§
    9²¸j§~  ã                   óŠ  — d dl mZ d dlmZ d dlZd dlmc mZ d dlm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 ddl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$m%Z% ddl&m'Z'm(Z( ddl)m*Z* ddl+m,Z,m-Z-m.Z. ddl/m0Z0 ddl1m2Z2  G d„ ded¬¦  «        Z3 G d„ dej4        ¦  «        Z5 ed¦  «         G d„ dej4        ¦  «        ¦   «         Z6 G d „ d!ej4        ¦  «        Z7e G d"„ d#ej4        ¦  «        ¦   «         Z8 G d$„ d%ej4        ¦  «        Z9d&„ Z: ed'¦  «        dHd(„¦   «         Z;d)ej<        d*e=d+ej<        fd,„Z>	 dId.ej4        d/ej<        d0ej<        d1ej<        d2ej<        dz  d3e?d4e?d5e%e'         fd6„Z@ ee;¦  «         G d7„ d8ej4        ¦  «        ¦   «         ZA G d9„ d:e¦  «        ZBe( G d;„ d<e#¦  «        ¦   «         ZC G d=„ d>ej4        ¦  «        ZDe( G d?„ d@eC¦  «        ¦   «         ZE	 	 	 dJdBej<        eFej<                 z  dz  dCe=dz  d2ej<        dz  d+ej<        e=z  fdD„ZGe( G dE„ dFeCe¦  «        ¦   «         ZHg dG¢ZIdS )Ké    )ÚCallable)Ú	TypedDictN)Únné   )Úinitialization)ÚACT2FN)ÚCacheÚDynamicCache)ÚGenerationMixin)Úuse_experts_implementationÚuse_kernel_forward_from_hubÚuse_kernelized_func)Úcreate_causal_mask)ÚGradientCheckpointingLayer)ÚMoeCausalLMOutputWithPastÚMoeModelOutputWithPast)ÚROPE_INIT_FUNCTIONSÚdynamic_rope_update)ÚALL_ATTENTION_FUNCTIONSÚPreTrainedModel)ÚUnpack)ÚTransformersKwargsÚauto_docstring)Údeprecate_kwarg)Úcan_return_tupleÚmaybe_autocastÚmerge_with_config_defaults)Úcapture_outputsé   )ÚGraniteMoeSharedConfigc                   ód   — e Zd ZU dZej        ed<   ej        ed<   eed<   eed<   ej        ed<   dS )ÚGraniteFlashAttentionKwargsaT  
    Keyword arguments for advanced Flash Attention, causal-conv1d, and mamba_ssm kernel usage.
    Use cases include padding-free training and fewer `torch.compile` graph breaks.

    cu_seq_lens_q (`torch.LongTensor`):
        Gets cumulative sequence length for query state.
    cu_seq_lens_k (`torch.LongTensor`):
        Gets cumulative sequence length for key state.
    max_length_q (`int`):
        Maximum sequence length for query state.
    max_length_k (`int`):
        Maximum sequence length for key state.
    seq_idx (`torch.IntTensor):
        Index of each packed sequence.
    Úcu_seq_lens_qÚcu_seq_lens_kÚmax_length_qÚmax_length_kÚseq_idxN)	Ú__name__Ú
__module__Ú__qualname__Ú__doc__ÚtorchÚ
LongTensorÚ__annotations__ÚintÚ	IntTensor© ó    úŽ/root/private/project/pallet_detection_v2/.venv/lib/python3.11/site-packages/transformers/models/granitemoeshared/modeling_granitemoeshared.pyr"   r"   .   sb   € € € € € € ðð ð  Ô#Ð#Ð#Ñ#ØÔ#Ð#Ð#Ñ#ØÐÐÑØÐÐÑØŒ_ÐÐÑÐÐr2   r"   F)Útotalc                   óL   ‡ — e Zd ZdZdefˆ fd„Zdej        dej        fd„Zˆ xZ	S )ÚGraniteMoeSharedMLPz~
    MLP layer for shared experts

    Args:
        config:
            Configuration object with model hyperparameters.
    Úconfigc                 óD  •— t          ¦   «                              ¦   «          |j        | _        |j        | _        t
          |j                 | _        t          j	        | j        | j        dz  d¬¦  «        | _
        t          j	        | j        | j        d¬¦  «        | _        d S )Né   F©Úbias)ÚsuperÚ__init__Úhidden_sizeÚ
input_sizeÚshared_intermediate_sizer   Ú
hidden_actÚ
activationr   ÚLinearÚinput_linearÚoutput_linear©Úselfr7   Ú	__class__s     €r3   r=   zGraniteMoeSharedMLP.__init__O   s…   ø€ Ý‰Œ×ÒÑÔÐà Ô,ˆŒØ!Ô:ˆÔÝ  Ô!2Ô3ˆŒÝœI d¤o°tÔ7GÈ!Ñ7KÐRWÐXÑXÔXˆÔÝœY tÔ'7¸¼ÈuÐUÑUÔUˆÔÐÐr2   Úhidden_statesÚreturnc                 óÐ   — |                       |¦  «        }|                     dd¬¦  «        }|                      |d         ¦  «        |d         z  }|                      |¦  «        }|S )Nr9   éÿÿÿÿ©Údimr   r   )rD   ÚchunkrB   rE   )rG   rI   Úchunked_hidden_statess      r3   ÚforwardzGraniteMoeSharedMLP.forwardX   sj   € Ø×)Ò)¨-Ñ8Ô8ˆØ -× 3Ò 3°A¸2Ð 3Ñ >Ô >ÐØŸšÐ(=¸aÔ(@ÑAÔAÐDYÐZ[ÔD\Ñ\ˆØ×*Ò*¨=Ñ9Ô9ˆØÐr2   ©
r(   r)   r*   r+   r    r=   r,   ÚTensorrQ   Ú__classcell__©rH   s   @r3   r6   r6   F   s|   ø€ € € € € ðð ðVÐ5ð Vð Vð Vð Vð Vð Vð U¤\ð °e´lð ð ð ð ð ð ð ð r2   r6   ÚRMSNormc                   óT   ‡ — 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GraniteMoeSharedRMSNormç�íµ ÷Æ°>ÚepsrJ   Nc                 ó¬   •— t          ¦   «                              ¦   «          t          j        t	          j        |¦  «        ¦  «        | _        || _        dS )zF
        GraniteMoeSharedRMSNorm is equivalent to T5LayerNorm
        N)r<   r=   r   Ú	Parameterr,   ÚonesÚweightÚvariance_epsilon)rG   r>   rZ   rH   s      €r3   r=   z GraniteMoeSharedRMSNorm.__init__b   sD   ø€ õ 	‰Œ×ÒÑÔÐÝ”l¥5¤:¨kÑ#:Ô#:Ñ;Ô;ˆŒØ #ˆÔÐÐr2   rI   c                 ó  — |j         }|                     t          j        ¦  «        }|                     d¦  «                             dd¬¦  «        }|t          j        || j        z   ¦  «        z  }| j        |                     |¦  «        z  S )Nr9   rL   T)Úkeepdim)	ÚdtypeÚtor,   Úfloat32ÚpowÚmeanÚrsqrtr_   r^   )rG   rI   Úinput_dtypeÚvariances       r3   rQ   zGraniteMoeSharedRMSNorm.forwardj   s|   € Ø#Ô)ˆØ%×(Ò(­¬Ñ7Ô7ˆØ ×$Ò$ QÑ'Ô'×,Ò,¨R¸Ð,Ñ>Ô>ˆØ%­¬°H¸tÔ?TÑ4TÑ(UÔ(UÑUˆØŒ{˜]×-Ò-¨kÑ:Ô:Ñ:Ð:r2   c                 óH   — t          | j        j        ¦  «        › d| j        › �S )Nz, eps=)Útupler^   Úshaper_   )rG   s    r3   Ú
extra_reprz"GraniteMoeSharedRMSNorm.extra_reprq   s&   € Ý˜œÔ)Ñ*Ô*ÐIÐI°$Ô2GÐIÐIÐIr2   )rY   )
r(   r)   r*   Úfloatr=   r,   rS   rQ   rm   rT   rU   s   @r3   rX   rX   `   sŒ   ø€ € € € € ð$ð $¨ð $¸$ð $ð $ð $ð $ð $ð $ð; U¤\ð ;°e´lð ;ð ;ð ;ð ;ðJð Jð Jð Jð Jð Jð Jr2   rX   c                   ór   ‡ — e Zd ZdZdefˆ fd„Zdej        deej        ej        ej        f         fd„Z	ˆ xZ
S )ÚGraniteMoeSharedTopKRoutera¡  Top-k gating that returns the routing decisions without grouping tokens by expert.

