
        Args:
            config ([`~transformers.PreTrainedConfig`]):
                The model configuration.
            device (`torch.device`):
                The device to use for initialization of the inverse frequencies.
            seq_len (`int`, *optional*):
                The current sequence length. Unused for this type of RoPE.
        Returns:
            Tuple of (`torch.Tensor`, `float`), containing the inverse frequencies for the RoPE embeddings and the
            post-processing scaling factor applied to the computed cos/sin (unused in this type of RoPE).
        Ú
rope_thetaÚhead_dimNg      ð?r   r8   ©r;   )rq   r;   )	rl   Úgetattrr2   Únum_attention_headsr-   ÚarangeÚint64r<   rK   )rU   rq   rs   ÚbaseÚdimÚattention_factorrc   s          r4   rm   z4Qwen3RotaryEmbedding.compute_default_rope_parametersi   s�   € ð& ×%Ñ% lÑ3ˆÜ�f˜j¨$Ó/Òc°6×3EÑ3EÈ×IcÑIcÑ3cˆàÐð Ø”U—\‘\ ! S¨!´5·;±;Ô?×BÑBÈ&ÔX]×XcÑXcÐBÓdÐgjÑjÑkñ
ˆð Ð)Ð)Ð)r5   c                 óN  — | j                   d d d …d f   j                  «       j                  |j                  d   dd«      j	                  |j
                  «      }|d d …d d d …f   j                  «       }t        |j
                  j                  t        «      r/|j
                  j                  dk7  r|j
                  j                  nd}t        |d¬«      5  |j                  «       |j                  «       z  j                  dd«      }t        j                  ||fd¬	«      }|j                  «       | j                  z  }|j                  «       | j                  z  }	d d d «       j	                  |j                   ¬
«      	j	                  |j                   ¬
«      fS # 1 sw Y   ŒAxY w)Nr   r9   r"   ÚmpsÚcpuF)Údevice_typeÚenabledr8   ©r}   rw   )rc   rK   ÚexpandrF   r<   rq   Ú
isinstanceÚtypeÚstrr   Ú	transposer-   ÚcatÚcosrn   Úsinr;   )
r1   r`   Úposition_idsÚinv_freq_expandedÚposition_ids_expandedr‚   ÚfreqsÚembr‹   rŒ   s
             r4   rC   zQwen3RotaryEmbedding.forward‡   sR  € ð !ŸM™M¨$²°4¨-Ñ8×>Ñ>Ó@×GÑGÈ×HZÑHZÐ[\ÑH]Ð_aÐcdÓe×hÑhÐij×iqÑiqÓrÐØ ,ªQ°²a¨ZÑ 8× >Ñ >Ó @Ðä'1°!·(±(·-±-ÄÔ'EÈ!Ï(É(Ï-É-Ð[`ÒJ`�a—h‘h—m’mÐfkˆÜ¨¸UÔCñ 	5Ø&×,Ñ,Ó.Ð1F×1LÑ1LÓ1NÑN×YÑYÐZ[Ð]^Ó_ˆEÜ—)‘)˜U E˜N°Ô3ˆCØ—'‘'“)˜d×4Ñ4Ñ4ˆCØ—'‘'“)˜d×4Ñ4Ñ4ˆC÷		5ð �v‰v˜AŸG™GˆvÓ$ c§f¡f°1·7±7 fÓ&;Ð;Ð;÷	5ð 	5ús   ÃBFÆF$r_   )NNN)rH   rI   rJ   r-   rL   Ú__annotations__r#   r+   Ústaticmethodr   ÚintrE   rK   rm   Úno_gradr   rC   rM   rN   s   @r4   rb   rb   V   s˜   ø… Ø�l‰lÓñV˜{õ Vð  à%)Ø+/Ø"ñ*Ø˜dÑ"ð*à˜Ñ(ð*ð �t‘ð*ð 
ˆ~˜uÐ$Ñ	%ò	*ó ð*ð: €U‡]�]ƒ_Øñ<ó ó ô<r5   rb   c                 óš   — | dd| j                   d   dz  …f   }| d| j                   d   dz  d…f   }t        j                  | |fd¬«      S )z*Rotates half the hidden dims of the input..Nr9   r8   r„   )rF   r-   rŠ   )r`   Úx1Úx2s      r4   Úrotate_halfr™   —   sZ   € à	
ˆ3Ð"�!—'‘'˜"‘+ Ñ"Ð"Ð"Ñ	#€BØ	
ˆ3�—‘˜‘˜qÑ Ñ"Ð"Ñ	#€BÜ�9‰9�r�c˜2�Y BÔ'Ð'r5   Úrotary_pos_embc                 óž   — |j                  |«      }|j                  |«      }| |z  t        | «      |z  z   }||z  t        |«      |z  z   }||fS )a…  Applies 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.
