§
    ‚ŠtjÑ  ã                   ó  — d dl mZ d dlmZmZ d dl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 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 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*m+Z+m,Z, ddl-m.Z.m/Z/ ddl0m1Z1 ddl2m3Z3 ddl4m5Z5  e,j6        e7¦  «        Z8 G d„ ded¬¦  «        Z9 G d„ dej:        ¦  «        Z;d„ Z<d ej=        d!e>d"ej=        fd#„Z?	 dId%ej:        d&ej=        d'ej=        d(ej=        d)ej=        dz  d*e@d+e@d,e&e(         fd-„ZAdJd.„ZB G d/„ d0ej:        ¦  «        ZC G d1„ d2ej        j:        ¦  «        ZDd3ej=        d4e>fd5„ZEd6„ ZFd7„ ZGd8„ ZH G d9„ d:ej:        ¦  «        ZI G d;„ d<ej:        ¦  «        ZJ ed=¦  «         G d>„ d?ej:        ¦  «        ¦   «         ZK G d@„ dAe¦  «        ZLe) G dB„ dCe$¦  «        ¦   «         ZMe) G dD„ dEeM¦  «        ¦   «         ZNe) G dF„ dGeMe¦  «        ¦   «         ZOg dH¢ZPdS )Ké    )ÚCallable)ÚOptionalÚ	TypedDictN)Únné   )Úinitialization)ÚACT2FN)ÚCacheÚDynamicCache)ÚGenerationMixin)Úuse_kernel_forward_from_hub)Úforce_accelerate_hooks)Úlazy_load_kernel)Úcreate_causal_maskÚcreate_recurrent_attention_mask)ÚGradientCheckpointingLayer)ÚBaseModelOutputWithPastÚCausalLMOutputWithPast)ÚROPE_INIT_FUNCTIONSÚdynamic_rope_update)ÚALL_ATTENTION_FUNCTIONSÚPreTrainedModel)ÚUnpack)ÚTransformersKwargsÚauto_docstringÚcan_return_tupleÚis_torchdynamo_compilingÚlogging)Úmaybe_autocastÚmerge_with_config_defaults)Úresolve_internal_import)Úcapture_outputsé   )ÚBambaConfigc                   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BambaFlashAttentionKwargsaU  
    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    úf/var/www/html/CA-Chatbot/venv/lib/python3.11/site-packages/transformers/models/bamba/modeling_bamba.pyr&   r&   7   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r6   r&   F)Útotalc                   óÔ   ‡ — 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BambaRotaryEmbeddingÚinv_freqNÚconfigc                 ó²  •— t          ¦   «                              ¦   «          |j        | _        |j        | _        || _        | j        j        d         | _        | j        }| j        dk    rt          | j                 } || j        |¦  «        \  }| _
        |                      d|d¬¦  «         |                      d|                     ¦   «         d¬¦  «         d S )NÚ	rope_typeÚdefaultr;   F)Ú
persistentÚoriginal_inv_freq)ÚsuperÚ__init__Úmax_position_embeddingsÚmax_seq_len_cachedÚoriginal_max_seq_lenr<   Úrope_parametersr>   Úcompute_default_rope_parametersr   Úattention_scalingÚregister_bufferÚclone)Úselfr<   ÚdeviceÚrope_init_fnr;   Ú	__class__s        €r7   rC   zBambaRotaryEmbedding.__init__R   sÊ   ø€ Ý‰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�$Ô(à×Ò˜Z¨¸eÐÑDÔDÐDØ×ÒÐ0°(·.².Ñ2BÔ2BÈuÐÑUÔUÐUÐUÐUr6   rM   ztorch.deviceÚseq_lenÚreturnztorch.Tensorc                 óü   — | j         d         }t          | dd¦  «        p| j        | j        z  }d}d|t	          j        d|dt          j        ¬¦  «                             |t          j        ¬¦  «        |z  z  z  }||fS )	a¨  
        Computes the inverse frequencies according to the original RoPE implementation
        Args:
            config ([`~transformers.PreTrainedConfig`]):
                The model configuration.
            device (`torch.device`):
                The device to use for initialization of the inverse frequencies.
            seq_len (`int`, *optional*):
                The current sequence length. Unused for this type of RoPE.
        Returns:
            Tuple of (`torch.Tensor`, `float`), containing the inverse frequencies for the RoPE embeddings and the
            post-processing scaling factor applied to the computed cos/sin (unused in this type of RoPE).
        Ú
rope_thetaÚhead_dimNg      ð?r   é   ©Údtype©rM   rW   )	rG   ÚgetattrÚhidden_sizeÚnum_attention_headsr0   ÚarangeÚint64ÚtoÚfloat)r<   rM   rP   ÚbaseÚdimÚattention_factorr;   s          r7   rH   z4BambaRotaryEmbedding.compute_default_rope_parametersb   sŒ   € ð& Ô% lÔ3ˆÝ�f˜j¨$Ñ/Ô/Ðc°6Ô3EÈÔIcÑ3cˆàÐð Ø•U”\ ! S¨!µ5´;Ð?Ñ?Ô?×BÒBÈ&ÕX]ÔXcÐBÑdÔdÐgjÑjÑkñ
ˆð Ð)Ð)Ð)r6   c                 óN  — | j         d d d …d f                              ¦   «                              |j        d         dd¦  «                             |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   éÿÿÿÿr#   ÚmpsÚcpuF)Údevice_typeÚenabledrU   ©ra   rV   )r;   r_   ÚexpandÚshaper^   rM   Ú
isinstanceÚtypeÚstrr   Ú	transposer0   ÚcatÚcosrI   ÚsinrW   )
rL   ÚxÚposition_idsÚinv_freq_expandedÚposition_ids_expandedrg   ÚfreqsÚembrq   rr   s
             r7   ÚforwardzBambaRotaryEmbedding.forward€   s·  € ð !œM¨$°°°°4¨-Ô8×>Ò>Ñ@Ô@×GÒGÈÔHZÐ[\ÔH]Ð_aÐcdÑeÔe×hÒhÐijÔiqÑrÔrÐØ ,¨Q¨Q¨Q°°a°a°a¨ZÔ 8× >Ò >Ñ @Ô @Ðå'1°!´(´-ÅÑ'EÔ'EÐkÈ!Ì(Ì-Ð[`ÒJ`ÐJ`�a”h”m�mÐfkˆÝ¨¸UÐCÑCÔCð 	5ð 	5Ø&×,Ò,Ñ.Ô.Ð1F×1LÒ1LÑ1NÔ1NÑN×YÒYÐZ[Ð]^Ñ_Ô_ˆEÝ”)˜U E˜N°Ð3Ñ3Ô3ˆCØ—'’'‘)”)˜dÔ4Ñ4ˆCØ—'’'‘)”)˜dÔ4Ñ4ˆCð		5ð 	5ð 	5ñ 	5ô 	5ð 	5ð 	5ð 	5ð 	5ð 	5ð 	5øøøð 	5ð 	5ð 	5ð 	5ð �vŠv˜AœGˆvÑ$Ô$ c§f¢f°1´7 fÑ&;Ô&;Ð;Ð;s   ÃBE&Å&E*Å-E*©N©NNN)r,   r-   r.   r0   ÚTensorr2   r$   rC   Ústaticmethodr   r3   Útupler_   rH   Úno_gradr   ry   Ú__classcell__©rO   s   @r7   r:   r:   O   sù   ø€ € € € € € ØŒlÐÐÑðVð V˜{ð Vð Vð Vð Vð Vð Vð  à%)Ø+/Ø"ð*ð *Ø˜dÑ"ð*à˜Ô(ð*ð �t‘ð*ð 
ˆ~˜uÐ$Ô	%ð	*ð *ð *ñ „\ð*ð: €U„]�_„_Øð<ð <ñ Ôñ „_ð<ð <ð <ð <ð <r6   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..Nrd   rU   ri   )rk   r0   rp   )rs   Úx1Úx2s      r7   Úrotate_halfr…   �   s]   € à	
ˆ3Ð"�!”'˜"”+ Ñ"Ð"Ð"Ô	#€BØ	
ˆ3�”˜”˜qÑ Ð"Ð"Ð"Ô	#€BÝŒ9�r�c˜2�Y BÐ'Ñ'Ô'Ð'r6   Úhidden_statesÚn_reprQ   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)rk   rj   Úreshape)r†   r‡   ÚbatchÚnum_key_value_headsÚslenrT   s         r7   Ú	repeat_kvr�   —   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Tr6   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 )NrU   r   rd   )ra   rW   )ÚpÚtrainingr#   )r�   Únum_key_value_groupsr0   Úmatmulro   r   Ú
functionalÚsoftmaxÚfloat32r^   rW   r•   r™   Ú
contiguous)r�   r�   r‘   r’   r“   r”   r•   r–   Ú
key_statesÚvalue_statesÚattn_weightsÚattn_outputs               r7   Úeager_attention_forwardr¤   £   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$Ð$r6   c                 ó˜  — |                      |¦  «        }|                      |¦  «        }|j        d         }| dd|…f         | d|d…f         }}|dd|…f         |d|d…f         }	}||z  t          |¦  «        |z  z   }
||z  t          |¦  «        |z  z   }t          j        |
|gd¬¦  «        }
t          j        ||	gd¬¦  «        }|
|fS )a»  Applies Rotary Position Embedding to the query and key tensors.