    Returns ``(top_k_index, top_k_weights, router_logits)``; the grouping/scattering used to live
    here (via ``expert_size.tolist()``, which broke fullgraph compile) and now happens inside the
    experts forward via ``use_experts_implementation`` so the default ``grouped_mm`` / ``batched_mm``
    paths can compile cleanly.
    r7   c                 óä   •— t          ¦   «                              ¦   «          |j        | _        |j        | _        t          j        t          j	        | j        |j
        ¦  «        ¦  «        | _        d S ©N)r<   r=   Únum_local_expertsÚnum_expertsÚnum_experts_per_tokÚtop_kr   r\   r,   Úemptyr>   r^   rF   s     €r3   r=   z#GraniteMoeSharedTopKRouter.__init__~   sU   ø€ Ý‰Œ×ÒÑÔÐØ!Ô3ˆÔØÔ/ˆŒ
Ý”l¥5¤;¨tÔ/?ÀÔASÑ#TÔ#TÑUÔUˆŒˆˆr2   rI   rJ   c                 óô   — t          j        || j        ¦  «                             ¦   «         }|                     | j        d¬¦  «        \  }}t          j        |d¬¦  «                             |¦  «        }|||fS )NrL   rM   )	ÚFÚlinearr^   rn   Útopkrv   r,   ÚsoftmaxÚtype_as)rG   rI   Úrouter_logitsÚtop_k_logitsÚtop_k_indexÚtop_k_weightss         r3   rQ   z"GraniteMoeSharedTopKRouter.forward„   so   € Ýœ °´Ñ<Ô<×BÒBÑDÔDˆØ$1×$6Ò$6°t´zÀrÐ$6Ñ$JÔ$JÑ!ˆ�kÝœ l¸Ð;Ñ;Ô;×CÒCÀMÑRÔRˆØ˜M¨=Ð8Ð8r2   )r(   r)   r*   r+   r    r=   r,   rS   rk   rQ   rT   rU   s   @r3   rp   rp   u   sŽ   ø€ € € € € ðð ðVÐ5ð Vð Vð Vð Vð Vð Vð9 U¤\ð 9°e¸E¼LÈ%Ì,ÐX]ÔXdÐ<dÔ6eð 9ð 9ð 9ð 9ð 9ð 9ð 9ð 9r2   rp   c                   óh   ‡ — e Zd ZdZdefˆ fd„Zdej        dej        dej        dej        fd„Zˆ xZ	S )	ÚGraniteMoeSharedExpertsz2Collection of expert weights stored as 3D tensors.r7   c                 ó´  •— t          ¦   «                              ¦   «          |j        | _        |j        | _        |j        | _        t          j	        t          j        | j        d| j        z  | j        ¦  «        ¦  «        | _        t          j	        t          j        | j        | j        | j        ¦  «        ¦  «        | _        t          |j                 | _        d S )Nr9   )r<   r=   rs   rt   r>   Ú
hidden_dimÚintermediate_sizeÚintermediate_dimr   r\   r,   rw   Úgate_up_projÚ	down_projr   rA   Úact_fnrF   s     €r3   r=   z GraniteMoeSharedExperts.__init__�   s£   ø€ Ý‰Œ×ÒÑÔÐØ!Ô3ˆÔØ Ô,ˆŒØ &Ô 8ˆÔÝœL­¬°TÔ5EÀqÈ4ÔK`ÑG`ÐbfÔbqÑ)rÔ)rÑsÔsˆÔÝœ¥e¤k°$Ô2BÀDÄOÐUYÔUjÑ&kÔ&kÑlÔlˆŒÝ˜VÔ.Ô/ˆŒˆˆr2   rI   r€   r�   rJ   c                 ó€  — t          j        |¦  «        }t          j        ¦   «         5  t           j        j                             || j        ¬¦  «        }|                     ddd¦  «        }t          j        | 	                    d¬¦  «        d¦  «         
                    ¦   «         }d d d ¦  «         n# 1 swxY w Y   |D ]þ}|d         }|| j        k    rŒt          j        ||         ¦  «        \  }}	||	         }
t          j                             |
| j        |         ¦  «                             dd¬¦  «        \  }}|                      |¦  «        |z  }t          j                             || j        |         ¦  «        }|||	|d f         z  }|                     d|	|                     |j        ¦  «        ¦  «         Œÿ|S )N)Únum_classesr9   r   r   )rL   éþÿÿÿrM   rL   )r,   Ú
zeros_likeÚno_gradr   Ú
functionalÚone_hotrt   ÚpermuteÚgreaterÚsumÚnonzeroÚwhererz   rˆ   rO   rŠ   r‰   Ú
index_add_rc   rb   )rG   rI   r€   r�   Úfinal_hidden_statesÚexpert_maskÚ
expert_hitÚ
expert_idxÚ	top_k_posÚ	token_idxÚcurrent_stateÚgateÚupÚcurrent_hidden_statess                 r3   rQ   zGraniteMoeSharedExperts.forward˜   sø  € õ $Ô.¨}Ñ=Ô=ÐÝŒ]‰_Œ_ð 	Sð 	SÝœ(Ô-×5Ò5°kÈtÔO_Ð5Ñ`Ô`ˆKØ%×-Ò-¨a°°AÑ6Ô6ˆKÝœ {§¢¸8 Ñ'DÔ'DÀaÑHÔH×PÒPÑRÔRˆJð	Sð 	Sð 	Sñ 	Sô 	Sð 	Sð 	Sð 	Sð 	Sð 	Sð 	Søøøð 	Sð 	Sð 	Sð 	Sð