        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.
    )Ú	unsqueezer™   )ÚqÚkr‹   rŒ   Úunsqueeze_dimÚq_embedÚk_embeds          r4   Úapply_rotary_pos_embr¢   ž   sY   € ð& �-‰-˜Ó
&€CØ
�-‰-˜Ó
&€CØ�3‰wœ; q›>¨CÑ/Ñ0€GØ�3‰wœ; q›>¨CÑ/Ñ0€GØ�GÐÐr5   r6   Ún_repr(   c                 óª   — | j                   \  }}}}|dk(  r| S | dd…dd…ddd…dd…f   j                  |||||«      } | j                  |||z  ||«      S )zÔ
    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)rF   r…   Úreshape)r6   r£   ÚbatchÚnum_key_value_headsÚslenrv   s         r4   Ú	repeat_kvr©   ¸   so   € ð
 2?×1DÑ1DÑ.€EÐ  hØ�‚zØÐØ!¢!¢Q¨ªa²Ð"2Ñ3×:Ñ:¸5ÐBUÐW\Ð^bÐdlÓm€MØ× Ñ  Ð(;¸eÑ(CÀTÈ8ÓTÐTr5   ÚmoduleÚqueryÚkeyÚvalueÚattention_maskÚscalingÚdropoutÚkwargsc                 ó  — t        || j                  «      }t        || j                  «      }	t        j                  ||j	                  dd«      «      |z  }
|�|
|z   }
t
        j                  j                  |
dt        j                  ¬«      j                  |j                  «      }
t
        j                  j                  |
|| j                  ¬«      }
t        j                  |
|	«      }|j	                  dd«      j                  «       }||
fS )Nr8   r   r9   )r}   r;   )ÚpÚtrainingr"   )r©   Únum_key_value_groupsr-   Úmatmulr‰   r   Ú
functionalÚsoftmaxr=   r<   r;   r°   r´   Ú
contiguous)rª   r«   r¬   r­   r®   r¯   r°   r±   Ú
key_statesÚvalue_statesÚattn_weightsÚattn_outputs               r4   Úeager_attention_forwardr¾   Ä   sà   € ô ˜3 × ;Ñ ;Ó<€JÜ˜U F×$?Ñ$?Ó@€Lä—<‘<  z×';Ñ';¸A¸qÓ'AÓBÀWÑL€LØÐ!Ø# nÑ4ˆä—=‘=×(Ñ(¨¸2ÄUÇ]Á]Ð(ÓS×VÑVÐW\×WbÑWbÓc€LÜ—=‘=×(Ñ(¨¸È6Ï?É?Ð(Ó[€LÜ—,‘,˜|¨\Ó:€KØ×'Ñ'¨¨1Ó-×8Ñ8Ó:€Kà˜Ð$Ð$r5   c                   óô   ‡ — e Zd ZdZdedefˆ fd„Z	 ddej                  de	ej                  ej                  f   dej                  dz  d	e
dz  d
ee   de	ej                  ej                  dz  f   fd„Zˆ xZS )ÚQwen3Attentionz=Multi-headed attention from 'Attention Is All You Need' paperrU   Ú	layer_idxc                 ó�  •— t         ‰| �  «        t        |d«      r|j                  |   nd | _        || _        || _        t        |d|j                  |j                  z  «      | _