    Removes the interleaving of cos and sin from GLM

    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.
    rd   .Nri   )Ú	unsqueezerk   r…   r0   rp   )ÚqÚkrq   rr   Úunsqueeze_dimÚ
rotary_dimÚq_rotÚq_passÚk_rotÚk_passÚq_embedÚk_embeds               r7   Úapply_rotary_pos_embr±   ½   sô   € ð( �-Š-˜Ñ
&Ô
&€CØ
�-Š-˜Ñ
&Ô
&€Cð ”˜2”€JØ�c˜;˜J˜;Ð&Ô'¨¨3°
°°Ð+;Ô)<ˆ6€EØ�c˜;˜J˜;Ð&Ô'¨¨3°
°°Ð+;Ô)<ˆ6€Eð �s‰{�{¨5Ñ1Ô1°CÑ7Ñ8€GØ�s‰{�{¨5Ñ1Ô1°CÑ7Ñ8€Gõ Œi˜ &Ð)¨rÐ2Ñ2Ô2€GÝŒi˜ &Ð)¨rÐ2Ñ2Ô2€GØ�GÐÐr6   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BambaAttentionz=Multi-headed attention from 'Attention Is All You Need' paperr<   Ú	layer_idxc                 ó®  •— t          ¦   «                              ¦   «          || _        || _        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        ¬¦  «        | _        d S )NrT   g      à¿T©Úbias)rB   rC   r<   r´   rY   rZ   r[   rT   r‹   rš   r”   Úattention_dropoutÚ	is_causalr   ÚLinearÚattention_biasÚq_projÚk_projÚv_projÚo_proj)rL   r<   r´   rO   s      €r7   rC   zBambaAttention.__init__æ   sB  ø€ Ý‰Œ×ÒÑÔÐØˆŒØ"ˆŒÝ ¨
°FÔ4FÈ&ÔJdÑ4dÑeÔeˆŒØ$*Ô$>À&ÔB\Ñ$\ˆÔ!Ø”} dÑ*ˆŒØ!'Ô!9ˆÔØˆŒå”iØÔ Ô :¸T¼]Ñ JÐQWÔQfð
ñ 
ô 
ˆŒõ ”iØÔ Ô :¸T¼]Ñ JÐQWÔQfð
ñ 
ô 
ˆŒõ ”iØÔ Ô :¸T¼]Ñ JÐQWÔQfð
ñ 
ô 
ˆŒõ ”iØÔ&¨¬Ñ6¸Ô8JÐQWÔQfð
ñ 
ô 
ˆŒˆˆr6   Nr†   Úposition_embeddingsr“   Úpast_key_valuesr–   rQ   c                 ó"  — |j         d d…         }g |¢d‘| j        ‘R }|                      |¦  «                             |¦  «                             dd¦  «        }|                      |¦  «                             |¦  «                             dd¦  «        }	|                      |¦  «                             |¦  «                             dd¦  «        }
|\  }}t          ||	||¦  «        \  }}	|�|                     |	|
| j	        ¦  «        \  }	}
t          j        | j        j        t          ¦  «        } || ||	|
|f| j        sdn| j        | j        dœ|¤Ž\  }} |j        g |¢d‘R Ž                      ¦   «         }|                      |¦  «        }||fS )Nrd   r#   rU   rŽ   )r•   r”   )rk   rT   r¼   Úviewro   r½   r¾   r±   Úupdater´   r   Úget_interfacer<   Ú_attn_implementationr¤   r™   r¸   r”   r‰   rŸ   r¿   )rL   r†   rÀ   r“   rÁ   r–   Úinput_shapeÚhidden_shapeÚquery_statesr    r¡   rq   rr   Úattention_interfacer£   r¢   s                   r7   ry   zBambaAttention.forwardý   sÀ  € ð $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(r6   r{   )r,   r-   r.   r/   r$   r3   rC   r0   r|   r~   r
   r   r   ry   r€   r�   s   @r7   r³   r³   ã   så   ø€ € € € € ØGÐGð
˜{ð 
°sð 
ð 
ð 
ð 
ð 
ð 
ð4 IMØ.2Ø(,ð&)ð &)à”|ð&)ð # 5¤<°´Ð#=Ô>ÀÑEð&)ð œ tÑ+ð	&)ð
  ™ð&)ð Ð+Ô,ð&)ð 
ˆuŒ|˜Uœ\Ð)Ô	*ð&)ð &)ð &)ð &)ð &)ð &)ð &)ð &)r6   r³   c                   ó(   ‡ — e Zd Zdˆ fd„	Zdd„Zˆ xZS )ÚBambaRMSNormGatedç�íµ ÷Æ°>c                 ó¬   •— t          ¦   «                              ¦   «          t          j        t	          j        |¦  «        ¦  «        | _        || _        d S rz   ©rB   rC   r   Ú	Parameterr0   ÚonesÚweightÚvariance_epsilon©rL   rZ   ÚepsrO   s      €r7   rC   zBambaRMSNormGated.__init__'  sB   ø€ Ý‰Œ×ÒÑÔÐÝ”l¥5¤:¨kÑ#:Ô#:Ñ;Ô;ˆŒØ #ˆÔÐÐr6   Nc                 óœ  — |j         }|                     t          j        ¦  «        }|�?|t          j                             |                     t          j        ¦  «        ¦  «        z  }|                     d¦  «                             dd¬¦  «        }|t          j	        || j
        z   ¦  «        z  }| j        |                     |¦  «        z  S ©NrU   rd   T)Úkeepdim)rW   r^   r0   rž   r   rœ   ÚsiluÚpowÚmeanÚrsqrtrÓ   rÒ   )rL   r†   ÚgateÚinput_dtypeÚvariances        r7   ry   zBambaRMSNormGated.forward,  sª   € Ø#Ô)ˆØ%×(Ò(­¬Ñ7Ô7ˆàÐØ)­B¬M×,>Ò,>¸t¿wºwÅuÄ}Ñ?UÔ?UÑ,VÔ,VÑVˆMØ ×$Ò$ QÑ'Ô'×,Ò,¨R¸Ð,Ñ>Ô>ˆØ%­¬°H¸tÔ?TÑ4TÑ(UÔ(UÑUˆàŒ{˜]×-Ò-¨kÑ:Ô:Ñ:Ð:r6   ©rÍ   rz   ©r,   r-   r.   rC   ry   r€   r�   s   @r7   rÌ   rÌ   &  sQ   ø€ € € € € ð$ð $ð $ð $ð $ð $ð
	;ð 	;ð 	;ð 	;ð 	;ð 	;ð 	;ð 	;r6   rÌ   Úinput_tensorÚpad_sizec                 ó¦   — t          | j        ¦  «        dk    r
ddddd|ddfnddd|ddf}t          j        j                             | |dd¬¦  «        S )z‚
    Padding x tensor with `pad_size` on the seq_len dim (dim=1)

    Assumes that we only have tensors of either size 4 or 3
    é   r   Úconstant)Úmoder’   )Úlenrk   r0   r   rœ   Úpad)râ   rã   Ú	pad_shapes      r7   Úpad_tensor_by_sizerë   ;  sj   € õ 47°|Ô7IÑ3JÔ3JÈaÒ3OÐ3O��A�q˜!˜Q ¨!¨QÐ/Ð/ÐVWÐYZÐ\]Ð_gÐijÐlmÐUn€IåŒ8Ô×"Ò" <°ÀÐSTÐ"ÑUÔUÐUr6   c                 ó"  — t          | |¦  «        } t          | j        ¦  «        dk    r.|                      | j        d         d|| j        d         ¦  «        S |                      | j        d         d|| j        d         | j        d         ¦  «        S )zÀ
    Padding input_tensor with `pad_size` on the seq_len dim (dim=1) and
    simultaneously splitting it into chunk sequences.

    Assumes that we only have tensors of either size 4 or 3
    r   r   rd   rU   )rë   rè   rk   r‰   )râ   rã   Ú
chunk_sizes      r7   Úreshape_into_chunksrî   F  s’   € õ & l°HÑ=Ô=€Lå
ˆ<ÔÑÔ !Ò#Ð#à×#Ò# LÔ$6°qÔ$9¸2¸zÈ<ÔK]Ð^_ÔK`ÑaÔaÐað ×#Ò#ØÔ˜qÔ! 2 z°<Ô3EÀaÔ3HÈ,ÔJ\Ð]^ÔJ_ñ
ô 
ð 	
r6   c                 ó  — |                       d¦  «        } | d         j        g |                       ¦   «         ¢|‘R Ž } t          j        t          j        ||| j        t          j        ¬¦  «        d¬¦  «        }|                      | d¦  «        } t          j        | d¬¦  «        }t          j        t          j        ||| j        t          j        ¬¦  «        d¬¦  «        }|                     | t          j	         ¦  «        }|S )zo
    More stable segment sum calculation. Uses cumulative sums and masking instead of direct subtractions.
    rd   ©.NrX   )Údiagonalr   éþÿÿÿri   )
Úsizerj   r0   ÚtrilrÑ   rM   ÚboolÚmasked_fillÚcumsumÚinf)râ   rí   ÚmaskÚtensor_segsums       r7   Úsegment_sumrû   Z  só   € ð ×"Ò" 2Ñ&Ô&€Jð 2�< 	Ô*Ô1ÐS°<×3DÒ3DÑ3FÔ3FÐSÈ
ÐSÐSÐS€LåŒ:•e”j ¨ZÀÔ@SÕ[`Ô[eÐfÑfÔfÐqsÐtÑtÔt€DØ×+Ò+¨T¨E°1Ñ5Ô5€Lå”L °2Ð6Ñ6Ô6€Mõ Œ:•e”j ¨ZÀÔ@SÕ[`Ô[eÐfÑfÔfÐqrÐsÑsÔs€DØ!×-Ò-¨t¨eµe´i°ZÑ@Ô@€MØÐr6   c                 ó¦   — |�N|j         d         dk    r=|j         d         dk    r,| j        }| |dd…dd…df         z                       |¦  «        } | S )zm
    Tunes out the hidden states for padding tokens, see https://github.com/state-spaces/mamba/issues/66
    Nr#   r   )rk   rW   r^   )r†   r“   rW   s      r7   Úapply_mask_to_padding_statesrý   n  si   € ð
 Ð! nÔ&:¸1Ô&=ÀÒ&AÐ&AÀnÔFZÐ[\ÔF]Ð`aÒFaÐFaØÔ#ˆØ&¨¸¸¸¸1¸1¸1¸d¸
Ô)CÑC×GÒGÈÑNÔNˆàÐr6   c            
       ó  ‡ — e Zd ZdZdedefˆ fd„Z	 	 	 ddej        de	dz  dej        dz  d	ej
        dz  fd
„Z	 	 dde	dz  dej        dz  fd„Z ed¦  «        	 	 	 dde	dz  dej        dz  d	ej
        dz  fd„¦   «         Zˆ xZS )Ú
BambaMixeruP  
    Compute âˆ†, A, B, C, and D the state space parameters and compute the `contextualized_states`.