 %ð 
	nð 
	nˆJØ# AœˆJØ˜TÔ-Ò-Ð-ØÝ#(¤;¨{¸:Ô/FÑ#GÔ#GÑ ˆI�yØ)¨)Ô4ˆMÝ”}×+Ò+¨M¸4Ô;LÈZÔ;XÑYÔY×_Ò_Ð`aÐgiÐ_ÑjÔj‰HˆD�"Ø$(§K¢K°Ñ$5Ô$5¸Ñ$:Ð!Ý$&¤M×$8Ò$8Ð9NÐPTÔP^Ð_iÔPjÑ$kÔ$kÐ!Ø$9¸MÈ)ÐU^Ð`dÐJdÔ<eÑ$eÐ!Ø×*Ò*¨1¨iÐ9N×9QÒ9QÐReÔRkÑ9lÔ9lÑmÔmÐmÐmà"Ð"s   ¨A>B2Â2B6Â9B6rR   rU   s   @r3   rƒ   rƒ   ‹   s�   ø€ € € € € à<Ð<ð0Ð5ð 0ð 0ð 0ð 0ð 0ð 0ð#à”|ð#ð ”\ð#ð ”|ð	#ð
 
Œð#ð #ð #ð #ð #ð #ð #ð #r2   rƒ   c                   óL   ‡ — e Zd ZdZdefˆ fd„Zdej        dej        fd„Zˆ xZ	S )ÚGraniteMoeSharedMoEzISparsely-gated mixture-of-experts block: router decides, experts compute.r7   c                 ó°   •— t          ¦   «                              ¦   «          |j        | _        t	          |¦  «        | _        t          |¦  «        | _        d S rr   )r<   r=   r>   r?   rp   Úrouterrƒ   ÚexpertsrF   s     €r3   r=   zGraniteMoeSharedMoE.__init__¶   sE   ø€ Ý‰Œ×ÒÑÔÐØ Ô,ˆŒÝ0°Ñ8Ô8ˆŒÝ.¨vÑ6Ô6ˆŒˆˆr2   Úlayer_inputrJ   c                 óö   — |                      ¦   «         \  }}}|                     d|¦  «        }|                      |¦  «        \  }}}|                      |||¦  «        }	|	                     ||| j        ¦  «        S )NrL   )ÚsizeÚreshaper¥   r¦   Úviewr?   )
rG   r§   ÚbszÚlengthÚemb_sizerI   r€   r�   Ú_Úlayer_outputs
             r3   rQ   zGraniteMoeSharedMoE.forward¼   sv   € Ø +× 0Ò 0Ñ 2Ô 2ÑˆˆV�XØ#×+Ò+¨B°Ñ9Ô9ˆØ(,¯ª°MÑ(BÔ(BÑ%ˆ�] AØ—|’| M°;ÀÑNÔNˆØ× Ò   f¨d¬oÑ>Ô>Ð>r2   rR   rU   s   @r3   r£   r£   ³   sq   ø€ € € € € ØSÐSð7Ð5ð 7ð 7ð 7ð 7ð 7ð 7ð? 5¤<ð ?°E´Lð ?ð ?ð ?ð ?ð ?ð ?ð ?ð ?r2   r£   c                 óœ   — | 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..NrL   r9   rM   )rl   r,   Úcat)ÚxÚx1Úx2s      r3   Úrotate_halfr¶   Ä   s]   € à	
ˆ3Ð"�!”'˜"”+ Ñ"Ð"Ð"Ô	#€BØ	
ˆ3�”˜”˜qÑ Ð"Ð"Ð"Ô	#€BÝŒ9�r�c˜2�Y BÐ'Ñ'Ô'Ð'r2   Úrotary_pos_embc                 ó¾   — |                      |¦  «        }|                      |¦  «        }| |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kÚcosÚsinÚunsqueeze_dimÚq_embedÚk_embeds          r3   Úapply_rotary_pos_embrÁ   Ë   sc   € ð& �-Š-˜Ñ
&Ô
&€CØ
�-Š-˜Ñ
&Ô
&€CØ�3‰w�; q™>œ>¨CÑ/Ñ0€GØ�3‰w�; q™>œ>¨CÑ/Ñ0€GØ�GÐÐr2   rI   Ún_reprJ   c                 ó¸   — | j         \  }}}}|dk    r| S | dd…dd…ddd…dd…f                              |||||¦  «        } |                      |||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)rl   Úexpandrª   )rI   rÂ   ÚbatchÚnum_key_value_headsÚslenÚhead_dims         r3   Ú	repeat_kvrÉ   å   s„   € ð
 2?Ô1DÑ.€EÐ  hØ�‚z€zØÐØ! ! ! ! Q Q Q¨¨a¨a¨a°°°Ð"2Ô3×:Ò:¸5ÐBUÐW\Ð^bÐdlÑmÔm€MØ× Ò  Ð(;¸eÑ(CÀTÈ8ÑTÔTÐTr2   ç        ÚmoduleÚqueryÚkeyÚvalueÚattention_maskÚscalingÚdropoutÚkwargsc                 ó  — t          || j        ¦  «        }t          || j        ¦  «        }	t          j        ||                     dd¦  «        ¦  «        |z  }
|�|
|z   }
t
          j                             |
dt          j        ¬¦  «         	                    |j
        ¦  «        }
t
          j                             |
|| j        ¬¦  «        }
t          j        |
|	¦  «        }|                     dd¦  «                             ¦   «         }||
fS )Nr9   r   rL   )rN   rb   )ÚpÚtrainingr   )rÉ   Únum_key_value_groupsr,   ÚmatmulÚ	transposer   r�   r|   rd   rc   rb   rÑ   rÕ   Ú
contiguous)rË   rÌ   rÍ   rÎ   rÏ   rÐ   rÑ   rÒ   Ú
key_statesÚvalue_statesÚattn_weightsÚattn_outputs               r3   Úeager_attention_forwardrÞ   ñ   sé   € õ ˜3 Ô ;Ñ<Ô<€JÝ˜U FÔ$?Ñ@Ô@€Lå”<  z×';Ò';¸A¸qÑ'AÔ'AÑBÔBÀWÑL€LØÐ!Ø# nÑ4ˆå”=×(Ò(¨¸2ÅUÄ]Ð(ÑSÔS×VÒVÐW\ÔWbÑcÔc€LÝ”=×(Ò(¨¸È6Ì?Ð(Ñ[Ô[€LÝ”,˜|¨\Ñ:Ô:€KØ×'Ò'¨¨1Ñ-Ô-×8Ò8Ñ:Ô:€Kà˜Ð$Ð$r2   c                   óÎ   ‡ — 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z  dej        dz  d	e
dz  d