        |j                  |j                  z  | _        | j                  dz  | _        |j                  | _        d| _        t!        j"                  |j                  |j                  | j                  z  |j$                  ¬«      | _        t!        j"                  |j                  |j                  | j                  z  |j$                  ¬«      | _        t!        j"                  |j                  |j                  | j                  z  |j$                  ¬«      | _        t!        j"                  |j                  | j                  z  |j                  |j$                  ¬«      | _        t/        | j                  |j0                  ¬«      | _        t/        | j                  |j0                  ¬«      | _        | j                  dk(  r|j6                  | _        y d | _        y )NÚlayer_typesrv   g      à¿TrS   ©r'   Úsliding_attention)r*   r+   ÚhasattrrÃ   Ú
layer_typerU   rÁ   rx   r2   ry   rv   r§   rµ   r¯   Úattention_dropoutÚ	is_causalr   rW   Úattention_biasÚq_projÚk_projÚv_projÚo_projr&   Úrms_norm_epsÚq_normÚk_normÚsliding_window©r1   rU   rÁ   r3   s      €r4   r+   zQwen3Attention.__init__á   s¿  ø€ Ü‰ÑÔÜ;BÀ6È=Ô;Y˜&×,Ñ,¨YÒ7Ð_cˆŒØˆŒØ"ˆŒÜ ¨
°F×4FÑ4FÈ&×JdÑJdÑ4dÓeˆŒØ$*×$>Ñ$>À&×B\ÑB\Ñ$\ˆÔ!Ø—}‘} dÑ*ˆŒØ!'×!9Ñ!9ˆÔØˆŒä—i‘iØ×Ñ × :Ñ :¸T¿]¹]Ñ JÐQW×QfÑQfô
ˆŒô —i‘iØ×Ñ × :Ñ :¸T¿]¹]Ñ JÐQW×QfÑQfô
ˆŒô —i‘iØ×Ñ × :Ñ :¸T¿]¹]Ñ JÐQW×QfÑQfô
ˆŒô —i‘iØ×&Ñ&¨¯©Ñ6¸×8JÑ8JÐQW×QfÑQfô
ˆŒô # 4§=¡=°f×6IÑ6IÔJˆŒÜ" 4§=¡=°f×6IÑ6IÔJˆŒØ7;·±ÐJ]Ò7]˜f×3Ñ3ˆÕÐcgˆÕr5   Nr6   Úposition_embeddingsr®   Úpast_key_valuesr±   r(   c                 ó\  — |j                   d d }g |¢d‘| j                  ‘­}| j                  | j                  |«      j	                  |«      «      j                  dd«      }| j                  | j                  |«      j	                  |«      «      j                  dd«      }	| j                  |«      j	                  |«      j                  dd«      }
|\  }}t        ||	||«      \  }}	|� |j                  |	|
| j                  «      \  }	}
t        j                  | j                  j                  t         «      } || ||	|
