§
    ‚ŠtjÆ›  ã                   óØ  — d Z ddlmZ ddlZddlmZ ddlmZ ddlm	Z	 ddl
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 ddlmZ ddlmZmZmZmZ ddl m!Z!m"Z" ddl#m$Z$ ddl%m&Z& ddl'm(Z( ddl)m*Z*m+Z+m,Z,m-Z-m.Z.m/Z/ ddl0m1Z1m2Z2m3Z3m4Z4m5Z5 ddl6m7Z7  ej8        e9¦  «        Z: G d„ ded¬¦  «        Z; G d„ de.¦  «        Z<d3d„Z=e" G d „ d!e*¦  «        ¦   «         Z> G d"„ d#e1¦  «        Z? G d$„ d%ej@        ¦  «        ZA G d&„ d'e,¦  «        ZB G d(„ d)e-¦  «        ZC G d*„ d+e(¦  «        ZDe G d,„ d-e¦  «        ¦   «         ZEe G d.„ d/eE¦  «        ¦   «         ZF G d0„ d1e+¦  «        ZGg d2¢ZHdS )4zPyTorch Bamba model.é    )Ú	TypedDictN)Únné   )Úinitialization)ÚACT2FN)ÚCacheÚDynamicCache)Úforce_accelerate_hooks)Úlazy_load_kernel)Úcreate_causal_maskÚcreate_recurrent_attention_mask)ÚBaseModelOutputWithPastÚCausalLMOutputWithPast)ÚPreTrainedModel)ÚUnpack)Úauto_docstringÚcan_return_tupleÚis_torchdynamo_compilingÚlogging)Úmerge_with_config_defaultsÚno_inherit_decorator)Úresolve_internal_import)Úcapture_outputsé   )ÚJambaAttentionDecoderLayer)ÚLlamaAttentionÚLlamaForCausalLMÚLlamaMLPÚLlamaRMSNormÚLlamaRotaryEmbeddingÚrotate_half)ÚMambaRMSNormGatedÚapply_mask_to_padding_statesÚpad_tensor_by_sizeÚreshape_into_chunksÚsegment_sumé   )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    úe/var/www/html/CA-Chatbot/venv/lib/python3.11/site-packages/transformers/models/bamba/modular_bamba.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r:   r*   F)Útotalc                   ó   — e Zd ZdS )ÚBambaRotaryEmbeddingN©r0   r1   r2   r9   r:   r;   r>   r>   U   ó   € € € € € Ø€Dr:   r>   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.
    éÿÿÿÿ.N©Údim)Ú	unsqueezeÚshaper!   r4   Úcat)ÚqÚkÚcosÚsinÚunsqueeze_dimÚ
rotary_dimÚq_rotÚq_passÚk_rotÚk_passÚq_embedÚk_embeds               r;   Úapply_rotary_pos_embrT   Z   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r:   c                   ó   — e Zd ZdS )ÚBambaAttentionNr?   r9   r:   r;   rV   rV   €   s   € € € € € à€Dr:   rV   c                   ó   — e Zd ZdS )ÚBambaRMSNormGatedNr?   r9   r:   r;   rX   rX   …   r@   r:   rX   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
    ÚconfigÚ	layer_idxc           	      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 )Nr   r'   )Úin_channelsÚout_channelsÚbiasÚkernel_sizeÚgroupsÚpadding)r`   ©Úepsz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)?ÚsuperÚ__init__Úmamba_n_headsÚ	num_headsÚhidden_sizeÚmamba_d_stateÚssm_state_sizeÚmamba_d_convÚconv_kernel_sizer7   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_headÚhead_dimÚmamba_chunk_sizeÚ