    A, D are input independent (see Mamba paper [1] Section 3.5.2 "Interpretation of A" for why A isn't selective)
    âˆ†, B, C are input-dependent (this is a key difference between Mamba and the linear time invariant S4,
    and is why Mamba is called **selective** state spaces)

    The are a few differences between this and Mamba2Mixer:
    - The variable use_precomputed_states is slightly different due to the hybrid cache structure
    - There's a few non-obvious bugs fixed with batching in the slow path that exist in main
    - Some extra variables that our layer doesn't need have been removed
    - We ported most of the refactors in https://github.com/huggingface/transformers/pull/35154, which is (as of Dec 18, 2024) unmerged
    r<   r´   c           	      ó  •— t          ¦   «                              ¦   «          |j        | _        |j        | _        |j        | _        |j        | _        t          |j
        | j        z  ¦  «        | _        || _        |j        | _        |j        | _        t"          |j                 | _        |j        | _        |j        | _        |j        | _        |j        | _        |j        | _        |j        | _        |j        | _        |j        | _        | j        d| j        z  | j        z  z   | _         tC          j"        | j         | j         |j        | j        | j         | j        dz
  ¬¦  «        | _#        | j        | j         z   | j        z   }tC          j$        | j        || j        ¬¦  «        | _%        tC          j&        tO          j(        | j        ¦  «        ¦  «        | _)        tO          j*        d| j        dz   ¦  «        }tC          j&        tO          j+        |¦  «        ¦  «        | _,        t[          | j        | j        ¬¦  «        | _.        tC          j&        tO          j(        | j        ¦  «        ¦  «        | _/        tC          j$        | j        | j        | j        ¬¦  «        | _0        tc          d¦  «        }te          |dd ¦  «        a3te          |dd ¦  «        a4tc          d	¦  «        }tk          |d
¬¦  «        a6tk          |d¬¦  «        a7tk          |d¬¦  «        a8ts          tl          tn          tp          th          tf          f¦  «        a:tt          stv           <                    d¦  «         ntv           <                    d¦  «         |j=        |         | _>        d S )NrU   r#   )Úin_channelsÚout_channelsr·   Úkernel_sizeÚgroupsÚpaddingr¶   ©rÕ   zcausal-conv1dÚcausal_conv1d_updateÚcausal_conv1d_fnz	mamba-ssmz8ops.triton.selective_state_update.selective_state_update)Úchained_pathz1ops.triton.ssd_combined.mamba_chunk_scan_combinedz8ops.triton.ssd_combined.mamba_split_conv1d_scan_combineda  The fast path is not available because one of `(selective_state_update, causal_conv1d_fn, causal_conv1d_update)` is None. Falling back to the naive implementation. To install follow https://github.com/state-spaces/mamba/#installation and https://github.com/Dao-AILab/causal-conv1dzDThe fast path for Bamba will be used when running the model on a GPU)?rB   rC   Úmamba_n_headsÚ	num_headsrZ   Úmamba_d_stateÚssm_state_sizeÚmamba_d_convÚconv_kernel_sizer3   Úmamba_expandÚintermediate_sizer´   Úmamba_conv_biasÚuse_conv_biasÚ
hidden_actÚ
activationr	   ÚactÚmamba_proj_biasÚuse_biasÚrms_norm_epsÚlayer_norm_epsilonÚmamba_n_groupsÚn_groupsÚmamba_d_headrT   Úmamba_chunk_sizerí   Útime_step_limitÚtime_step_minÚtime_step_maxÚconv_dimr   ÚConv1dÚconv1drº   Úin_projrÐ   r0   rÑ   Údt_biasr\   ÚlogÚA_logrÌ   ÚnormÚDÚout_projr   rY   r  r  r!   Úselective_state_updateÚmamba_chunk_scan_combinedÚ mamba_split_conv1d_scan_combinedÚallÚis_fast_path_availableÚloggerÚwarning_onceÚlayer_typesÚ
layer_type)rL   r<   r´   Úprojection_sizeÚAÚcausal_conv1dÚ	mamba_ssmrO   s          €r7   rC   zBambaMixer.__init__‰  s:  ø€ Ý‰Œ×ÒÑÔÐØÔ-ˆŒØ!Ô-ˆÔØ$Ô2ˆÔØ &Ô 3ˆÔÝ!$ VÔ%8¸4Ô;KÑ%KÑ!LÔ!LˆÔØ"ˆŒØ#Ô3ˆÔØ Ô+ˆŒÝ˜&Ô+Ô,ˆŒØÔ.ˆŒà"(Ô"5ˆÔàÔ-ˆŒØÔ+ˆŒØ Ô1ˆŒà%Ô5ˆÔØ#Ô1ˆÔØ#Ô1ˆÔàÔ.°°T´]Ñ1BÀTÔEXÑ1XÑXˆŒÝ”iØœØœØÔ'ØÔ-Ø”=ØÔ)¨AÑ-ð
ñ 
ô 
ˆŒð Ô0°4´=Ñ@À4Ä>ÑQˆÝ”yØÔØØ”ð
ñ 
ô 
ˆŒõ ”|¥E¤J¨t¬~Ñ$>Ô$>Ñ?Ô?ˆŒõ ŒL˜˜DœN¨QÑ.Ñ/Ô/ˆÝ”\¥%¤)¨A¡,¤,Ñ/Ô/ˆŒ
Ý% dÔ&<À$ÔBYÐZÑZÔZˆŒ	Ý”�eœj¨¬Ñ8Ô8Ñ9Ô9ˆŒåœ	 $Ô"8¸$Ô:JÐQUÔQ^Ð_Ñ_Ô_ˆŒõ )¨Ñ9Ô9ˆÝ& }Ð6LÈdÑSÔSÐÝ" =Ð2DÀdÑKÔKÐõ % [Ñ1Ô1ˆ	Ý!8ØÐ$^ð"
ñ "
ô "
Ðõ %<ØÐ$Wð%
ñ %
ô %
Ð!õ ,CØÐ$^ð,
ñ ,
ô ,
Ð(õ
 "%å&Ý)Ý0Ý Ý$ðñ"
ô "
Ðõ &ð 	hÝ×Òð>ñô ð ð õ ×ÒÐ fÑgÔgÐgà Ô,¨YÔ7ˆŒˆˆr6   Nr†   Úcache_paramsr“   r+   c                 ó2  — t          ||¦  «        }|                      |¦  «        }|j        \  }}}| j        | j        z  }	|d uo|                     | j        ¦  «        }
|
r:|j        | j                 j        d         }|j        | j                 j	        d         }|
�rž|dk    �r—| 
                    d¦  «                             | j        | j        | j        gd¬¦  «        \  }}}t          ||| j        j         
                    d¦  «        | j        j        | j        ¦  «        }t)          j        || j        |	|	gd¬¦  «        \  }}}t)          j        | j                             ¦   «         ¦  «         }|d d …d df         d d …d d …d f                              d| j        | j        ¦  «                             t(          j        ¬¦  «        }|d d …d d …d f                              dd| j        ¦  «        }| j        d d …d df                              d| j        ¦  «        }| j        d d …d df                              d| j        ¦  «        }|                     || j        |j        d         | j        z  ¦  «        }|                     || j        |j        d         | j        z  ¦  «        }|                     || j        | j        ¦  «        }t?          |||||||d |d¬¦