ee         de	ej        ej        f         fd„Zˆ xZS )ÚGraniteMoeSharedAttentionz=Multi-headed attention from 'Attention Is All You Need' paperr7   Ú	layer_idxc                 ó¨  •— t          ¦   «                              ¦   «          || _        || _        t	          |d|j        |j        z  ¦  «        | _        |j        |j        z  | _	        |j
        | _        |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        ¬¦  «        | _        d S )NrÈ   Tr:   )r<   r=   r7   rá   Úgetattrr>   Únum_attention_headsrÈ   rÆ   rÖ   Úattention_multiplierrÐ   Úattention_dropoutÚ	is_causalr   rC   Úattention_biasÚq_projÚk_projÚv_projÚo_proj©rG   r7   rá   rH   s      €r3   r=   z"GraniteMoeSharedAttention.__init__  s>  ø€ Ý‰Œ×ÒÑÔÐØˆŒØ"ˆŒÝ ¨
°FÔ4FÈ&ÔJdÑ4dÑeÔeˆŒØ$*Ô$>À&ÔB\Ñ$\ˆÔ!ØÔ2ˆŒØ!'Ô!9ˆÔØˆŒå”iØÔ Ô :¸T¼]Ñ JÐQWÔQfð
ñ 
ô 
ˆŒõ ”iØÔ Ô :¸T¼]Ñ JÐQWÔQfð
ñ 
ô 
ˆŒõ ”iØÔ Ô :¸T¼]Ñ JÐQWÔQfð
ñ 
ô 
ˆŒõ ”iØÔ&¨¬Ñ6¸Ô8JÐQWÔQfð
ñ 
ô 
ˆŒˆˆr2   NrI   Úposition_embeddingsrÏ   Úpast_key_valuesrÒ   rJ   c                 ó"  — |j         d d…         }g |¢d‘| j        ‘R }|                      |¦  «                             |¦  «                             dd¦  «        }|                      |¦  «                             |¦  «                             dd¦  «        }	|                      |¦  «                             |¦  «                             dd¦  «        }
|\  }}t          ||	||¦  «        \  }}	|�|                     |	|
| j	        ¦  «        \  }	}
t          j        | j        j        t          ¦  «        } || ||	|
|f| j        sdn| j        | j        dœ|¤Ž\  }} |j        g |¢d‘R Ž                      ¦   «         }|                      |¦  «        }||fS )NrL   r   r9   rÊ   )rÑ   rÐ   )rl   rÈ   ré   r«   rØ   rê   rë   rÁ   Úupdaterá   r   Úget_interfacer7   Ú_attn_implementationrÞ   rÕ   ræ   rÐ   rª   rÙ   rì   )rG   rI   rî   rÏ   rï   rÒ   Úinput_shapeÚhidden_shapeÚquery_statesrÚ   rÛ   r¼   r½   Úattention_interfacerÝ   rÜ   s                   r3   rQ   z!GraniteMoeSharedAttention.forward%  sÀ  € ð $Ô)¨#¨2¨#Ô.ˆØ8˜Ð8 bÐ8¨$¬-Ð8Ð8ˆà—{’{ =Ñ1Ô1×6Ò6°|ÑDÔD×NÒNÈqÐRSÑTÔTˆØ—[’[ Ñ/Ô/×4Ò4°\ÑBÔB×LÒLÈQÐPQÑRÔRˆ
Ø—{’{ =Ñ1Ô1×6Ò6°|ÑDÔD×NÒNÈqÐRSÑTÔTˆà&‰ˆˆSÝ#7¸ÀjÐRUÐWZÑ#[Ô#[Ñ ˆ�jàÐ&Ø'6×'=Ò'=¸jÈ,ÐX\ÔXfÑ'gÔ'gÑ$ˆJ˜å(?Ô(MØŒKÔ,Õ.Eñ)
ô )
Ðð %8Ð$7ØØØØØð	%
ð  $œ}ÐH�C�C°$Ô2HØ”Lð	%
ð 	%
ð ð	%
ð 	%
Ñ!ˆ�\ð *�kÔ)Ð;¨;Ð;¸Ð;Ð;Ð;×FÒFÑHÔHˆØ—k’k +Ñ.Ô.ˆØ˜LÐ(Ð(r2   )NNN)r(   r)   r*   r+   r    r/   r=   r,   rS   rk   r	   r   r   rQ   rT   rU   s   @r3   rà   rà   
  sæ   ø€ € € € € àGÐGð
Ð5ð 
À#ð 
ð 
ð 
ð 
ð 
ð 
ð4 IMØ.2Ø(,ð&)ð &)à”|ð&)ð # 5¤<°´Ð#=Ô>ÀÑEð&)ð œ tÑ+ð	&)ð
  ™ð&)ð Ð+Ô,ð&)ð 
ˆuŒ|˜Uœ\Ð)Ô	*ð&)ð &)ð &)ð &)ð &)ð &)ð &)ð &)r2   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
dz  deej        ej        f         dz  dee         deej        eej        ej        f         dz  f         fd„Zˆ xZS )ÚGraniteMoeSharedDecoderLayerr7   rá   c                 óŽ  •— t          ¦   «                              ¦   «          |j        | _        t          ||¬¦  «        | _        t          |j        |j        ¬¦  «        | _        t          |j        |j        ¬¦  «        | _        t          |¦  «        | _
        |j        | _        |j        dk    rd nt          |¦  «        | _        d S )N)r7   rá   ©rZ   r   )r<   r=   r>   rà   Ú	self_attnrX   Úrms_norm_epsÚinput_layernormÚpost_attention_layernormr£   Úblock_sparse_moeÚresidual_multiplierr@   r6   Ú
shared_mlprí   s      €r3   r=   z%GraniteMoeSharedDecoderLayer.__init__O  s°   ø€ Ý‰Œ×ÒÑÔÐØ!Ô-ˆÔÝ2¸&ÈIÐVÑVÔVˆŒÝ6°vÔ7IÈvÔObÐcÑcÔcˆÔÝ(?ÀÔ@RÐX^ÔXkÐ(lÑ(lÔ(lˆÔ%Ý 3°FÑ ;Ô ;ˆÔØ#)Ô#=ˆÔ Ø"(Ô"AÀQÒ"FÐ"F˜$˜$ÕL_Ð`fÑLgÔLgˆŒˆˆr2   NFrI   rÏ   Úposition_idsrï   Úoutput_attentionsÚ	use_cacherî   rÒ   rJ   c                 ó4  — |}	|                       |¦  «        } | j        d|||||||dœ|¤Ž\  }}