chunk_sizeÚtime_step_limitÚtime_step_minÚtime_step_maxÚconv_dimr   ÚConv1dÚconv1dÚLinearÚin_projÚ	Parameterr4   ÚonesÚdt_biasÚarangeÚlogÚA_logrX   ÚnormÚDÚout_projr   Úgetattrrf   rg   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)Úselfr[   r\   Úprojection_sizeÚAÚcausal_conv1dÚ	mamba_ssmÚ	__class__s          €r;   rj   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r:   NÚhidden_statesÚcache_paramsÚattention_maskr/   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'   rB   rC   .©ÚdtypeT)Úzr�   Údt_softplusg        ÚinfÚdt_limitF)r’   r‚   r/   rw   Úrmsnorm_weightÚrmsnorm_epsÚoutproj_weightÚoutproj_biasÚheaddimÚngroupsÚnorm_before_gateÚreturn_final_statesr   )ÚsiluÚswish)ÚxÚweightr`   rw   r/   )r‚   r’   rª   r/   rµ   r�   r«   Úinitial_states)2r#   rŠ   rF   r~   ro   Úhas_previous_stater\   ÚlayersÚconv_statesÚrecurrent_statesÚsqueezeÚsplitrs   r†   rl   rf   rˆ   r¹   r`   rw   r4   Úexpr�   ÚfloatÚexpandr€   ÚtoÚfloat32r�   r’   Úviewr•   r‘   r“   rƒ   Útrainingr—   r‚   Úvariance_epsilonÚ	transposerG   r   Ú
functionalÚpadrq   Úupdate_conv_staterx   rg   r–   Úupdate_recurrent_state)rž   r¤   r¥   r¦   r/   Úprojected_statesÚ
batch_sizeÚseq_lenÚ_Úgroups_time_state_sizeÚuse_precomputed_statesÚ
conv_stateÚrecurrent_stateÚgateÚhidden_states_B_CÚdtÚBÚCr    r�   r’   Úhidden_states_reshapedÚoutÚdt_limit_kwargsr½   Úscan_outputÚ	ssm_states                              r;   Úcuda_kernels_forwardzBambaMixer.cuda_kernels_forwardö   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
r:   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 )NrB   rC   r'   r   r   .).N).NNr¨   ©Údevice)rã   r©   )rD   Úoutput_sizec                 ó<   •— g | ]}t          |‰‰j        ¦  «        ‘ŒS r9   )r%   r‚   )Ú.0ÚtÚpad_sizerž   s     €€r;   ú
<listcomp>z,BambaMixer.torch_forward.<locals>.<listcomp>  s)   ø€ Ð%zÐ%zÐ%zÐ\]Õ&9¸!¸XÀtÄÑ&WÔ&WÐ%zÐ%zÐ%zr:   r   é   éþÿÿÿ)r©   rã   )r'   r   )9rF   r©   r#   rŠ   rÀ   rs   r†   rl   rÉ   r»   r\   r¼   r½   rÌ   rq   r4   Úsumrˆ   r¹   r¿   ru   r`   rx   rG   r   rÊ   rË   r~   ro   rÁ   r�   rÂ   r¾   rã   rÃ   r€   r�   ÚsoftplusrÄ   Úclamprƒ   rÅ   ÚreshapeÚ
contiguousrÍ   rÆ   Úbmmr’   Úrepeat_interleaver‚   r$   ÚpermuteÚcumsumr&   Ú
zeros_liker‘   r“   )3rž   Úinput_statesr¥   r¦   rÏ   rÐ   rÑ   r©   rÎ   rÖ   r×   rØ   rÓ   rÔ   r½   r¤   rÙ   rÚ   r    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_statesrß   Ústate_decay_outÚC_times_statesÚstate_decay_out_permutedÚY_offrÞ   Úcontextualized_statesrè   s3   `                                                 @r;   Útorch_forwardzBambaMixer.torch_forward›  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$Ð$r:   rˆ   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   )r™   rŠ   r¹   rã   Útyper   rà   ÚNotImplementedErrorr©   rF   rÄ   r  )rž   r¤   r¥   r¦   r/   Úkwargsr©   s          r;   ÚforwardzBambaMixer.forwardb  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$¸JÔ-GÑG×KÒKÈEÑRÔRˆMà×!Ò! -°¸~ÑNÔNÐNr:   )NNN)NN)r0   r1   r2   r3   r(   r7   rj   r4   ÚTensorr   r8   rà   r  r
   r  Ú__classcell__©r£   s   @r;   rZ   rZ   Š   s†  ø€ € € € € ðð ð\8˜{ð \8°sð \8ð \8ð \8ð \8ð \8ð \8ðB &*Ø.2Ø*.ðbð bà”|ðbð ˜d‘lðbð œ tÑ+ð	bð
 ” 4Ñ'ðbð bð bð bðP &*Ø.2ð	D%ð D%ð ˜d‘lðD%ð œ tÑ+ð	D%ð D%ð D%ð D%ðN Ð˜HÑ%Ô%ð &*Ø.2Ø*.ðOð Oð ˜d‘lðOð œ tÑ+ð	Oð