  «
        }|                     || j        | j        z  ¦  «        }|                       ||¦  «        }|  !                    |¦  «        d d …d df         }�n\t)          j        | j                             ¦   «         ¦  «         }| j"        d	t/          d
¦  «        fk    ri nd| j"        i}| j#        r�|€�tI          || j        j         
                    d¦  «        | j        j        | j        |f| j        | j%        || j        | j         j        | j         j&        | j!        j        | j!        j        | j        | j        dddœ|¤Ž}�nu|                     | j        | j        | j        gd¬¦  «        \  }}}| '                    dd¦  «        }|
rt)          j(        ||gd¬¦  «        }|�PtR          j*         +                    || j,        |j        d         z
  df¦  «        }| -                    || j        ¦  «         | j        dvr>|  .                    |                      |¦  «        dd |j        d         …f         ¦  «        }n@t_          || j        j         
                    d¦  «        | j        j        | j        |¬¦  «        }|
r|d d …d d …| d …f         }| '                    dd¦  «        }t          ||¦  «        }t)          j        || j        |	|	gd¬¦  «        \  }}}ta          |                     ||d| j        ¦  «        |||                     ||| j        d¦  «        |                     ||| j        d¦  «        f| j%        | j        d |d| j        d|
r|nd dœ|¤Ž\  }}|�|�| 1                    || j        ¦  «        }|                     ||d¦  «        }|                       ||¦  «        }|  !                    |¦  «        }|S )Nr   r#   rd   ri   .rV   T)Úzr&  Údt_softplusrŽ   rø   Údt_limitF)r*  rí   r+   r  Úrmsnorm_weightÚrmsnorm_epsÚoutproj_weightÚoutproj_biasÚheaddimÚngroupsÚnorm_before_gateÚreturn_final_statesrU   )rÙ   Úswish)rs   rÒ   r·   r  r+   )rí   r*  r;  r+   rE  r&  r<  Úinitial_states)2rý   r%  rk   r  r  Úhas_previous_stater´   ÚlayersÚconv_statesÚrecurrent_statesÚsqueezeÚsplitr  r"  r  r  r$  rÒ   r·   r  r0   Úexpr(  r_   rj   rT   r^   rž   r&  r*  rÃ   r,  r)  r+  r  r™   r.  rí   rÓ   ro   rp   r   rœ   ré   r  Úupdate_conv_stater  r  r-  Úupdate_recurrent_state)rL   r†   r9  r“   r+   Úprojected_statesÚ
batch_sizerP   Ú_Úgroups_time_state_sizeÚuse_precomputed_statesÚ
conv_stateÚrecurrent_staterÝ   Úhidden_states_B_CÚdtÚBÚCr6  r&  r*  Úhidden_states_reshapedÚoutÚdt_limit_kwargsrJ  Úscan_outputÚ	ssm_states                              r7   Úcuda_kernels_forwardzBambaMixer.cuda_kernels_forwardç  s  € õ 5°]ÀNÑSÔSˆØŸ<š<¨Ñ6Ô6Ðð "/Ô!4Ñˆ
�G˜QØ!%¤°Ô1DÑ!DÐà!-°TÐ!9Ð!m¸l×>]Ò>]Ð^bÔ^lÑ>mÔ>mÐØ!ð 	VØ%Ô,¨T¬^Ô<ÔHÈÔKˆJØ*Ô1°$´.ÔAÔRÐSTÔUˆOð "ñ L	1 g°¢l¡lØ*:×*BÒ*BÀ1Ñ*EÔ*E×*KÒ*KØÔ'¨¬¸¼ÐGÈRð +Lñ +ô +Ñ'ˆDÐ# Rõ
 !5Ø!ØØ”Ô"×*Ò*¨1Ñ-Ô-Ø”Ô Ø”ñ!ô !Ðõ #(¤+Ø!ØÔ'Ð)?ÐAWÐXØð#ñ #ô #ÑˆM˜1˜aõ ”˜4œ:×+Ò+Ñ-Ô-Ñ.Ô.Ð.ˆAØ�!�!�!�T˜3�,”    1 1 1 d 
Ô+×2Ò2°2°t´}ÀdÔFYÑZÔZ×]Ò]ÕdiÔdqÐ]ÑrÔrˆAØ�A�A�A�q�q�q˜$�J”×&Ò& r¨2¨t¬}Ñ=Ô=ˆBØ”l 1 1 1 d¨C <Ô0×7Ò7¸¸D¼MÑJÔJˆGØ”�q�q�q˜$ �|Ô$×+Ò+¨B°´Ñ>Ô>ˆAØ—’�z 4¤=°!´'¸!´*ÀÄÑ2MÑNÔNˆAØ—’�z 4¤=°!´'¸!´*ÀÄÑ2MÑNÔNˆAØ%2×%7Ò%7¸
ÀDÄNÐTXÔTaÑ%bÔ%bÐ"Ý2ØØ&ØØØØØØØØ ðñ ô ˆMð *×.Ò.¨z¸4¼>ÈDÌMÑ;YÑZÔZˆMØ ŸIšI m°TÑ:Ô:ˆMð —-’- Ñ.Ô.¨q¨q¨q°$¸¨|Ô<ˆC‰Cõ ”˜4œ:×+Ò+Ñ-Ô-Ñ.Ô.Ð.ˆAØ$(Ô$8¸SÅ%ÈÁ,Ä,Ð<OÒ$OÐ$O˜b˜bÐV`ÐbfÔbvÐUwˆOð Œ}ð X1 Ð!5Ý6Ø$Ø”KÔ&×.Ò.¨qÑ1Ô1Ø”KÔ$Ø”LØðð ”fØ#œØ#Ø#œØ#'¤9Ô#3Ø $¤	Ô :Ø#'¤=Ô#7Ø!%¤Ô!3Ø œMØ œMØ%*Ø(-ð#ð ð$ &ð%ð �‘ð, /?×.DÒ.DØÔ+¨T¬]¸D¼NÐKÐQSð /Eñ /ô /Ñ+�Ð'¨ð
 %6×$?Ò$?ÀÀ1Ñ$EÔ$EÐ!Ø)ð [õ ).¬	°:Ð?PÐ2QÐWYÐ(ZÑ(ZÔ(ZÐ%ØÐ+Ý"$¤-×"3Ò"3Ø)ØÔ.Ð1BÔ1HÈÔ1LÑLÈaÐPñ#ô #�Kð !×2Ò2°;ÀÄÑOÔOÐOà”?Ð*;Ð;Ð;Ø(,¯ª°·²Ð=NÑ1OÔ1OÐPSÐUrÐWhÔWnÐoqÔWrÐUrÐPrÔ1sÑ(tÔ(tÐ%Ð%å(8Ø+Ø#œ{Ô1×9Ò9¸!Ñ<Ô<Ø!œ[Ô-Ø#'¤?Ø 'ð)ñ )ô )Ð%ð *ð KØ(9¸!¸!¸!¸Q¸Q¸QÀÀÀ	À	¸/Ô(JÐ%Ø$5×$?Ò$?ÀÀ1Ñ$EÔ$EÐ!å$@ÐARÐTbÑ$cÔ$cÐ!Ý&+¤kØ%ØÔ+Ð-CÐE[Ð\Øð'ñ 'ô 'Ñ#�˜q !õ *CØ!×&Ò& z°7¸BÀÄÑNÔNØØØ—F’F˜: w°´¸rÑBÔBØ—F’F˜: w°´¸rÑBÔBð*ð  $œØ”fØØ#Ø(,Ø œLØ $Ø6LÐ#V ? ?ÐRVð*ð *ð &ð*ð *Ñ&�˜Yð$ Ð(¨\Ð-EØ ,× CÒ CÀIÈtÌ~Ñ ^Ô ^�Ià)×.Ò.¨z¸7ÀBÑGÔG�à"Ÿiši¨°TÑ:Ô:�ð —m’m KÑ0Ô0�Øˆ
r6   c                 óÞ  ‡ ‡2— |j         \  }}}|j        }t          ||¦  «        }‰                      |¦  «        }|                     ‰ j        ‰ j        ‰ j        gd¬¦  «        \  }	}
}|
                     dd¦  «        }
|d uo| 	                    ‰ j
        ¦  «        }|r|j        ‰ j
                 j        d         }|r“|dk    r�|                     |
‰ j
        ¦  «        d‰ j         d …f         }t          j        |‰ j        j                             d¦  «        z  d¬¦  «        }
‰ j        r|
‰ j        j        z   }
‰                      |
¦  «        }
nÎ|rt          j        ||
gd¬¦  «        }
|�Pt0          j                             |
‰ j        |
j         d         z
  df¦  «        }|                     |‰ j
        ¦  «         ‰                      ‰                      |
¦  «        dd |
j         d         …f         ¦  «        }
|r|
d| d …f         }
|
                     dd¦  «        }
t          |
|¦  «        }
t          j        |
‰ j        ‰ j        ‰ j        z  ‰ j        ‰ j        z  gd¬¦  «        \  }}}t          j        ‰ j                             ¦   «         ¦  «         }|�rf|dk    �r_|j        ‰ j