|	|| j        z  z   }|}	|                      |¦  «        }|                      |¦  «        }| j        €|}n||                      |¦  «        z   }|	|| j        z  z   }|S )N)rI   rÏ   r  rï   r  r  rî   r1   )rþ   rü   r  rÿ   r   r  )rG   rI   rÏ   r  rï   r  r  rî   rÒ   Úresidualr¯   Úmoe_hidden_statess               r3   rQ   z$GraniteMoeSharedDecoderLayer.forwardY  sÜ   € ð !ˆØ×,Ò,¨]Ñ;Ô;ˆð *˜4œ>ð 	
Ø'Ø)Ø%Ø+Ø/ØØ 3ð	
ð 	
ð ð	
ð 	
Ñˆ�qð ! =°4Ô3KÑ#KÑKˆà ˆØ×5Ò5°mÑDÔDˆØ ×1Ò1°-Ñ@Ô@ÐàŒ?Ð"Ø-ˆMˆMà-°·²ÀÑ0NÔ0NÑNˆMØ  =°4Ô3KÑ#KÑKˆØÐr2   )NNNFFN)r(   r)   r*   r    r/   r=   r,   rS   r-   r	   Úboolrk   r   r"   ÚFloatTensorrQ   rT   rU   s   @r3   rù   rù   N  s2  ø€ € € € € ðhÐ5ð hÀ#ð hð hð hð hð hð hð /3Ø04Ø(,Ø).Ø!&ØHLð%ð %à”|ð%ð œ tÑ+ð%ð Ô&¨Ñ-ð	%ð
  ™ð%ð   $™;ð%ð ˜$‘;ð%ð # 5¤<°´Ð#=Ô>ÀÑEð%ð Ð4Ô5ð%ð 
ˆuÔ  %¨Ô(9¸5Ô;LÐ(LÔ"MÐPTÑ"TÐTÔ	Uð%ð %ð %ð %ð %ð %ð %ð %r2   rù   c                   ó†   ‡ — 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 ej        ¦   «         ˆ fd„¦   «         Zˆ xZS )ÚGraniteMoeSharedPreTrainedModelr7   ÚmodelTrù   rï   )rI   Ú
attentionsc                 óŠ  •— t          ¦   «                              |¦  «         t          |t          ¦  «        rNt	          j        |j        d| j        j        ¬¦  «         t	          j        |j	        d| j        j        ¬¦  «         d S t          |t          ¦  «        r(t	          j        |j        d| j        j        ¬¦  «         d S d S )NrÊ   )rf   Ústd)r<   Ú_init_weightsÚ
isinstancerƒ   ÚinitÚnormal_rˆ   r7   Úinitializer_ranger‰   rp   r^   )rG   rË   rH   s     €r3   r  z-GraniteMoeSharedPreTrainedModel._init_weights’  s·   ø€ å‰Œ×Ò˜fÑ%Ô%Ð%Ý�fÕ5Ñ6Ô6ð 	UÝŒL˜Ô,°3¸D¼KÔ<YÐZÑZÔZÐZÝŒL˜Ô)°¸¼Ô9VÐWÑWÔWÐWÐWÐWÝ˜Õ :Ñ;Ô;ð 	UÝŒL˜œ¨S°d´kÔ6SÐTÑTÔTÐTÐTÐTð	Uð 	Ur2   )r(   r)   r*   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_backendrù   rà   Ú_can_record_outputsr,   r�   r  rT   rU   s   @r3   r  r  �  s±   ø€ € € € € € à"Ð"Ð"Ñ"ØÐØ&*Ð#Ø7Ð8ÐØ#4Ð"5ÐØÐØ€NØÐØ!ÐØ"&Ðà5Ø/ðð Ðð
 €U„]�_„_ðUð Uð Uð Uñ „_ðUð Uð Uð Uð Ur2   r  c            	       óð   ‡ — e Zd Z edd¬¦  «        d
defˆ fd„¦   «         Ze edd¬¦  «        	 d
dedeej	        e
f         fd„¦   «         ¦   «         Z ej        ¦   «         ed	„ ¦   «         ¦   «         Zˆ xZS )ÚGraniteMoeSharedRotaryEmbeddingÚdevicez5.18)ÚversionNr7   c                 ó¾  •— t          ¦   «                              ¦   «          |j        | _        |j        | _        || _        | j        j        d         | _        | j        }| j        dk    rt          | j                 } || j        |¦  «        \  }| _
        t          j        |d¬¦  «        | _        t          j        |                     ¦   «         d¬¦  «        | _        d S )NÚ	rope_typeÚdefaultF)Ú
persistent)r<   r=   Úmax_position_embeddingsÚmax_seq_len_cachedÚoriginal_max_seq_lenr7   Úrope_parametersr%  Úcompute_default_rope_parametersr   Úattention_scalingr   ÚBufferÚinv_freqÚcloneÚoriginal_inv_freq)rG   r7   r"  Úrope_init_fnr/  rH   s        €r3   r=   z(GraniteMoeSharedRotaryEmbedding.__init__�  sÀ   ø€ å‰Œ×ÒÑÔÐØ"(Ô"@ˆÔØ$*Ô$BˆÔ!àˆŒàœÔ4°[ÔAˆŒØ!%Ô!EˆØŒ>˜YÒ&Ð&Ý.¨t¬~Ô>ˆLØ+7¨<¸¼ÀVÑ+LÔ+LÑ(ˆ�$Ô(åœ	 (°uÐ=Ñ=Ô=ˆŒÝ!#¤¨8¯>ª>Ñ+;Ô+;ÈÐ!NÑ!NÔ!NˆÔÐÐr2   rJ   c                 óä   — | j         d         }t          | dd¦  «        p| j        | j        z  }d}d|t	          j        d|dt          j        ¬¦  «        |z  z  z  }|                     |¦  «        |fS )a¾  
        Computes the inverse frequencies according to the original RoPE implementation
        Args:
            config ([`~transformers.PreTrainedConfig`]):
                The model configuration.
        Returns:
            Tuple of (`torch.Tensor`, `float`), containing the inverse frequencies for the RoPE embeddings and the
            post-processing scaling factor applied to the computed cos/sin (unused in this type of RoPE).