 ” 4Ñ'ðOð Oð Oñ &Ô%ðOð Oð Oð Oð Or:   rZ   c                   ó   — e Zd ZdS )ÚBambaMLPNr?   r9   r:   r;   r  r  y  r@   r:   r  c                   ó   — e Zd ZdS )ÚBambaRMSNormNr?   r9   r:   r;   r  r  }  r@   r:   r  c                   ó  ‡ — e Zd Zddededefˆ fd„Z	 	 	 	 	 ddej        d	ej        dz  d
ej	        dz  de
dz  dedz  deej        ej        f         dz  dee         deej        eej        ej        f         dz  f         fd„Zˆ xZS )ÚBambaDecoderLayerÚlinear_attentionr[   r\   r�   c                 ó6  •— t          ¦   «                              ||¦  «         | `d}|dk    rt          nd } ||¦  «        | _        || _        |dk    rt          ||¬¦  «        | _        d S |dk    rt          ||¦  «        | _        d S t          d|›�¦  «        ‚)Nr'   r"  )r[   r\   Úfull_attentionzInvalid layer_type: )
ri   rj   Ú	self_attnr  Úfeed_forwardÚ
block_typerZ   ÚmambarV   Ú
ValueError)rž   r[   r\   r�   Únum_expertsÚffn_layer_classr£   s         €r;   rj   zBambaDecoderLayer.__init__‚  s¯   ø€ Ý‰Œ×Ò˜ Ñ+Ô+Ð+àˆNàˆØ&1°QÒ&6Ð&6�(˜(¸DˆØ+˜O¨FÑ3Ô3ˆÔà$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Cr:   NFr¤   r¦   Úposition_idsÚpast_key_valuesÚ	use_cacheÚposition_embeddingsr  Úreturnc           
      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¤   r¥   r¦   r$  )r¤   r¦   r,  r-  r.  r/  r9   )Úinput_layernormr'  r(  r%  Úpre_ff_layernormr&  )
rž   r¤   r¦   r,  r-  r.  r/  r  ÚresidualÚself_attn_weightss
             r;   r  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$7ð0ð 0ð ð0ð 0Ñ,ˆMÐ,ð ! =Ñ0ˆà ˆØ×-Ò-¨mÑ<Ô<ˆØ×)Ò)¨-Ñ8Ô8ˆØ  =Ñ0ˆàÐ/Ð/Ð/r:   )r"  )NNNFN)r0   r1   r2   r(   r7   Ústrrj   r4   r  r5   r   ÚboolÚtupler   r*   ÚFloatTensorr  r  r  s   @r;   r!  r!  �  s.  ø€ € € € € ðDð D˜{ð D°sð DÈð Dð Dð Dð Dð Dð Dð( /3Ø04Ø(,Ø!&ØHLð(0ð (0à”|ð(0ð œ tÑ+ð(0ð Ô&¨Ñ-ð	(0ð
  ™ð(0ð ˜$‘;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 (0r:   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!  r-  )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'   )ri   Ú_init_weightsÚ
isinstancerZ   ÚinitÚones_r�   Úcopy_r�   r4   r�   rŽ   rl   r’   )rž   Úmoduler£   s     €r;   r?  z"BambaPreTrainedModel._init_weightsÎ  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 	!r:   )r0   r1   r2   r(   r6   Úbase_model_prefixÚsupports_gradient_checkpointingÚ_no_split_modulesÚ_skip_keys_device_placementÚ_supports_flash_attnÚ_supports_sdpaÚ_is_statefulÚ_can_compile_fullgraphr!  rV   Ú_can_record_outputsr4   Úno_gradr?  r  r  s   @r;   r;  r;  ¾  s¡   ø€ € € € € € àÐÐÑØÐØ&*Ð#Ø,Ð-ÐØ#4Ð"5ÐØÐØ€NØ€LØ!Ðà*Ø$ðð Ðð
 €U„]�_„_ð!ð !ð !ð !ñ „_ð!ð !ð !ð !ð !r:   r;  c                   óâ   ‡ — e Zd Zdefˆ fd„Zeee	 	 	 	 	 	 ddej	        dz  dej
        dz  dej	        dz  dedz  dej        dz  d	edz  d
ee         defd„¦   «         ¦   «         ¦   «         Zˆ xZS )Ú