|f| j"                  sdn| j$                  | j&                  | j(                  dœ|¤Ž\  }} |j*                  g |¢d‘­Ž j-                  «       }| j/                  |«      }||fS )Nr9   r"   r8   ç        )r°   r¯   rÒ   )rF   rv   rÐ   rË   Úviewr‰   rÑ   rÌ   rÍ   r¢   ÚupdaterÁ   r   Úget_interfacerU   Ú_attn_implementationr¾   r´   rÈ   r¯   rÒ   r¥   r¹   rÎ   )r1   r6   rÔ   r®   rÕ   r±   Úinput_shapeÚhidden_shapeÚquery_statesrº   r»   r‹   rŒ   Úattention_interfacer½   r¼   s                   r4   rC   zQwen3Attention.forwardü   s®  € ð $×)Ñ)¨#¨2Ð.ˆØ8˜Ð8 bÐ8¨$¯-©-Ñ8ˆà—{‘{ 4§;¡;¨}Ó#=×#BÑ#BÀ<Ó#PÓQ×[Ñ[Ð\]Ð_`ÓaˆØ—[‘[ §¡¨]Ó!;×!@Ñ!@ÀÓ!NÓO×YÑYÐZ[Ð]^Ó_ˆ
Ø—{‘{ =Ó1×6Ñ6°|ÓD×NÑNÈqÐRSÓTˆà&‰ˆˆSÜ#7¸ÀjÐRUÐWZÓ#[Ñ ˆ�jàÐ&Ø'6×'=Ñ'=¸jÈ,ÐX\×XfÑXfÓ'gÑ$ˆJ˜ä(?×(MÑ(MØ�K‰K×,Ñ,Ô.Eó)
Ðñ %8ØØØØØð
%
ð  $Ÿ}š}‘C°$×2HÑ2HØ—L‘LØ×.Ñ.ñ
%
ð ñ
%
Ñ!ˆ�\ð *�k×)Ñ)Ð;¨;Ð;¸Ò;×FÑFÓHˆØ—k‘k +Ó.ˆØ˜LÐ(Ð(r5   r_   )rH   rI   rJ   Ú__doc__r#   r”   r+   r-   rL   rE   r   r   r   rC   rM   rN   s   @r4   rÀ   rÀ   Ý   s¡   ø„ áGðh˜{ð h°sõ hð@ )-ñ')à—|‘|ð')ð # 5§<¡<°·±Ð#=Ñ>ð')ð Ÿ™ tÑ+ð	')ð
  ™ð')ð Ð-Ñ.ð')ð 
ˆu�|‰|˜UŸ\™\¨DÑ0Ð0Ñ	1÷')r5   rÀ   c                   ó  ‡ — e Zd Zdedefˆ fd„Z	 	 	 	 	 ddej                  dej                  dz  dej                  dz  de	dz  d	e
dz  d
eej                  ej                  f   dz  dee   dej                  fd„Zˆ xZS )ÚQwen3DecoderLayerrU   rÁ   c                 ó   •— t         ‰| �  «        |j                  | _        t        ||¬«      | _        t        |«      | _        t        |j                  |j                  ¬«      | _	        t        |j                  |j                  ¬«      | _
        y )N)rU   rÁ   rÄ   )r*   r+   r2   rÀ   Ú	self_attnrP   Úmlpr&   rÏ   Úinput_layernormÚpost_attention_layernormrÓ   s      €r4   r+   zQwen3DecoderLayer.__init__'  sl   ø€ Ü‰ÑÔØ!×-Ñ-ˆÔä'¨vÀÔKˆŒä˜FÓ#ˆŒÜ+¨F×,>Ñ,>ÀF×DWÑDWÔXˆÔÜ(4°V×5GÑ5GÈV×M`ÑM`Ô(aˆÕ%r5   Nr6   r®   r�   rÕ   Ú	use_cacherÔ   r±   r(   c           
      óÀ   — |}| j                  |«      } | j                  d||||||dœ|¤Ž\  }}	||z   }|}| j                  |«      }| j                  |«      }||z   }|S )N)r6   r®   r�   rÕ   rè   rÔ   © )ræ   rä   rç   rå   )
r1   r6   r®   r�   rÕ   rè   rÔ   r±   ÚresidualÚ_s
             r4   rC   zQwen3DecoderLayer.forward1  s�   € ð !ˆØ×,Ñ,¨]Ó;ˆà)˜4Ÿ>™>ð 
Ø'Ø)Ø%Ø+ØØ 3ñ
ð ñ
Ñˆ�qð ! =Ñ0ˆð !ˆØ×5Ñ5°mÓDˆØŸ™ Ó/ˆØ  =Ñ0ˆØÐr5   )NNNFN)rH   rI   rJ   r#   r”   r+   r-   rL   Ú
LongTensorr   ÚboolrE   r   r   rC   rM   rN   s   @r4   râ   râ   &  s¾   ø„ ðb˜{ð b°sõ bð /3Ø04Ø(,Ø!&ØHLñà—|‘|ðð Ÿ™ tÑ+ðð ×&Ñ&¨Ñ-ð	ð
  ™ðð ˜$‘;ðð # 5§<¡<°·±Ð#=Ñ>ÀÑEðð Ð+Ñ,ðð 
�‰÷r5   râ   c                   óJ   — e Zd ZU eed<   dZdZdgZdgZdZ	dZ