        Ú
rope_thetarÈ   Ng      ð?r   r9   ©rb   )r+  rã   r>   rä   r,   Úarangern   rc   )r7   r"  rÒ   ÚbaserN   Úattention_factorr/  s          r3   r,  z?GraniteMoeSharedRotaryEmbedding.compute_default_rope_parameters®  sz   € ð Ô% lÔ3ˆÝ�f˜j¨$Ñ/Ô/Ðc°6Ô3EÈÔIcÑ3cˆàÐà˜$¥5¤<°°3¸ÅÄÐ#MÑ#MÔ#MÐPSÑ#SÑTÑUˆØ�{Š{˜6Ñ"Ô"Ð$4Ð4Ð4r2   c                 óú  — | j         d d d …d f                              |j        d         dd¦  «                             t          j        |j        ¬¦  «        }|d d …d d d …f                              ¦   «         }t          |j        j        t          ¦  «        r|j        j        dk    r|j        j        nd}t          |d¬¦  «        5  ||z                       dd	¦  «        }t	          j        ||fd¬
¦  «        }|                     ¦   «         | j        z  }|                     ¦   «         | j        z  }	d d d ¦  «         n# 1 swxY w Y   |                     |j        ¬¦  «        |	                     |j        ¬¦  «        fS )Nr   rL   r   ©rb   r"  ÚmpsÚcpuF)Údevice_typeÚenabledr9   rM   r5  )r/  rÄ   rl   rc   r,   rn   r"  r  ÚtypeÚstrr   rØ   r²   r¼   r-  r½   rb   )
rG   r³   r  Úinv_freq_expandedÚposition_ids_expandedr=  ÚfreqsÚembr¼   r½   s
             r3   rQ   z'GraniteMoeSharedRotaryEmbedding.forwardÄ  s›  € ð ŒM˜$    4˜-Ô(×/Ò/°Ô0BÀ1Ô0EÀrÈ1ÑMÔM×PÒPÕW\ÔWbÐklÔksÐPÑtÔtð 	ð !-¨Q¨Q¨Q°°a°a°a¨ZÔ 8× >Ò >Ñ @Ô @Ðå'1°!´(´-ÅÑ'EÔ'EÐkÈ!Ì(Ì-Ð[`ÒJ`ÐJ`�a”h”m�mÐfkˆå¨¸UÐCÑCÔCð 	5ð 	5Ø&Ð)>Ñ>×IÒIÈ!ÈQÑOÔOˆEÝ”)˜U E˜N°Ð3Ñ3Ô3ˆCØ—'’'‘)”)˜dÔ4Ñ4ˆCØ—'’'‘)”)˜dÔ4Ñ4ˆCð		5ð 	5ð 	5ñ 	5ô 	5ð 	5ð 	5ð 	5ð 	5ð 	5ð 	5øøøð 	5ð 	5ð 	5ð 	5ð �vŠv˜AœGˆvÑ$Ô$ c§f¢f°1´7 fÑ&;Ô&;Ð;Ð;s   ÃA*D<Ä<E ÅE rr   )r(   r)   r*   r   r    r=   Ústaticmethodrk   r,   rS   rn   r,  r�   r   rQ   rT   rU   s   @r3   r!  r!  œ  sø   ø€ € € € € Ø€_�X vÐ.Ñ.Ô.ðOð OÐ5ð Oð Oð Oð Oð Oñ /Ô.ðOð  Ø€_�X vÐ.Ñ.Ô.à/3ð5ð 5Ø&ð5à	ˆuŒ|˜UÐ"Ô	#ð5ð 5ð 5ñ /Ô.ñ „\ð5ð( €U„]�_„_Øð<ð <ñ Ôñ „_ð<ð <ð <ð <ð <r2   r!  c                   óâ   ‡ — 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GraniteMoeSharedModelr7   c                 óö  •‡— t          ¦   «                              ‰¦  «         ‰j        | _        ‰j        | _        t          j        ‰j        ‰j        | j        ¦  «        | _        t          j	        ˆfd„t          ‰j        ¦  «        D ¦   «         ¦  «        | _        t          ‰j        ‰j        ¬¦  «        | _        t!          ‰¬¦  «        | _        d| _        ‰j        | _        |                      ¦   «          d S )Nc                 ó0   •— g | ]}t          ‰|¦  «        ‘ŒS r1   )rù   )Ú.0rá   r7   s     €r3   ú
<listcomp>z2GraniteMoeSharedModel.__init__.<locals>.<listcomp>à  s$   ø€ ÐnÐnÐnÀÕ)¨&°)Ñ<Ô<ÐnÐnÐnr2   rû   ©r7   F)r<   r=   Úpad_token_idÚpadding_idxÚ
vocab_sizer   Ú	Embeddingr>   Úembed_tokensÚ
ModuleListÚrangeÚnum_hidden_layersÚlayersrX   rý   Únormr!  Ú
rotary_embÚgradient_checkpointingÚembedding_multiplierÚ	post_initrF   s    `€r3   r=   zGraniteMoeSharedModel.__init__Ù  sà   øø€ Ý‰Œ×Ò˜Ñ Ô Ð Ø!Ô.ˆÔØ Ô+ˆŒåœL¨Ô):¸FÔ<NÐPTÔP`ÑaÔaˆÔÝ”mØnÐnÐnÐnÍeÐTZÔTlÑNmÔNmÐnÑnÔnñ
ô 
ˆŒõ ,¨FÔ,>ÀFÔDWÐXÑXÔXˆŒ	Ý9ÀÐHÑHÔHˆŒØ&+ˆÔ#Ø$*Ô$?ˆÔ!ð 	�ŠÑÔÐÐÐr2   NÚ	input_idsrÏ   r  rï   Úinputs_embedsr  rÒ   rJ   c           
      óZ  — |d u |d uz  rt          d¦  «        ‚|r|€t          | j        ¬¦  «        }|€|                      |¦  «        }|€V|�|                     ¦   «         nd}t          j        |j        d         |j        ¬¦  «        |z   }| 	                    d¦  «        }t          | j        ||||¬¦  «        }	|| j        z  }|}
|                      |
|¦  «        }| j        d | j        j        …         D ]} ||
f||	|||dœ|¤Ž}
Œ|                      |
¦  «        }
t!          |
|¬¦  «        S )	Nz:You must specify exactly one of input_ids or inputs_embedsrL  r   r   )r"  )r7   r\  rÏ   rï   r  )rî   rÏ   r  rï   r  )Úlast_hidden_staterï   )Ú
ValueErrorr
   r7   rQ  Úget_seq_lengthr,   r6  rl   r"  r¹   r   rY  rW  rU  rT  rV  r   )rG   r[  rÏ   r  rï   r\  r  rÒ   Úpast_seen_tokensÚcausal_maskrI   rî   Údecoder_layers                r3   rQ   zGraniteMoeSharedModel.forwardê  s—  € ð ˜Ð -°tÐ";Ñ<ð 	[ÝÐYÑZÔZÐZàð 	?˜Ð0Ý*°$´+Ð>Ñ>Ô>ˆOàÐ Ø ×-Ò-¨iÑ8Ô8ˆMàÐØCRÐC^˜×=Ò=Ñ?Ô?Ð?ÐdeÐÝ œ<¨Ô(;¸AÔ(>À}ÔG[Ð\Ñ\Ô\Ð_oÑoˆLØ'×1Ò1°!Ñ4Ô4ˆLå(Ø”;Ø'Ø)Ø+Ø%ð
ñ 
ô 
ˆð &¨Ô(AÑAˆØ%ˆð #Ÿošo¨m¸\ÑJÔJÐà!œ[Ð)H¨4¬;Ô+HÐ)HÔIð 		ð 		ˆMØ)˜MØðà$7Ø*Ø)Ø /Ø#ðð ð ðð ˆMˆMð Ÿ	š	 -Ñ0Ô0ˆå%Ø+Ø+ð
ñ 
ô 
ð 	
r2   )NNNNNN)r(   r)   r*   r    r=   r   r   r   r,   r-   rS   r	   r
  r	  r   r   r   rQ   rT   rU   s   @r3   rG  rG  ×  s  ø€ € € € € ðÐ5ð ð ð ð ð ð ð"  ØØð .2Ø.2Ø04Ø(,Ø26Ø!%ð5
ð 5
àÔ# dÑ*ð5
ð œ tÑ+ð5
ð Ô&¨Ñ-ð	5
ð
  ™ð5
ð Ô(¨4Ñ/ð5
ð ˜$‘;ð5
ð Ð+Ô,ð5
ð 
 ð5
ð 5
ð 5
ñ „^ñ „_ñ  Ôð5
ð 5
ð 5
ð 5
ð 5
r2   rG  r9   Úgate_logitsrt   c           	      óÎ  — | �t          | t          ¦  «        sdS | d         j        }t          j        |t          j        |¬¦  «        }t          j        |t          j        |¬¦  «        }d}|�4|                     d¦  «                             |t          j        ¬¦  «        }| D �]‰}	t          j        j	         
                    |	                     |¦  «        d¬¦  «        }
t          j        |
|d¬¦  «        \  }}|€z|t          j        |                     d¦  «        |¬¦  «                             ¦   «         z   }||
                     ¦   «                              d¬¦  «        z   }||
j        d         z   }ŒÒ|t          j        |t          j        |¬¦  «                             d|                     d¦  «        |                     |¦  «        ¦  «        z   }||
                     ¦   «         |                     d¦  «        z                       d¬¦  «        z   }||                     ¦   «         z   }�Œ‹||z  }||z  }t          j        ||                     d¦  «        z  ¦  «        }||z  S )	aÄ  
    Computes auxiliary load balancing loss as in Switch Transformer - implemented in Pytorch.