                 j         d         j!        }|d d …dd d …f         d d …d df         }|                     dd¦  «         "                    ||j         d         ‰ j#        ¦  «        }‰ j$        d          "                    ‰ j$        j         d         ‰ j#        ¦  «        }t          j        j         %                    || &                    |j        ¦  «        z   ¦  «        }t          j'        |‰ j(        d         ‰ j(        d         ¦  «        }|d          "                    ‰ j        ‰ j#        ‰ j        ¦  «         &                    t          j)        ¬	¦  «        }t          j        |d         |z  ¦  «         &                    |¬
¦  «        }| *                    |‰ j        d¦  «        dd d d …f         }| "                    |‰ j        ‰ j        ‰ j        z  |j         d         ¦  «         +                    ¦   «         }| *                    |d|j         d         ¦  «        }|d         |dd d d …f         z  }| *                    |d‰ j#        ¦  «        }||d         z   &                    |¬
¦  «        }|j        ‰ j
                 j         d         |z  |z   }| ,                    |‰ j
        ¦  «        }| *                    |‰ j        d¦  «        dd d d …f         }| "                    |‰ j        ‰ j        ‰ j        z  |j         d         ¦  «         +                    ¦   «         }| *                    |d|j         d         ¦  «        }| &                    |j!        |j        ¬¦  «        }| -                    |‰ j        z  ‰ j#        ‰ j        ¦  «        }| -                    |‰ j        z  ‰ j        d¦  «        }t          j.        ||¦  «        }| -                    |‰ j        ‰ j#        ¦  «        }‰ j/        d          "                    ‰ j/        j         d         ‰ j#        ¦  «        }|||z  z    &                    |j        ¦  «        }| *                    |d¦  «        d d …d df         }�n0t0          j         %                    |‰ j$        z   ¦  «        }t          j'        |‰ j(        d         ‰ j(        d         ¦  «        }| *                    ||d‰ j#        ¦  «                             ¦   «         }| *                    ||d‰ j        ¦  «                             ¦   «         }| *                    ||d‰ j        ¦  «                             ¦   «         }| 0                    ‰ j        ‰ j        z  d‰ j        ¬¦  «        }| 0                    ‰ j        ‰ j        z  d‰ j        ¬¦  «        }‰ j1        |‰ j1        z  z
  ‰ j1        z  Š2‰ j/        d         te          |‰2¦  «        z  }||d         z  }| &                    |j        ¦  «        |z  }ˆ2ˆ fd„||||fD ¦   «         \  }}}}| 3                    dddd¦  «        }t          j4        |d¬¦  «        }t          j        tk          |¦  «        ¦  «        }|d d …d d …d d …d d d …d d …f         |d d …d d …d d d …d d …d d …f         z  } |                      d¬¦  «        }!|!d         | 3                    ddddd¦  «        d         z  }"|"                     d¬¦  «        }#|#d         |d d …d d …d f         z                       d¬¦  «        }$t          j        |d d …d d …d d …dd …f         |z
  ¦  «        }%||% 3                    dddd¦  «        d         z  }&|&dd d d …f         |d         z                       d¬¦  «        }'|rF|j        ‰ j
                 j         d         d d …d f          &                    |'j        |'j!        ¬¦  «        nt          j6        |'d d …d d…f         ¦  «        }(t          j        |(|'gd¬¦  «        }'t          j        tk          t0          j                             |d d …d d …d d …df         d¦  «        ¦  «        ¦  «        })|)                     dd¦  «        })|)d         |'d d …d d …d df         z                       d¬¦  «        }*|*d d …d d…f         |*d d …df         }+}'t          j        |¦  «        },|dd d d …f         |'d d …d d …d df         z  }-|, 3                    dddd¦  «        }.|-                     d¦  «        |.d         z  }/|$|/z   }| *                    |d‰ j        ‰ j#        ¦  «        }||z   }‰2dk    r|d d …d |…d d …d d …f         }| *                    ||d¦  «        }|+�|�| ,                    |+‰ j
        ¦  «        }+‰  7                    ||	¦  «        }0‰  8                    |0 &                    |¦  «        ¦  «        }1|1S )Nrd   ri   r#   rU   r   .rð   ).NNrV   ©rM   rX   )ra   Úoutput_sizec                 ó<   •— g | ]}t          |‰‰j        ¦  «        ‘ŒS r5   )rî   rí   )Ú.0Útrã   rL   s     €€r7   ú
<listcomp>z,BambaMixer.torch_forward.<locals>.<listcomp>  s)   ø€ Ð%zÐ%zÐ%zÐ\]Õ&9¸!¸XÀtÄÑ&WÔ&WÐ%zÐ%zÐ%zr6   r   rå   rò   )rW   rM   )r#   r   )9rk   rW   rý   r%  rM  r  r"  r  ro   rH  r´   rI  rJ  rO  r  r0   Úsumr$  rÒ   rL  r  r·   r  rp   r   rœ   ré   r  r  rN  r(  r_   rK  rM   rj   rT   r&  Úsoftplusr^   Úclampr  rž   r‰   rŸ   rP  rÃ   Úbmmr*  Úrepeat_interleaverí   rë   Úpermuter÷   rû   Ú
zeros_liker)  r+  )3rL   Úinput_statesr9  r“   rR  rP   rS  rW   rQ  rÝ   rX  rY  rU  rV  rJ  r†   rZ  r[  r6  Úcache_devicer&  ÚdAÚdBÚdBxÚ
ssm_statesÚssm_states_reshapedÚ
C_reshapedÚyr*  Ú
D_residualÚA_cumsumÚLÚG_intermediateÚGÚM_intermediateÚMÚY_diagÚdecay_statesÚB_decayÚstatesÚprevious_statesÚdecay_chunkÚ
new_statesr`  Ústate_decay_outÚC_times_statesÚstate_decay_out_permutedÚY_offr_  Úcontextualized_statesrã   s3   `                                                 @r7   Útorch_forwardzBambaMixer.torch_forwardŒ  st  øø€ ð ".Ô!3Ñˆ
�G˜QØÔ"ˆõ 4°LÀ.ÑQÔQˆØŸ<š<¨Ñ5Ô5ÐØ&6×&<Ò&<ØÔ'¨¬¸¼ÐGÈRð '=ñ '
ô '