dZdZdZeedœZy)ÚQwen3PreTrainedModelrU   ÚmodelTrâ   rÕ   )r6   Ú
attentionsN)rH   rI   rJ   r#   r’   Úbase_model_prefixÚsupports_gradient_checkpointingÚ_no_split_modulesÚ_skip_keys_device_placementÚ_supports_flash_attnÚ_supports_sdpaÚ_supports_flex_attnÚ_can_compile_fullgraphÚ_supports_attention_backendrâ   rÀ   Ú_can_record_outputsrê   r5   r4   rð   rð   Q  sQ   … àÓØÐØ&*Ð#Ø,Ð-ÐØ#4Ð"5ÐØÐØ€NØÐà!ÐØ"&Ðà*Ø$ñÑr5   rð   c                   óò   ‡ — e Zd Zdefˆ fd„Zeee	 	 	 	 	 	 ddej                  dz  dej                  dz  dej                  dz  dedz  dej                  dz  d	edz  d
ee   defd„«       «       «       Zˆ xZS )Ú
Qwen3ModelrU   c           	      óF  •— t         ‰| �  |«       |j                  | _        |j                  | _        t        j                  |j                  |j                  | j                  «      | _        t        j                  t        |j                  «      D �cg c]  }t        ||«      ‘Œ c}«      | _        t        |j                  |j                  ¬«      | _        t#        |¬«      | _        d| _        d| j(                  j*                  v | _        | j/                  «        y c c}w )NrÄ   ©rU   FrÅ   )r*   r+   Úpad_token_idÚpadding_idxÚ
vocab_sizer   Ú	Embeddingr2   Úembed_tokensÚ
ModuleListÚrangeÚnum_hidden_layersrâ   Úlayersr&   rÏ   Únormrb   Ú
rotary_embÚgradient_checkpointingrU   rÃ   Úhas_sliding_layersÚ	post_initrÓ   s      €r4   r+   zQwen3Model.__init__f  sÝ   ø€ Ü‰Ñ˜Ô Ø!×.Ñ.ˆÔØ ×+Ñ+ˆŒäŸL™L¨×):Ñ):¸F×<NÑ<NÐPT×P`ÑP`ÓaˆÔÜ—m‘mÜCHÈ×IaÑIaÓCbÖc°iÔ˜v yÕ1Òcó
ˆŒô ! ×!3Ñ!3¸×9LÑ9LÔMˆŒ	Ü.°fÔ=ˆŒØ&+ˆÔ#Ø"5¸¿¹×9PÑ9PÐ"PˆÔð 	�‰Õùò ds   ÂDNÚ	input_idsr®   r�   rÕ   Úinputs_embedsrè   r±   r(   c           
      óú  — |d u |d uz  rt        d«      ‚|€| j                  |«      }|r|€t        | j                  ¬«      }|€V|�|j	                  «       nd}t        j                  |j                  d   |j                  ¬«      |z   }|j                  d«      }t        |x}	t        «      s9| j                  ||||dœ}
dt        di |
¤Ži}	| j                  rt        di |
¤Ž|	d<   |}| j                  ||«      }t!        | j"                  d | j                  j$                   «      D ].  \  }} ||f|	| j                  j&                  |      ||||d	œ|¤Ž}Œ0 | j)                  |«      }t+        ||r|¬
«      S d ¬