BambaModelr[   c           	      óJ  •— t          ¦   «                              |¦  «         |j        | _        |j        | _        t          j        |j        |j        | j        ¦  «        | _        g }t          |j
        ¦  «        D ]2}|                     t          |||j        |         ¬¦  «        ¦  «         Œ3t          j        |¦  «        | _        |j        | _        t#          |j        |j        ¬¦  «        | _        t)          |¬¦  «        | _        d| _        |                      ¦   «          d S )N)r\   r�   rd   ©r[   F)ri   rj   Úpad_token_idÚpadding_idxÚ
vocab_sizer   Ú	Embeddingrm   Úembed_tokensÚrangeÚnum_hidden_layersÚappendr!  Úlayers_block_typeÚ
ModuleListr¼   Ú_attn_implementationr  r{   Úfinal_layernormr>   Ú
rotary_embÚgradient_checkpointingÚ	post_init)rž   r[   Údecoder_layersÚir£   s       €r;   rj   zBambaModel.__init__Ù  s	  ø€ Ý‰Œ×Ò˜Ñ Ô Ð Ø!Ô.ˆÔØ Ô+ˆŒåœL¨Ô):¸FÔ<NÐPTÔP`ÑaÔaˆÔØˆÝ�vÔ/Ñ0Ô0ð 	rð 	rˆAØ×!Ò!Õ"3°FÀaÐTZÔTlÐmnÔToÐ"pÑ"pÔ"pÑqÔqÐqÐqÝ”m NÑ3Ô3ˆŒà$*Ô$?ˆÔ!Ý+¨FÔ,>ÀFÔDWÐXÑXÔXˆÔÝ.°fÐ=Ñ=Ô=ˆŒà&+ˆÔ#à�ŠÑÔÐÐÐr:   NÚ	input_idsr¦   r,  r-  Úinputs_embedsr.  r  r0  c           
      ó„  — |d u |d uz  rt          d¦  «        ‚|€|                      |¦  «        }|}|r|€t          | j        ¬¦  «        }|€9t	          j        |j        d         |j        ¬¦  «                             d¦  «        }t          |x}	t          ¦  «        s&| j        ||||dœ}
t          di |
¤Žt          di |
¤Ždœ}	|                      ||¬¦  «        }t          | j        ¦  «        D ]-\  }} ||f|	| j        j        |                  ||||d	œ|¤Ž\  }}Œ.|                      |¦  «        }t%          ||¬
¦  «        S )Nz:You must specify exactly one of input_ids or inputs_embedsrR  r'   râ   r   )r[   re  r¦   r-  r,  )r$  r"  )r,  )r¦   r,  r-  r.  r/  )Úlast_hidden_stater-  r9   )r)  rW  r	   r[   r4   rŽ   rF   rã   rE   r@  Údictr   r   r_  Ú	enumerater¼   r[  r^  r   )rž   rd  r¦   r,  r-  re  r.  r  r¤   Úcausal_mask_mappingÚmask_kwargsr/  rc  Údecoder_layerÚattn_weightss                  r;   r  zBambaModel.forwardì  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$7ð+ð +ð ð+ð +Ñ'ˆM˜<˜<ð ×,Ò,¨]Ñ;Ô;ˆå&Ø+Ø+ð
ñ 
ô 
ð 	
r:   )NNNNNN)r0   r1   r2   r(   rj   r   r   r   r4   r5   r  r   r9  r7  r   r*   r   r  r  r  s   @r;   rP  rP  ×  s  ø€ € € € € ð˜{ð ð ð ð ð ð ð&  ØØð .2Ø.2Ø04Ø(,Ø26Ø!%ð7
ð 7
àÔ# dÑ*ð7
ð œ tÑ+ð7
ð Ô&¨Ñ-ð	7
ð
  ™ð7
ð Ô(¨4Ñ/ð7
ð ˜$‘;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
r:   rP  c                   ó   ‡ — e Zd 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BambaForCausalLMc                 óŠ   •— t          ¦   «                              |¦  «         |j        | _        |                      ¦   «          d S )N)ri   rj   Úz_loss_coefficientra  )rž   r[   r£   s     €r;   rj   zBambaForCausalLM.__init__*  s>   ø€ Ý‰Œ×Ò˜Ñ Ô Ð Ø"(Ô";ˆÔð 	�ŠÑÔÐÐÐr:   Nr   rd  r¦   r,  r-  re  Úlabelsr.  Úlogits_to_keepr0  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."