Ñ#ˆÐ ð .×7Ò7¸¸!Ñ<Ô<Ðà!-°TÐ!9Ð!m¸l×>]Ò>]Ð^bÔ^lÑ>mÔ>mÐØ!ð 	LØ%Ô,¨T¬^Ô<ÔHÈÔKˆJð "ð 	B g°¢l lØ&×8Ò8Ð9JÈDÌNÑ[Ô[Ð\_ÐbfÔbwÐawÐaxÐaxÐ\xÔyˆKå %¤	Ø˜dœkÔ0×8Ò8¸Ñ;Ô;Ñ;Àð!ñ !ô !Ðð Ô!ð IØ$5¸¼Ô8HÑ$HÐ!Ø $§¢Ð):Ñ ;Ô ;ÐÐà%ð WÝ$)¤I¨zÐ;LÐ.MÐSUÐ$VÑ$VÔ$VÐ!ØÐ'Ý œm×/Ò/Ø%¨Ô(=Ð@QÔ@WÐXZÔ@[Ñ([Ð]^Ð'_ñô �ð ×.Ò.¨{¸D¼NÑKÔKÐKà $§¢¨¯ªÐ5FÑ)GÔ)GÈÐMjÐO`ÔOfÐgiÔOjÐMjÐHjÔ)kÑ lÔ lÐØ%ð FØ$5°c¸G¸8¸9¸9°nÔ$EÐ!Ø 1× ;Ò ;¸A¸qÑ AÔ AÐå8Ð9JÈNÑ[Ô[ÐÝ#œkØØÔ# T¤]°TÔ5HÑ%HÈ$Ì-ÐZ^ÔZmÑJmÐnØð
ñ 
ô 
Ñˆ�q˜!õ ŒY�t”z×'Ò'Ñ)Ô)Ñ*Ô*Ð*ˆØ!ñ F	[ g°¢l¡là'Ô.¨t¬~Ô>ÔOÐPQÔRÔYˆLð �A�A�A�q˜!˜!˜!�G”˜Q˜Q˜Q  c˜\Ô*ˆBØ—’˜a Ñ#Ô#×*Ò*¨:°r´xÀ´|ÀTÄ]ÑSÔSˆBà”l 9Ô-×4Ò4°T´\Ô5GÈÔ5JÈDÌMÑZÔZˆGå”Ô$×-Ò-¨b°7·:²:¸b¼hÑ3GÔ3GÑ.GÑHÔHˆBÝ”˜R Ô!5°aÔ!8¸$Ô:NÈqÔ:QÑRÔRˆBØ�/Ô"×)Ò)¨$¬.¸$¼-ÈÔI\Ñ]Ô]×`Ò`ÕglÔgtÐ`ÑuÔuˆAå”)˜B˜yœM¨AÑ-Ñ.Ô.×2Ò2¸,Ð2ÑGÔGˆBð
 —	’	˜* d¤m°RÑ8Ô8¸¸dÀAÀAÀA¸ÔFˆAØ—’˜ T¤]°D´NÀdÄmÑ4SÐUVÔU\Ð]_ÔU`ÑaÔa×lÒlÑnÔnˆAØ—	’	˜* b¨!¬'°"¬+Ñ6Ô6ˆAà�I”  3¨¨a¨a¨a <¤Ñ0ˆBð *×1Ò1°*¸bÀ$Ä-ÑPÔPˆMØ˜ iÔ0Ñ0×4Ò4¸LÐ4ÑIÔIˆCð &Ô,¨T¬^Ô<ÔMÈaÔPÐSUÑUÐX[Ñ[ˆJØ%×<Ò<¸ZÈÌÑXÔXˆJð —	’	˜* d¤m°RÑ8Ô8¸¸dÀAÀAÀA¸ÔFˆAØ—’˜ T¤]°D´NÀdÄmÑ4SÐUVÔU\Ð]_ÔU`ÑaÔa×lÒlÑnÔnˆAØ—	’	˜* b¨!¬'°"¬+Ñ6Ô6ˆAð $Ÿš¨a¬h¸a¼g˜ÑFÔFˆJà",§/¢/°*¸t¼~Ñ2MÈtÌ}Ð^bÔ^qÑ"rÔ"rÐØŸš 
¨T¬^Ñ ;¸TÔ=PÐRSÑTÔTˆJÝ”	Ð-¨zÑ:Ô:ˆAØ—’�z 4¤>°4´=ÑAÔAˆAð ”�yÔ!×(Ò(¨¬¬°a¬¸$¼-ÑHÔHˆAØ�] QÑ&Ñ&×*Ò*¨1¬7Ñ3Ô3ˆAð —	’	˜* bÑ)Ô)¨!¨!¨!¨T°3¨,Ô7ˆA‰Aõ ”×'Ò'¨¨T¬\Ñ(9Ñ:Ô:ˆBÝ”˜R Ô!5°aÔ!8¸$Ô:NÈqÔ:QÑRÔRˆBØ)×1Ò1°*¸gÀrÈ4Ì=ÑYÔY×_Ò_ÑaÔaˆMØ—	’	˜* g¨r°4Ô3FÑGÔG×MÒMÑOÔOˆAØ—	’	˜* g¨r°4Ô3FÑGÔG×MÒMÑOÔOˆAØ×#Ò# D¤N°d´mÑ$CÈÐX\ÔXfÐ#ÑgÔgˆAØ×#Ò# D¤N°d´mÑ$CÈÐX\ÔXfÐ#ÑgÔgˆAØœ¨'°D´OÑ*CÑCÀtÄÑVˆHàœ 	Ô*Õ-?ÀÈxÑ-XÔ-XÑXˆJð *¨B¨y¬MÑ9ˆMØ—’�]Ô(Ñ)Ô)¨BÑ.ˆAð &{Ð%zÐ%zÐ%zÐ%zÐboÐqrÐtuÐwxÐayÐ%zÑ%zÔ%zÑ"ˆM˜1˜a ð —	’	˜!˜Q  1Ñ%Ô%ˆAÝ”| A¨2Ð.Ñ.Ô.ˆHõ ”	�+ a™.œ.Ñ)Ô)ˆAð ˜q˜q˜q ! ! ! Q Q Q¨¨a¨a¨a°°°Ð2Ô3°a¸¸¸¸1¸1¸1¸dÀAÀAÀAÀqÀqÀqÈ!È!È!Ð8KÔ6LÑLˆNØ×"Ò" rÐ"Ñ*Ô*ˆAð ˜yœ\¨A¯IªI°a¸¸A¸qÀ!Ñ,DÔ,DÀYÔ,OÑOˆNØ×"Ò" rÐ"Ñ*Ô*ˆAð ˜	”l ]°1°1°1°a°a°a¸°:Ô%>Ñ>×CÒCÈÐCÑJÔJˆFõ !œ9 X¨a¨a¨a°°°°A°A°A°r°s°s¨lÔ%;¸hÑ%FÑGÔGˆLØ˜,×.Ò.¨q°"°b¸!Ñ<Ô<¸YÔGÑGˆGØ˜c 4¨¨¨˜lÔ+¨m¸IÔ.FÑF×KÒKÐPQÐKÑRÔRˆFð *ð5�Ô# D¤NÔ3ÔDÀQÔGÈÈÈÈ4ÈÔP×SÒSÐZ`ÔZfÐouÔo|ÐSÑ}Ô}Ð}åÔ% f¨Q¨Q¨Q°°°¨U¤mÑ4Ô4ð õ
 ”Y °Ð8¸aÐ@Ñ@Ô@ˆFÝœ)¥Kµ´×0AÒ0AÀ(È1È1È1ÈaÈaÈaÐQRÐQRÐQRÐTVÈ;ÔBWÐY_Ñ0`Ô0`Ñ$aÔ$aÑbÔbˆKØ%×/Ò/°°1Ñ5Ô5ˆKØ% oÔ6¸ÀÀÀÀ1À1À1ÀdÈCÀÔ9PÑP×UÒUÐZ[ÐUÑ\Ô\ˆJØ *¨1¨1¨1¨c¨r¨c¨6Ô 2°J¸q¸q¸qÀ"¸uÔ4E�IˆFõ $œi¨Ñ1Ô1ˆOØ  T¨1¨1¨1 œo°°q°q°q¸!¸!¸!¸TÀ3°Ô0GÑGˆNØ'6×'>Ò'>¸qÀ!ÀQÈÑ'JÔ'JÐ$Ø#×'Ò'¨Ñ+Ô+Ð.FÀyÔ.QÑQˆEð ˜‘ˆAà—	’	˜* b¨$¬.¸$¼-ÑHÔHˆAà�J‘ˆAà˜!Š|ˆ|Ø�a�a�a˜˜'˜ 1 1 1 a a aÐ'Ô(�Ø—	’	˜* g¨rÑ2Ô2ˆAð Ð$¨Ð)AØ(×?Ò?À	È4Ì>ÑZÔZ�	à—i’i  4Ñ(Ô(ˆð
 !%§¢¨k¯nªn¸UÑ.CÔ.CÑ DÔ DÐØ$Ð$r6   r$  c                 ó|  — t           r>d| j        j        j        j        v r&t          ¦   «         s|                      ||||¦  «        S |�t          d¦  «        ‚|j        }|�G|j	        d         dk    r6|j	        d         dk    r%||d d …d d …d f         z   
                    |¦  «        }|                      |||¦  «        S )NÚcudaz\`seq_idx` support requires fast path support. Please install `mamba_ssm` and `causal_conv1d`r#   r   )r0  r%  rÒ   rM   rm   r   ra  ÚNotImplementedErrorrW   rk   r^   rŒ  )rL   r†   r9  r“   r+   r–   rW   s          r7   ry   zBambaMixer.forwardS  sÝ   € õ "ð 	c f°´Ô0CÔ0JÔ0OÐ&OÐ&OÕXpÑXrÔXrÐ&OØ×,Ò,¨]¸LÈ.ÐZaÑbÔbÐbØÐÝ%Ønñô ð ð Ô#ˆØÐ%¨.Ô*>¸qÔ*AÀAÒ*EÐ*EÈ.ÔJ^Ð_`ÔJaÐdeÒJeÐJeà*¨^¸A¸A¸A¸q¸q¸qÀ$¸JÔ-GÑG×KÒKÈEÑRÔRˆMà×!Ò! -°¸~ÑNÔNÐNr6   r{   )NN)r,   r-   r.   r/   r$   r3   rC   r0   r|   r
   r4   ra  rŒ  r   ry   r€   r�   s   @r7   rÿ   rÿ   {  s†  ø€ € € € € ðð ð\8˜{ð \8°sð \8ð \8ð \8ð \8ð \8ð \8ðB &*Ø.2Ø*.ðbð bà”|ðbð ˜d‘lðbð œ tÑ+ð	bð
 ” 4Ñ'ðbð bð bð bðP &*Ø.2ð	D%ð D%ð ˜d‘lðD%ð œ tÑ+ð	D%ð D%ð D%ð D%ðN Ð˜HÑ%Ô%ð &*Ø.2Ø*.ðOð Oð ˜d‘lðOð œ tÑ+ð	Oð
 ” 4Ñ'ðOð Oð Oñ &Ô%ðOð Oð Oð Oð Or6   rÿ   c                   ó$   ‡ — e Zd Zˆ fd„Zd„ Zˆ xZS )ÚBambaMLPc                 ó¶  •— t          ¦   «                              ¦   «          || _        |j        | _        |j        | _        t          j        | j        | j        |j        ¬¦  «        | _        t          j        | j        | j        |j        ¬¦  «        | _	        t          j        | j        | j        |j        ¬¦  «        | _
        t          |j                 | _        d S )Nr¶   )rB   rC   r<   rZ   r  r   rº   Úmlp_biasÚ	gate_projÚup_projÚ	down_projr	   r  Úact_fn©rL   r<   rO   s     €r7   rC   zBambaMLP.__init__k  s¯   ø€ Ý‰Œ×ÒÑÔÐØˆŒØ!Ô-ˆÔØ!'Ô!9ˆÔÝœ 4Ô#3°TÔ5KÐRXÔRaÐbÑbÔbˆŒÝ”y Ô!1°4Ô3IÐPVÔP_Ð`Ñ`Ô`ˆŒÝœ 4Ô#9¸4Ô;KÐRXÔRaÐbÑbÔbˆŒÝ˜VÔ.Ô/ˆŒˆˆr6   c                 ó¨   — |                       |                      |                      |¦  «        ¦  «        |                      |¦  «        z  ¦  «        }|S rz   )r–  r—  r”  r•  )rL   rs   r–  s      r7   ry   zBambaMLP.forwardu  sA   € Ø—N’N 4§;¢;¨t¯~ª~¸aÑ/@Ô/@Ñ#AÔ#AÀDÇLÂLÐQRÁOÄOÑ#SÑTÔTˆ	ØÐr6   rá   r�   s   @r7   r‘  r‘  j  sG   ø€ € € € € ð0ð 0ð 0ð 0ð 0ðð ð ð ð ð ð r6   r‘  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BambaRMSNormrÍ   rÕ   rQ   Nc                 ó¬   •— t          ¦   «                              ¦   «          t          j        t	          j        |¦  «        ¦  «        | _        || _        dS )z;
        BambaRMSNorm is equivalent to T5LayerNorm
        NrÏ   rÔ   s      €r7   rC   zBambaRMSNorm.__init__|  sD   ø€ õ 	‰Œ×ÒÑÔÐÝ”l¥5¤:¨kÑ#:Ô#:Ñ;Ô;ˆŒØ #ˆÔÐÐr6   r†   c                 ó  — |j         }|                     t          j        ¦  «        }|                     d¦  «                             dd¬¦  «        }|t          j        || j        z   ¦  «        z  }| j        |                     |¦  «        z  S r×   )	rW   r^   r0   rž   rÚ   rÛ   rÜ   rÓ   rÒ   )rL   r†   rÞ   rß   s       r7   ry   zBambaRMSNorm.forward„  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:r6   c                 óH   — t          | j        j        ¦  «        › d| j        › �S )Nz, eps=)r~   rÒ   rk   rÓ   )rL   s    r7   Ú