«      S )Nz:You must specify exactly one of input_ids or inputs_embedsr   r   r"   )rq   )rU   r  r®   rÕ   r�   Úfull_attentionrÅ   )r®   rÔ   r�   rÕ   rè   )Úlast_hidden_staterÕ   rê   )Ú
ValueErrorr  r	   rU   Úget_seq_lengthr-   rz   rF   rq   rœ   r†   Údictr   r  r   r  Ú	enumerater	  r  rÃ   r
  r   )r1   r  r®   r�   rÕ   r  rè   r±   Úpast_seen_tokensÚcausal_mask_mappingÚmask_kwargsr6   rÔ   ÚiÚdecoder_layers                  r4   rC   zQwen3Model.forwardw  s¾  € ð ˜Ð -°tÐ";Ò<ÜÐYÓZÐZàÐ Ø ×-Ñ-¨iÓ8ˆMá˜Ð0Ü*°$·+±+Ô>ˆOàÐØCRÐC^˜×=Ñ=Ô?ÐdeÐÜ Ÿ<™<¨×(;Ñ(;¸AÑ(>À}×G[ÑG[Ô\Ð_oÑoˆLØ'×1Ñ1°!Ó4ˆLô °Ð?Ð-ÄÔFð Ÿ+™+Ø!.Ø"0Ø#2Ø ,ñˆKð !Ô"4Ñ"C°{Ñ"Cð#Ðð ×&Ò&Ü;\Ñ;kÐ_jÑ;kÐ#Ð$7Ñ8à%ˆØ"Ÿo™o¨m¸\ÓJÐä )¨$¯+©+Ð6U¸¿¹×8UÑ8UÐ*VÓ Wò 		ÑˆAˆ}Ù)Øðà2°4·;±;×3JÑ3JÈ1Ñ3MÑNØ$7Ø)Ø /Ø#ñð ñ‰Mð		ð Ÿ	™	 -Ó0ˆÜ&Ø+Ù/8˜Oô
ð 	
à>Bô
ð 	
r5   )NNNNNN)rH   rI   rJ   r#   r+   r    r!   r   r-   rí   rL   r   ÚFloatTensorrî   r   r   r   rC   rM   rN   s   @r4   rþ   rþ   d  sÍ   ø„ ð˜{õ ð"  ØØð .2Ø.2Ø04Ø(,Ø26Ø!%ñ<
à×#Ñ# dÑ*ð<
ð Ÿ™ tÑ+ð<
ð ×&Ñ&¨Ñ-ð	<
ð
  ™ð<
ð ×(Ñ(¨4Ñ/ð<
ð ˜$‘;ð<
ð Ð+Ñ,ð<
ð 
!ò<
ó ó ó  ô<
r5   rþ   c                   óB  ‡ — e Zd ZddiZddiZddgdgfiZˆ fd„Zee	 	 	 	 	 	 	 	 dd	e	j                  dz  d
e	j                  dz  de	j                  dz  dedz  de	j                  dz  de	j                  dz  dedz  dee	j                  z  dee   defd„«       «       Zˆ xZS )ÚQwen3ForCausalLMzlm_head.weightzmodel.embed_tokens.weightÚlm_headÚcolwise_gather_outputr6   Úlogitsc                 óè   •— t         ‰| �  |«       t        |«      | _        |j                  | _        t        j                  |j                  |j                  d¬«      | _        | j                  «        y rR   )
r*   r+   rþ   rñ   r  r   rW   r2   r   r  r]   s     €r4   r+   zQwen3ForCausalLM.__init__¿  sU   ø€ Ü‰Ñ˜Ô Ü Ó'ˆŒ
Ø ×+Ñ+ˆŒÜ—y‘y ×!3Ñ!3°V×5FÑ5FÈUÔSˆŒð 	�‰Õr5   Nr  r®   r�   rÕ   r  Úlabelsrè   Úlogits_to_keepr±   r(   c	           
      óx  —  | j                   d||||||dœ|	¤Ž}
|
j                  }t        |t        «      rt	        | d«      n|}| j                  |dd…|dd…f   «      }d}|�* | j                  d||| j                  j                  dœ|	¤Ž}t        |||
j                  |
j                  |
j                  ¬«      S )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]`.

        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."