    See Switch Transformer (https://huggingface.co/papers/2101.03961) for more details. This function implements the loss
    function presented in equations (4) - (6) of the paper. It aims at penalizing cases where the routing between
    experts is too unbalanced.

    Args:
        gate_logits:
            Logits from the `gate`, should be a tuple of model.config.num_hidden_layers tensors of
            shape [batch_size X sequence_length, num_experts].
        num_experts:
            Number of experts
        top_k:
            The number of experts to route per-token, can be also interpreted as the `top-k` routing
            parameter.
        attention_mask (`torch.Tensor`, *optional*):
            The attention_mask used in forward function
            shape [batch_size X sequence_length] if not None.

    Returns:
        The auxiliary loss.
    Nr   r:  rÊ   rL   )r"  rb   rM   )Ú	minlength)r  rk   r"  r,   Úzerosrd   rª   rc   r   r�   r|   r{   Úbincountrn   r”   rl   Úscatter_add_Úrepeat_interleaver¹   )rd  rt   rv   rÏ   Úcompute_deviceÚtokens_per_expert_sumÚrouter_prob_sumÚ
total_rowsÚ	flat_maskÚ
layer_gateÚrouting_weightsr¯   Úselected_expertsÚtokens_per_expertÚrouter_prob_per_expertÚoverall_losss                   r3   Úload_balancing_loss_funcrv  %  sV  € ð: Ð¥*¨[½%Ñ"@Ô"@ÐØˆqð ! ”^Ô*€NÝ!œK¨½5¼=ÐQ_Ð`Ñ`Ô`ÐÝ”k +µU´]È>ÐZÑZÔZ€OØ€JàÐ!à"×*Ò*¨2Ñ.Ô.×1Ò1¸ÍuÌ}Ð1Ñ]Ô]ˆ	à!ð 6ñ 6ˆ
Ýœ(Ô-×5Ò5°j·m²mÀNÑ6SÔ6SÐY[Ð5Ñ\Ô\ˆÝ#œj¨¸%ÀRÐHÑHÔHÑˆÐØÐ!ð &­¬Ð7G×7OÒ7OÐPRÑ7SÔ7SÐ_jÐ(kÑ(kÔ(k×(qÒ(qÑ(sÔ(sÑsð "ð .°×0EÒ0EÑ0GÔ0G×0KÒ0KÐPQÐ0KÑ0RÔ0RÑRˆOØ# oÔ&;¸AÔ&>Ñ>ˆJˆJð %:½E¼KØ¥5¤=¸ð=ñ =ô =çŠl˜1Ð.×6Ò6°rÑ:Ô:¸I×<WÒ<WÐX]Ñ<^Ô<^Ñ_Ô_ñ%`Ð!ð .°×1FÒ1FÑ1HÔ1HÈ9×K^ÒK^Ð_aÑKbÔKbÑ1b×0gÒ0gÐlmÐ0gÑ0nÔ0nÑnˆOØ# i§m¢m¡o¤oÑ5ˆJ‰Jà-°
Ñ:ÐØ,¨zÑ9Ðå”9Ð.Ð1G×1QÒ1QÐRSÑ1TÔ1TÑTÑUÔU€LØ˜+Ñ%Ð%r2   c                   ó  ‡ — e Zd ZddiZddiZddgdgfiZddiZdef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z  fd„¦   «         ¦   «         Zˆ xZS )ÚGraniteMoeSharedForCausalLMzlm_head.weightzmodel.embed_tokens.weightÚlm_headÚcolwise_gather_outputrI   ÚlogitsÚkeep_full_weightr7   c                 ó^  •— t          ¦   «                              |¦  «         t          |¦  «        | _        |j        | _        t          j        |j        |j        d¬¦  «        | _        |j	        | _	        |j
        | _        |j        | _        |j        | _        |                      ¦   «          d S )NFr:   )r<   r=   rG  r  rO  r   rC   r>   ry  Úrouter_aux_loss_coefrs   rt   ru   Úlogits_scalingrZ  rF   s     €r3   r=   z$GraniteMoeSharedForCausalLM.__init__q  s–   ø€ Ý‰Œ×Ò˜Ñ Ô Ð Ý*¨6Ñ2Ô2ˆŒ
Ø Ô+ˆŒÝ”y Ô!3°VÔ5FÈUÐSÑSÔSˆŒØ$*Ô$?ˆÔ!Ø!Ô3ˆÔØ#)Ô#=ˆÔ Ø$Ô3ˆÔð 	�ŠÑÔÐÐÐr2   Nr   r[  rÏ   r  rï   r\  ÚlabelsÚoutput_router_logitsÚlogits_to_keeprJ   c	           	      ó2  — |�|n| j         j        } | j        d|||||dœ|	¤Ž}
|
j        }t	          |t
          ¦  «        rt          | d¦  «        n|}|                      |dd…|dd…f         ¦  «        }|| j         j        z  }d}|� | j	        ||fd| j         j
        i|	¤Ž}d}|rHt          |
