+
    "Ý³j5 ã                   ó  a € R# t>0 t ^ RIt^ RIt^ RIt^ RIt^ RIHtHtHt ^ RI	t	^ RI
Ht ^ RIt^ RIHt ^RIHt ^RIHt ^RIHtHt ^RIHt ^RIHt ^R	IHt ^R
IHt ^RIHtH t  ^RI!H"t"H#t#H$t$H%t%H&t& ^RI'H(t( ^RI)H*t*H+t+ ]! 4       t,R R lt-]P\                  ! 4       t// t0] ^ k / t1] ^k / t2] ^k R R lt3R R lt4R R lt5R R lt6R R lt7]Pp                  R R l4       t9R R lt:R R  lt; ! R! R"4      t<R".t=R# )$é    N)ÚDictÚOptionalÚTuple)Ú_ConvNd)ÚNamedModule)ÚQuantizeConfig)ÚFallbackStrategyÚ	SmoothMSE)Ú
get_device©Úenv_flag)Úsetup_logger)Ú
torch_sync)Úmse_optimal_quantÚsmooth_block)Úcompose_final_permÚcompute_global_permÚcompute_local_permsÚextend_perm_with_tailÚinvert_perm)Únpu_inverse_cholesky_factor)Ú
HF_OPTIMUMÚ	Quantizerc                ó$   € V ^8„  d   QhR\         /# )é   Úreturn)Úbool)Úformats   "Ú\/home/luca/screen-checking/.venv/lib/python3.14/site-packages/gptqmodel/quantization/gptq.pyÚ__annotate__r    )   s   € ÷ -ñ -œdñ -ó    c                 ó   € \        R4      # )zÕVerbose per-module hessian-inverse markers are opt-in.

The QuantizationRegionTimer measurement is always recorded and flushed
periodically, which provides aggregate stall isolation without per-module
line noise.
ÚGPTQMODEL_LOG_HESSIANr   © r!   r   Ú_log_hessian_verboser%   )   s   € ô Ð+Ó,Ð,r!   c                ó|   € V ^8„  d   QhR\         P                  R\        \        \        \
        ,          3,          /# ©r   Údevicer   ©Útorchr(   r   Ústrr   Úint)r   s   "r   r    r    ?   s,   € ÷ ñ œeŸl™lð ¬u´S¼(Ä3½-Ð5GÕ/Hñ r!   c                 ó^   € \         P                  ! V 4      pVP                  VP                  3# ©N©r*   r(   ÚtypeÚindex©r(   Údevs   & r   Ú_device_cache_keyr4   ?   s#   € Ü
�,Š,�vÓ
€CØ�8‰8�S—Y‘YÐÐr!   c                óf   € V ^8„  d   QhR\         P                  R\        \        \        3,          /# r'   )r*   r(   r   r+   r,   )r   s   "r   r    r    D   s&   € ÷ <ñ <¤%§,¡,ð <´5¼¼c¸µ?ñ <r!   c                ó€   € \         P                  ! V 4      pVP                  VP                  f   R3# VP                  3# )z@Order reductions by device identity, never worker arrival order.éÿÿÿÿr/   r2   s   & r   Ú_device_reduction_keyr8   D   s4   € ä
�,Š,�vÓ
€CØ�8‰8˜3Ÿ9™9Ò,�RÐ;Ð;°#·)±)Ð;Ð;r!   c                ó|   € V ^8„  d   QhR\         P                  R\        \        \        \
        ,          3,          /# r'   r)   )r   s   "r   r    r    J   s,   € ÷ %ñ %¤§¡ð %´%¼¼XÄc½]Ð8JÕ2Kñ %r!   c                 ó   € \        V 4      # r.   )r4   ©r(   s   &r   Ú_workspace_cache_keyr<   J   s   € Ü˜VÓ$Ð$r!   c                ó|   € V ^8„  d   QhR\         \        \        \        ,          3,          R\        P
                  /# )r   Úkeyr   )r   r+   r   r,   Ú	threadingÚLock)r   s   "r   r    r    N   s,   € ÷ 
ñ 
œœs¤H¬S¥MÐ1Õ2ð 
´y·~±~ñ 