extra_reprzBambaRMSNorm.extra_repr‹  s&   € Ý˜œÔ)Ñ*Ô*ÐIÐI°$Ô2GÐIÐIÐIr6   rà   )
r,   r-   r.   r_   rC   r0   r|   ry   r   r€   r�   s   @r7   rœ  rœ  z  sŒ   ø€ € € € € ð$ð $¨ð $¸$ð $ð $ð $ð $ð $ð $ð; U¤\ð ;°e´lð ;ð ;ð ;ð ;ðJð Jð Jð Jð Jð Jð Jr6   rœ  c                   ó  ‡ — e Zd Zddededefˆ fd„Z	 	 	 	 	 ddej        d	ej        dz  d
ej	        dz  de
dz  dedz  deej        ej        f         dz  dee         deej        eej        ej        f         dz  f         fd„Zˆ xZS )ÚBambaDecoderLayerÚlinear_attentionr<   r´   r4  c                 ó®  •— t          ¦   «                              ¦   «          d}|dk    rt          nd } ||¦  «        | _        t	          |j        |j        ¬¦  «        | _        t	          |j        |j        ¬¦  «        | _        || _	        |dk    rt          ||¬¦  «        | _        d S |dk    rt          ||¦  «        | _        d S t          d|›�¦  «        ‚)Nr#   r  r£  )r<   r´   Úfull_attentionzInvalid layer_type: )rB   rC   r‘  Úfeed_forwardrœ  rZ   r  Úinput_layernormÚpre_ff_layernormÚ
block_typerÿ   Úmambar³   Ú	self_attnÚ
ValueError)rL   r<   r´   r4  Únum_expertsÚffn_layer_classrO   s         €r7   rC   zBambaDecoderLayer.__init__�  sÜ   ø€ Ý‰Œ×ÒÑÔÐàˆØ&1°QÒ&6Ð&6�(˜(¸DˆØ+˜O¨FÑ3Ô3ˆÔÝ+¨FÔ,>ÀFÔDWÐXÑXÔXˆÔÝ ,¨VÔ-?ÀVÔEXÐ YÑ YÔ YˆÔà$ˆŒØÐ+Ò+Ð+Ý#¨6¸YÐGÑGÔGˆDŒJˆJˆJØÐ+Ò+Ð+Ý+¨F°IÑ>Ô>ˆDŒNˆNˆNåÐB°JÐBÐBÑCÔCÐCr6   NFr†   r“   rt   rÁ   Ú	use_cacherÀ   r–   rQ   c           
      ó&  — |}|                       |¦  «        }| j        dk    r | j        d|||dœ|¤Ž}d }	n"| j        dk    r | j        d||||||dœ|¤Ž\  }}	||z   }|}|                      |¦  «        }|                      |¦  «        }||z   }||	fS )Nr£  )r†   r9  r“   r¥  )r†   r“   rt   rÁ   r¯  rÀ   r5   )r§  r©  rª  r«  r¨  r¦  )
rL   r†   r“   rt   rÁ   r¯  rÀ   r–   ÚresidualÚself_attn_weightss
             r7   ry   zBambaDecoderLayer.forward¡  sÿ   € ð !ˆà×,Ò,¨]Ñ;Ô;ˆàŒ?Ð0Ò0Ð0Ø&˜DœJð Ø+Ø,Ø-ðð ð ð	ð ˆMð !%ÐÐØŒ_Ð 0Ò0Ð0Ø/=¨t¬~ð 0Ø+Ø-Ø)Ø /Ø#Ø$7ð0ð 0ð ð0ð 0Ñ,ˆMÐ,ð ! =Ñ0ˆà ˆØ×-Ò-¨mÑ<Ô<ˆØ×)Ò)¨-Ñ8Ô8ˆØ  =Ñ0ˆàÐ/Ð/Ð/r6   )r£  )NNNFN)r,   r-   r.   r$   r3   rn   rC   r0   r|   r1   r
   rõ   r~   r   r&   ÚFloatTensorry   r€   r�   s   @r7   r¢  r¢  �  s.  ø€ € € € € ðDð D˜{ð D°sð DÈð Dð Dð Dð Dð Dð Dð( /3Ø04Ø(,Ø!&ØHLð(0ð (0à”|ð(0ð œ tÑ+ð(0ð Ô&¨Ñ-ð	(0ð
  ™ð(0ð ˜$‘;ð(0ð # 5¤<°´Ð#=Ô>ÀÑEð(0ð Ð2Ô3ð(0ð 
ˆuÔ  %¨Ô(9¸5Ô;LÐ(LÔ"MÐPTÑ"TÐTÔ	Uð(0ð (0ð (0ð (0ð (0ð (0ð (0ð (0r6   r¢  c                   ó‚   ‡ — e Zd ZU eed<   dZdZdgZdgZdZ	dZ
dZdZeedœZ ej        ¦   «         ˆ fd„¦   «         Zˆ xZS )ÚBambaPreTrainedModelr<   ÚmodelTr¢  rÁ   )r†   Ú
attentionsc           
      ój  •— t          ¦   «                              |¦  «         t          |t          ¦  «        r{t	          j        |j        ¦  «         t	          j        |j        t          j
        t          j        d|j        dz   ¦  «        ¦  «        ¦  «         t	          j        |j        ¦  «         d S d S )Nr#   )rB   Ú_init_weightsrl   rÿ   ÚinitÚones_r&  Úcopy_r(  r0   r'  r\   r  r*  )rL   r�   rO   s     €r7   r¹  z"BambaPreTrainedModel._init_weightsÜ  s”   ø€ å‰Œ×Ò˜fÑ%Ô%Ð%Ý�f�jÑ)Ô)ð 	!ÝŒJ�v”~Ñ&Ô&Ð&ÝŒJ�v”|¥U¤Y­u¬|¸A¸vÔ?OÐRSÑ?SÑ/TÔ/TÑ%UÔ%UÑVÔVÐVÝŒJ�v”xÑ Ô Ð Ð Ð ð	!ð 	!r6   )r,   r-   r.   r$   r2   Úbase_model_prefixÚsupports_gradient_checkpointingÚ_no_split_modulesÚ_skip_keys_device_placementÚ_supports_flash_attnÚ_supports_sdpaÚ_is_statefulÚ_can_compile_fullgraphr¢  r³   Ú_can_record_outputsr0   r   r¹  r€   r�   s   @r7   rµ  rµ  Ì  s¡   ø€ € € € € € àÐÐÑØÐØ&*Ð#Ø,Ð-ÐØ#4Ð"5ÐØÐØ€NØ€LØ!Ðà*Ø$ðð Ðð
 €U„]�_„_ð!ð !ð !ð !ñ „_ð!ð !ð !ð !ð !r6   rµ  c                   óâ   ‡ — e Zd Zdefˆ fd„Zeee	 	 	 	 	 	 ddej	        dz  dej
        dz  dej	        dz  dedz  dej        dz  d	edz  d
ee         defd„¦   «         ¦   «         ¦   «         Zˆ xZS )Ú
BambaModelr<   c           	      óJ  •— t          ¦   «                              |¦  «         |j        | _        |j        | _        t          j        |j        |j        | j        ¦  «        | _        g }t          |j
        ¦  «        D ]2}|                     t          |||j        |         ¬¦  «        ¦  «         Œ3t          j        |¦  «        | _        |j        | _        t#          |j        |j        ¬¦  «        | _        t)          |¬¦  «        | _        d| _        |                      ¦   «          d S )N)r´   r4  r  ©r<   F)rB   rC   Úpad_token_idÚpadding_idxÚ
vocab_sizer   Ú	EmbeddingrZ   Úembed_tokensÚrangeÚnum_hidden_layersÚappendr¢  Úlayers_block_typeÚ
ModuleListrI  rÆ   rœ  r  Úfinal_layernormr:   Ú
rotary_embÚgradient_checkpointingÚ	post_init)rL   r<   Údecoder_layersÚirO   s       €r7   rC   zBambaModel.__init__ç  s	  ø€ Ý‰Œ×Ò˜Ñ Ô Ð Ø!Ô.ˆÔØ Ô+ˆŒåœL¨Ô):¸FÔ<NÐPTÔP`ÑaÔaˆÔØˆÝ�vÔ/Ñ0Ô0ð 	rð 	rˆAØ×!Ò!Õ"3°FÀaÐTZÔTlÐmnÔToÐ"pÑ"pÔ"pÑqÔqÐqÐqÝ”m NÑ3Ô3ˆŒà$*Ô$?ˆÔ!Ý+¨FÔ,>ÀFÔDWÐXÑXÔXˆÔÝ.°fÐ=Ñ=Ô=ˆŒà&+ˆÔ#à�ŠÑÔÐÐÐr6   NÚ	input_idsr“   rt   rÁ   Úinputs_embedsr¯  r–   rQ   c           
      ó„  — |d u |d uz  rt          d¦  «        ‚|€|                      |¦  «        }|}|r|€t          | j        ¬¦  «        }|€9t	          j        |j        d         |j        ¬¦  «                             d¦  «        }t          |x}	t          ¦  «        s&| j        ||||dœ}
t          di |
¤Žt          di |
¤Ždœ}	|                      ||¬¦  «        }t          | j        ¦  «        D ]-\  }} ||f|	| j        j        |                  ||||d	œ|¤Ž\  }}Œ.|                      |¦  «        }t%          ||¬