j        | j        | j        |¦  «        }|�%|| j        |                     |j        ¦  «        z  z  }t%          ||||
j        |
j        |
j        |
j        ¬¦  «        S )ax  
        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, GraniteMoeSharedForCausalLM

        >>> model = GraniteMoeSharedForCausalLM.from_pretrained("ibm/PowerMoE-3b")
        >>> tokenizer = AutoTokenizer.from_pretrained("ibm/PowerMoE-3b")

        >>> 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)r[  rÏ   r  rï   r\  rO  )ÚlossÚaux_lossr{  rï   rI   r  r~   r1   )r7   r�  r  r^  r  r/   Úslicery  r  Úloss_functionrO  rv  r~   rt   ru   r~  rc   r"  r   rï   rI   r  )rG   r[  rÏ   r  rï   r\  r€  r�  r‚  rÒ   ÚoutputsrI   Úslice_indicesr{  r„  r…  s                   r3   rQ   z#GraniteMoeSharedForCausalLM.forward~  s‘  € ðJ %9Ð$DÐ Ð È$Ì+ÔJjð 	ð �$”*ð 
ØØ)Ø%Ø+Ø'ð
ð 
ð ð
ð 
ˆð  Ô1ˆÝ8BÀ>ÕSVÑ8WÔ8WÐk�˜~˜o¨tÑ4Ô4Ð4Ð]kˆØ—’˜m¨A¨A¨A¨}¸a¸a¸aÐ,?Ô@ÑAÔAˆØ˜$œ+Ô4Ñ4ˆàˆØÐà%�4Ô%ØØðð ð  œ;Ô1ðð ð	ð ˆDð ˆØð 	MÝ/ØÔ%ØÔ ØÔ(Øñ	ô ˆHð Ð!Ø˜Ô1°H·K²KÀÄÑ4LÔ4LÑLÑL�Ý(ØØØØ#Ô3Ø!Ô/ØÔ)Ø!Ô/ð
ñ 
ô 
ð 	
r2   )NNNNNNNr   )r(   r)   r*   Ú_tied_weights_keysÚ_tp_planÚ_pp_planÚ
_fsdp_planr    r=   r   r   r,   r-   rS   r	   r
  r	  r/   rk   r   rQ   rT   rU   s   @r3   rx  rx  j  sk  ø€ € € € € à*Ð,GÐHÐØÐ2Ð3€HØ˜_Ð-°¨zÐ:Ð;€HØÐ/Ð0€JðÐ5ð ð ð ð ð ð ð Øð .2Ø.2Ø04Ø(,Ø26Ø*.Ø,0Ø-.ðQ
ð Q
àÔ# dÑ*ðQ
ð œ tÑ+ðQ
ð Ô&¨Ñ-ð	Q
ð
  ™ðQ
ð Ô(¨4Ñ/ðQ
ð Ô  4Ñ'ðQ
ð # T™kðQ
ð ˜eœlÑ*ðQ
ð 
Ð*Ñ	*ðQ
ð Q
ð Q
ñ Ôñ „^ðQ
ð Q
ð Q
ð Q
ð Q
r2   rx  )rx  rG  r  )r   )rÊ   )Nr9   N)JÚcollections.abcr   Útypingr   r,   Útorch.nn.functionalr   r�   ry   Ú r   r  Úactivationsr   Úcache_utilsr	   r
   Ú
generationr   Úintegrationsr   r   r   Úmasking_utilsr   Úmodeling_layersr   Úmodeling_outputsr   r   Úmodeling_rope_utilsr   r   Úmodeling_utilsr   r   Úprocessing_utilsr   Úutilsr   r   Úutils.deprecationr   Úutils.genericr   r   r   Úutils.output_capturingr   Úconfiguration_granitemoesharedr    r"   ÚModuler6   rX   rp   rƒ   r£   r¶   rÁ   rS   r/   rÉ   rn   rÞ   rà   rù   r  r!  rG  rk   rv  rx  Ú__all__r1   r2   r3   ú<module>r£     sÑ  ðð* %Ð $Ð $Ð $Ð $Ð $Ø Ð Ð Ð Ð Ð à €€€Ø Ð Ð Ð Ð Ð Ð Ð Ð Ø Ð Ð Ð Ð Ð à &Ð &Ð &Ð &Ð &Ð &Ø !Ð !Ð !Ð !Ð !Ð !Ø .Ð .Ð .Ð .Ð .Ð .Ð .Ð .Ø )Ð )Ð )Ð )Ð )Ð )Ø hÐ hÐ hÐ hÐ hÐ hÐ hÐ hÐ hÐ hØ /Ð /Ð /Ð /Ð /Ð /Ø 9Ð 9Ð 9Ð 9Ð 9Ð 9Ø QÐ QÐ QÐ QÐ QÐ QÐ QÐ QØ KÐ KÐ KÐ KÐ KÐ KÐ KÐ KØ FÐ FÐ FÐ FÐ FÐ FÐ FÐ FØ &Ð &Ð &Ð &Ð &Ð &Ø 7Ð 7Ð 7Ð 7Ð 7Ð 7Ð 7Ð 7Ø 0Ð 0Ð 0Ð 0Ð 0Ð 0Ø YÐ YÐ YÐ YÐ YÐ YÐ YÐ YÐ YÐ YØ 5Ð 5Ð 5Ð 5Ð 5Ð 5Ø BÐ BÐ BÐ BÐ BÐ Bðð ð ð ð  )°5ð ñ ô ð ð0ð ð ð ð ˜"œ)ñ ô ð ð4 Ð˜YÑ'Ô'ðJð Jð Jð Jð J˜bœiñ Jô Jñ (Ô'ðJð(9ð 9ð 9ð 9ð 9 ¤ñ 9ô 9ð 9ð, ð$#ð $#ð $#ð $#ð $#˜bœiñ $#ô $#ñ Ôð$#ðN?ð ?ð ?ð ?ð ?˜"œ)ñ ?ô ?ð ?ð"(ð (ð (ð ÐÐ-Ñ.Ô.ðð ð ñ /Ô.ðð2	U˜Uœ\ð 	U°#ð 	U¸%¼,ð 	Uð 	Uð 	Uð 	Uð& ð%ð %ØŒIð%àŒ<ð%ð 
Œð%ð Œ<ð	%ð
 ”L 4Ñ'ð%ð ð%ð ð%ð Ð'Ô(ð%ð %ð %ð %ð2 ÐÐ)Ñ*Ô*ð@)ð @)ð @)ð @)ð @) ¤	ñ @)ô @)ñ +Ô*ð@)ðF0ð 0ð 0ð 0ð 0Ð#=ñ 0ô 0ð 0ðf ðUð Uð Uð Uð U oñ Uô Uñ „ðUð48<ð 8<ð 8<ð 8<ð 8< b¤iñ 8<ô 8<ð 8<ðv ðJ
ð J
ð J
ð J
ð J
Ð;ñ J
ô J
ñ „ðJ
ð^ #Ø
Ø*.ð	B&ð B&Ø”  e¤lÔ 3Ñ3°dÑ:ðB&à�t‘ðB&ð ”L 4Ñ'ð	B&ð
 „\�CÑðB&ð B&ð B&ð B&ðJ ðf
ð f
ð f
ð f
ð f
Ð"AÀ?ñ f
ô f
ñ „ðf
ðR fÐ
eÐ
e€€€r2   