        ```)r  r®   r�   rÕ   r  rè   N)r"  r$  r  )Úlossr"  rÕ   r6   rò   rê   )rñ   r  r†   r”   Úslicer   Úloss_functionrU   r  r   rÕ   r6   rò   )r1   r  r®   r�   rÕ   r  r$  rè   r%  r±   Úoutputsr6   Úslice_indicesr"  r'  s                  r4   rC   zQwen3ForCausalLM.forwardÈ  sÚ   € ðH ,6¨4¯:©:ð ,
ØØ)Ø%Ø+Ø'Øñ,
ð ñ,
ˆð  ×1Ñ1ˆä8BÀ>ÔSVÔ8Wœ˜~˜o¨tÔ4Ð]kˆØ—‘˜mªA¨}ºaÐ,?Ñ@ÓAˆàˆØÐØ%�4×%Ñ%Ðp¨V¸FÈtÏ{É{×OeÑOeÑpÐioÑpˆDä%ØØØ#×3Ñ3Ø!×/Ñ/Ø×)Ñ)ô
ð 	
r5   )NNNNNNNr   )rH   rI   rJ   Ú_tied_weights_keysÚ_tp_planÚ_pp_planr+   r   r   r-   rí   rL   r   r  rî   r”   r   r   r   rC   rM   rN   s   @r4   r  r  ¹  s  ø„ à*Ð,GÐHÐØÐ2Ð3€HØ˜_Ð-°¨zÐ:Ð;€Hôð Øð .2Ø.2Ø04Ø(,Ø26Ø*.Ø!%Ø-.ñ;
à×#Ñ# dÑ*ð;
ð Ÿ™ tÑ+ð;
ð ×&Ñ&¨Ñ-ð	;
ð
  ™ð;
ð ×(Ñ(¨4Ñ/ð;
ð × Ñ  4Ñ'ð;
ð ˜$‘;ð;
ð ˜eŸl™lÑ*ð;
ð Ð+Ñ,ð;
ð 
 ò;
ó ó ô;
r5   r  c                   ó   — e Zd Zy)ÚQwen3ForSequenceClassificationN©rH   rI   rJ   rê   r5   r4   r0  r0    ó   „ Ør5   r0  c                   ó   — e Zd Zy)ÚQwen3ForTokenClassificationNr1  rê   r5   r4   r4  r4    r2  r5   r4  c                   ó   — e Zd ZdZy)ÚQwen3ForQuestionAnsweringÚtransformerN)rH   rI   rJ   ró   rê   r5   r4   r6  r6    s   „ Ø%Ñr5   r6  )r  r6  rð   rþ   r0  r4  )r"   )r×   )EÚcollections.abcr   Útypingr   r-   r   Úactivationsr   Úcache_utilsr   r	   Ú
generationr
   Úintegrationsr   r   r   Úmasking_utilsr   r   Úmodeling_flash_attention_utilsr   Úmodeling_layersr   r   r   r   Úmodeling_outputsr   r   Úmodeling_rope_utilsr   r   Úmodeling_utilsr   r   Úprocessing_utilsr   Úutilsr   r   r   Úutils.genericr   r    Úutils.output_capturingr!   Úconfiguration_qwen3r#   ÚModuler&   rP   rb   r™   r¢   rL   r”   r©   rK   r¾   rÀ   râ   rð   rþ   r  r0  r4  r6  Ú__all__rê   r5   r4   ú<module>rK     s  ðõ* %Ý ã Ý å !ß .Ý )ß fÑ fß RÝ B÷ó ÷ Pß Kß FÝ &ß IÑ Iß GÝ 5Ý ,ñ ˜YÓ'ôJ�2—9‘9ó Jó (ðJô(ˆr�y‰yô ô ><˜2Ÿ9™9ô ><òB(ñ Ð*Ó+òó ,ðð2	U˜UŸ\™\ð 	U°#ð 	U¸%¿,¹,ó 	Uð& ñ%Ø�I‰Ið%à�<‰<ð%ð 
�‰ð%ð �<‰<ð	%ð
 —L‘L 4Ñ'ð%ð ð%ð ð%ð Ð'Ñ(ó%ñ2 Ð)Ó*ôE)�R—Y‘Yó E)ó +ðE)ôP(Ð2ô (ðV ô˜?ó ó ðð$ ôQ
Ð%ó Q
ó ðQ
ðh ôK
Ð+¨_ó K
ó ðK
ô\	Ð%EÐG[ô 	ô	Ð"?ÐAUô 	ô&Ð ;Ð=Qô &ò�r5   