¦  «        S )Nz:You must specify exactly one of input_ids or inputs_embedsrÉ  r#   rc  r   )r<   rÛ  r“   rÁ   rt   )r¥  r£  )rt   )r“   rt   rÁ   r¯  rÀ   )Úlast_hidden_staterÁ   r5   )r¬  rÎ  r   r<   r0   r\   rk   rM   r¦   rl   Údictr   r   rÕ  Ú	enumeraterI  rÒ  rÔ  r   )rL   rÚ  r“   rt   rÁ   rÛ  r¯  r–   r†   Úcausal_mask_mappingÚmask_kwargsrÀ   rÙ  Údecoder_layerr¢   s                  r7   ry   zBambaModel.forwardú  sº  € ð ˜Ð -°tÐ";Ñ<ð 	[ÝÐYÑZÔZÐZàÐ Ø ×-Ò-¨iÑ8Ô8ˆMØ%ˆàð 	?˜Ð0Ý*°$´+Ð>Ñ>Ô>ˆOàÐÝ œ<¨Ô(;¸AÔ(>À}ÔG[Ð\Ñ\Ô\×fÒfÐghÑiÔiˆLå°Ð?Ð-ÅÑFÔFð 	ð œ+Ø!.Ø"0Ø#2Ø ,ðð ˆKõ #5Ð"CÐ"C°{Ð"CÐ"CÝ$CÐ$RÐ$RÀkÐ$RÐ$Rð#ð #Ðð #Ÿošo¨mÈ,˜oÑWÔWÐå )¨$¬+Ñ 6Ô 6ð 		ð 		ÑˆAˆ}Ø*7¨-Øð+à2°4´;Ô3PÐQRÔ3SÔTØ)Ø /Ø#Ø$7ð+ð +ð ð+ð +Ñ'ˆM˜<˜<ð ×,Ò,¨]Ñ;Ô;ˆå&Ø+Ø+ð
ñ 
ô 
ð 	
r6   )NNNNNN)r,   r-   r.   r$   rC   r    r"   r   r0   r1   r|   r
   r³  rõ   r   r&   r   ry   r€   r�   s   @r7   rÇ  rÇ  å  s  ø€ € € € € ð˜{ð ð ð ð ð ð ð&  ØØð .2Ø.2Ø04Ø(,Ø26Ø!%ð7
ð 7
àÔ# dÑ*ð7
ð œ tÑ+ð7
ð Ô&¨Ñ-ð	7
ð
  ™ð7
ð Ô(¨4Ñ/ð7
ð ˜$‘;ð7
ð Ð2Ô3ð7
ð 
!ð7
ð 7
ð 7
ñ „^ñ „_ñ  Ôð7
ð 7
ð 7
ð 7
ð 7
r6   rÇ  c                   ó   ‡ — 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fd„¦   «         ¦   «         Z	 	 	 	 	 	 dˆ fd„	Zˆ xZS )ÚBambaForCausalLMzlm_head.weightzmodel.embed_tokens.weightÚlm_headÚcolwise_gather_outputr†   Úlogitsc                 ó  •— t          ¦   «                              |¦  «         t          |¦  «        | _        |j        | _        t          j        |j        |j        d¬¦  «        | _        |j	        | _	        |  
                    ¦   «          d S )NFr¶   )rB   rC   rÇ  r¶  rÌ  r   rº   rZ   rå  Úz_loss_coefficientr×  r˜  s     €r7   rC   zBambaForCausalLM.__init__=  su   ø€ Ý‰Œ×Ò˜Ñ Ô Ð Ý Ñ'Ô'ˆŒ
Ø Ô+ˆŒÝ”y Ô!3°VÔ5FÈUÐSÑSÔSˆŒØ"(Ô";ˆÔð 	�ŠÑÔÐÐÐr6   Nr   rÚ  r“   rt   rÁ   rÛ  Úlabelsr¯  Úlogits_to_keeprQ   c	           
      ó(  —  | j         d
||||||dœ|	¤Ž}
|
j        }t          |t          ¦  «        rt	          | d¦  «        n|}|                      |dd…|dd…f         ¦  «        }d}|�‡ | j        d
||| j        j        dœ|	¤Ž}| j	        dk    ra| 
                    d¬¦  «                             |j        ¬¦  «                             d¦  «                             ¦   «         }|| j	        |z  z   }t          |||
j        |
j        |
j        ¬	¦  «        S )aJ  
        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, BambaForCausalLM

        >>> model = BambaForCausalLM.from_pretrained("...")
        >>> tokenizer = AutoTokenizer.from_pretrained("...")

        >>> 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."
        ```)rÚ  r“   rt   rÁ   rÛ  r¯  N)rç  rê  rÌ  r   rd   ri   rV   rU   )Úlossrç  rÁ   r†   r·  r5   )r¶  rÝ  rl   r3   Úslicerå  Úloss_functionr<   rÌ  ré  Ú	logsumexpr^   rW   rÚ   rÛ   r   rÁ   r†   r·  )rL   rÚ  r“   rt   rÁ   rÛ  rê  r¯  rë  r–   Úoutputsr†   Úslice_indicesrç  rí  Úz_losss                   r7   ry   zBambaForCausalLM.forwardG  sX  € ðH ,6¨4¬:ð ,
Ø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àˆØÐØ%�4Ô%Ðp¨V¸FÈtÌ{ÔOeÐpÐpÐioÐpÐpˆDØÔ&¨Ò*Ð*Ø×)Ò)¨bÐ)Ñ1Ô1×4Ò4¸4¼:Ð4ÑFÔF×JÒJÈ1ÑMÔM×RÒRÑTÔT�Ø˜dÔ5¸Ñ>Ñ>�å%ØØØ#Ô3Ø!Ô/ØÔ)ð
ñ 
ô 
ð 	
r6   TFc           
      óh   •— | j         j        |d<    t          ¦   «         j        |f||||||dœ|¤Ž}	|	S )Nrë  )rÁ   r“   rÛ  rt   r¯  Úis_first_iteration)r<   Únum_logits_to_keeprB   Úprepare_inputs_for_generation)rL   rÚ  rÁ   r“   rÛ  rt   r¯  rõ  r–   Úmodel_inputsrO   s             €r7   r÷  z.BambaForCausalLM.prepare_inputs_for_generationˆ  s^   ø€ ð $(¤;Ô#AˆÐÑ Ø<•u‘w”wÔ<Øð	
à+Ø)Ø'Ø%ØØ1ð	
ð 	
ð ð	
ð 	
ˆð Ðr6   )NNNNNNNr   )NNNNTF)r,   r-   r.   Ú_tied_weights_keysÚ_tp_planÚ_pp_planrC   r   r   r0   r1   r|   r
   r³  rõ   r3   r   ry   r÷  r€   r�   s   @r7   rä  rä  7  sn  ø€ € € € € à*Ð,GÐHÐØÐ2Ð3€HØ˜_Ð-°¨zÐ:Ð;€Hðð ð ð ð ð Øð .2Ø.2Ø04Ø(,Ø26Ø*.Ø!%Ø-.ð=
ð =
àÔ# dÑ*ð=
ð œ tÑ+ð=
ð Ô&¨Ñ-ð	=
ð
  ™ð=
ð Ô(¨4Ñ/ð=
ð Ô  4Ñ'ð=
ð ˜$‘;ð=
ð ˜eœlÑ*ð=
ð 
 ð=
ð =
ð =
ñ „^ñ Ôð=
ðD ØØØØØ ðð ð ð ð ð ð ð ð ð r6   rä  )rÇ  rä  rµ  )rŽ   )r#   )QÚcollections.abcr   Útypingr   r   r0   r   Ú r   rº  Úactivationsr	   Úcache_utilsr
   r   Ú
generationr   Úintegrationsr   Úintegrations.accelerater   Úintegrations.hub_kernelsr   Úmasking_utilsr   r   Úmodeling_layersr   Úmodeling_outputsr   r   Úmodeling_rope_utilsr   r   Úmodeling_utilsr   r   Úprocessing_utilsr   Úutilsr   r   r   r   r   Úutils.genericr   r    Úutils.import_utilsr!   Úutils.output_capturingr"   Úconfiguration_bambar$   Ú
get_loggerr,   r1  r&   ÚModuler:   r…   r|   r3   r�   r_   r¤   r±   r³   rÌ   rë   rî   rû   rý   rÿ   r‘  rœ  r¢  rµ  rÇ  rä  Ú__all__r5   r6   r7   ú<module>r     sz  ðð4 %Ð $Ð $Ð $Ð $Ð $Ø &Ð &Ð &Ð &Ð &Ð &Ð &Ð &à €€€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 7Ð 7Ð 7Ð 7Ð 7Ð 7Ø =Ð =Ð =Ð =Ð =Ð =Ø 8Ð 8Ð 8Ð 8Ð 8Ð 8Ø PÐ PÐ PÐ PÐ PÐ PÐ PÐ PØ 9Ð 9Ð 9Ð 9Ð 9Ð 9Ø OÐ OÐ OÐ OÐ OÐ OÐ OÐ OØ KÐ KÐ KÐ KÐ KÐ KÐ KÐ KØ FÐ FÐ FÐ FÐ FÐ FÐ FÐ FØ &Ð &Ð &Ð &Ð &Ð &Ø lÐ lÐ lÐ lÐ lÐ lÐ lÐ lÐ lÐ lÐ lÐ lÐ lÐ lØ GÐ GÐ GÐ GÐ GÐ GÐ GÐ GØ 9Ð 9Ð 9Ð 9Ð 9Ð 9Ø 5Ð 5Ð 5Ð 5Ð 5Ð 5Ø ,Ð ,Ð ,Ð ,Ð ,Ð ,ð 
ˆÔ	˜HÑ	%Ô	%€ðð ð ð ð  	°ð ñ ô ð ð0><ð ><ð ><ð ><ð ><˜2œ9ñ ><ô ><ð ><ðB(ð (ð (ð	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 %ð4#ð #ð #ð #ðL@)ð @)ð @)ð @)ð @)�R”Yñ @)ô @)ð @)ðF;ð ;ð ;ð ;ð ;˜œœñ ;ô ;ð ;ð*V U¤\ð V¸Sð Vð Vð Vð Vð
ð 
ð 
ð(ð ð ð(	ð 	ð 	ðlOð lOð lOð lOð lO�”ñ lOô lOð lOð^ð ð ð ð ˆrŒyñ ô ð ð  Ð˜YÑ'Ô'ðJð Jð Jð Jð J�2”9ñ Jô Jñ (Ô'ðJð(:0ð :0ð :0ð :0ð :0Ð2ñ :0ô :0ð :0ðz ð!ð !ð !ð !ð !˜?ñ !ô !ñ „ð!ð0 ðN
ð N
ð N
ð N
ð N
Ð%ñ N
ô N
ñ „ðN
ðb ðgð gð gð gð gÐ+¨_ñ gô gñ „ðgðT EÐ
DÐ
D€€€r6   