        ```)rd  r¦   r,  r-  re  r.  N)Úlogitsrr  rU  r   rB   rC   r¨   r   )Úlossru  r-  r¤   r=  r9   )r<  rg  r@  r7   ÚsliceÚlm_headÚloss_functionr[   rU  rq  Ú	logsumexprÄ   r©   ÚpowÚmeanr   r-  r¤   r=  )rž   rd  r¦   r,  r-  re  rr  r.  rs  r  Úoutputsr¤   Úslice_indicesru  rv  Úz_losss                   r;   r  zBambaForCausalLM.forward1  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 	
r:   TFc           
      óh   •— | j         j        |d<    t          ¦   «         j        |f||||||dœ|¤Ž}	|	S )Nrs  )r-  r¦   re  r,  r.  Úis_first_iteration)r[   Únum_logits_to_keepri   Úprepare_inputs_for_generation)rž   rd  r-  r¦   re  r,  r.  r�  r  Úmodel_inputsr£   s             €r;   rƒ  z.BambaForCausalLM.prepare_inputs_for_generationr  s^   ø€ ð $(¤;Ô#AˆÐÑ Ø<•u‘w”wÔ<Øð	
à+Ø)Ø'Ø%ØØ1ð	
ð 	
ð ð	
ð 	
ˆð Ðr:   )NNNNNNNr   )NNNNTF)r0   r1   r2   rj   r   r   r4   r5   r  r   r9  r7  r7   r   r  rƒ  r  r  s   @r;   ro  ro  )  sC  ø€ € € € € ðð ð ð ð ð Øð .2Ø.2Ø04Ø(,Ø26Ø*.Ø!%Ø-.ð=
ð =
àÔ# dÑ*ð=
ð œ tÑ+ð=
ð Ô&¨Ñ-ð	=
ð
  ™ð=
ð Ô(¨4Ñ/ð=
ð Ô  4Ñ'ð=
ð ˜$‘;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 r:   ro  )rP  ro  r;  )r'   )Ir3   Útypingr   r4   r   Ú r   rA  Úactivationsr   Úcache_utilsr   r	   Úintegrations.accelerater
   Úintegrations.hub_kernelsr   Úmasking_utilsr   r   Úmodeling_outputsr   r   Úmodeling_utilsr   Úprocessing_utilsr   Úutilsr   r   r   r   Úutils.genericr   r   Úutils.import_utilsr   Úutils.output_capturingr   Újamba.modeling_jambar   Úllama.modeling_llamar   r   r   r   r    r!   Úmamba2.modeling_mamba2r"   r#   r$   r%   r&   Úconfiguration_bambar(   Ú
get_loggerr0   rš   r*   r>   rT   rV   rX   ÚModulerZ   r  r  r!  r;  rP  ro  Ú__all__r9   r:   r;   ú<module>rš     sj  ðð& Ð à 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 8Ð 8Ð 8Ð 8Ð 8Ð 8Ø PÐ PÐ PÐ PÐ PÐ PÐ PÐ PØ OÐ OÐ OÐ OÐ OÐ OÐ OÐ OØ -Ð -Ð -Ð -Ð -Ð -Ø &Ð &Ð &Ð &Ð &Ð &Ø XÐ XÐ XÐ XÐ XÐ XÐ XÐ XÐ XÐ XÐ XÐ XØ MÐ MÐ MÐ MÐ MÐ MÐ MÐ MØ 9Ð 9Ð 9Ð 9Ð 9Ð 9Ø 5Ð 5Ð 5Ð 5Ð 5Ð 5Ø =Ð =Ð =Ð =Ð =Ð =ðð ð ð ð ð ð ð ð ð ð ð ð ð ð ð ðð ð ð ð ð ð ð ð ð ð ð ð ð ð -Ð ,Ð ,Ð ,Ð ,Ð ,ð 
ˆÔ	˜HÑ	%Ô	%€ðð ð ð ð  	°ð ñ ô ð ð0	ð 	ð 	ð 	ð 	Ð/ñ 	ô 	ð 	ð
#ð #ð #ð #ðL ð	ð 	ð 	ð 	ð 	�^ñ 	ô 	ñ Ôð	ð	ð 	ð 	ð 	ð 	Ð)ñ 	ô 	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 	ˆxñ 	ô 	ð 	ð	ð 	ð 	ð 	ð 	�<ñ 	ô 	ð 	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 `ð `ð `ð `Ð'ñ `ô `ð `ðF EÐ
DÐ
D€€€r:   