r!   c                 ó   € \         P                  V 4      pVe   V# \        ;_uu_ 4        \         P                  V 4      pVf   \        P                  ! 4       pV\         V &   R R R 4       V#   + '       g   i     T# ; ir.   )Ú_WORKSPACE_LOCKSÚgetÚ_WORKSPACE_LOCKS_GUARDr?   r@   )r>   Úlocks   & r   Ú_workspace_lockrF   N   sk   € Ü×Ñ Ó$€DØÒØˆç	Ö	Ü×#Ñ# CÓ(ˆØŠ<Ü—>’>Ó#ˆDØ$(Ô˜SÑ!÷	 
 ð
 €K÷ 
 Ö	ð
 €Kús   ª8A,Á,A=	c          
      ó’   € V ^8„  d   QhR\         \        P                  ,          R\        P                  R\        R\        R\
        /# )r   Ú	workspaceÚdtypeÚrequired_rowsÚcolsr   )r   r*   ÚTensorrI   r,   r   )r   s   "r   r    r    [   sG   € ÷ ñ ÜœŸ™Õ%ðä�;‰;ðô ðô ð	ô
 
ñr!   c                 óÆ   € V f   R# V P                   ^8w  d   R# V P                  V8w  d   R# V P                  ^,          V8w  d   R# V P                  ^ ,          V8  d   R# R# )NTF)ÚndimrI   Úshape)rH   rI   rJ   rK   s   &&&&r   Ú_needs_workspace_resizerP   [   sV   € ð ÒÙØ‡~�~˜ÔÙØ‡�˜%ÔÙØ‡��qÕ˜TÔ!ÙØ‡��qÕ˜MÔ)ÙÙr!   c                ó²   € V ^8„  d   QhR\         P                  R\         P                  R\        R\        R\        \         P
                  \        3,          /# )r   r(   rI   rK   rJ   r   )r*   r(   rI   r,   r   rL   r   )r   s   "r   r    r    o   sP   € ÷ .ñ .Ü�L‰Lð.ä�;‰;ð.ô ð.ô ð	.ô
 Œ5�<‰<œÐÕñ.r!   c           	   #   ó   "  € \        V 4      p\        V4      pT;_uu_ 4        \        P                  VR 4      pVR J;'       d    \	        VVVV4      '       * pV'       g&   \        V^4      p\        P                  ! W‚3WR7      pR R R 4        XX3x € V;_uu_ 4        V\        V&   R R R 4       R #   + '       g   i     L5; i  + '       g   i     R # ; i  T;_uu_ 4        X\        T&   R R R 4       i   + '       g   i     i ; i; i5i)N©rI   r(   )r<   rF   Ú_WORKSPACE_CACHEÚpoprP   Úmaxr*   Úempty)	r(   rI   rK   rJ   r>   rE   rH   ÚreusedÚrowss	   &&&&     r   Ú_lease_workspacerZ   n   sÚ   é € ô ˜vÓ
&€CÜ˜3Ó€Dß	ŠÜ$×(Ñ(¨¨dÓ3ˆ	Ø $Ð&÷ 
ð 
Ô/FØØØØó	0
ô ,
ˆ÷ Ü�} aÓ(ˆDÜŸš T L¸ÔMˆI÷ 
ð.Ø˜ÐÒçŠTØ$-Ô˜SÑ!÷ ‰T÷ 
�ú÷ �TˆTû�TŠTØ$-Ô˜SÑ!÷ �T�TˆTÿsj   ‚!D£=B3Á!%B3ÂDÂC Â
DÂ
CÂ)
DÂ3C	Â>DÃC	Ã	DÃDÃ%
C8	Ã/	DÃ8DÄDÄDc                óD   € V ^8„  d   QhR\         P                  R\        /# r'   )r*   r(   r   )r   s   "r   r    r    ‰   s   € ÷ ñ ¤e§l¡lð ´tñ r!   c                 ó¼  € \        V 4      p\        P                  V4      pVe   V# \        P                  ! V 4      pVP
                  R8X  d   R\        V&   R#  \        P                  ! R\        P                  VR7      p\        P                  ! R\        P                  VR7      p\        P                  ! WE4      pRpV\        V&   V#   \         d    Rp Li ; i)NÚmetaFrS   T)é   r^   )
r4   Ú_BF16_SUPPORT_CACHErC   r*   r(   r0   ÚzerosÚbfloat16ÚmatmulÚ	Exception)r(   Ú	cache_keyÚcachedr3   ÚaÚbÚ_Úsupports   &       r   Ú_device_supports_bfloat16rj   ‰   s»   € Ü! &Ó)€IÜ ×$Ñ$ YÓ/€FØÒØˆä
�,Š,�vÓ
€CØ
‡x�x�6ÔØ).Ô˜IÑ&ÙðÜ�KŠK˜¤e§n¡n¸SÔAˆÜ�KŠK˜¤e§n¡n¸SÔAˆÜ�LŠL˜ÓˆØˆð &-Ô˜	Ñ"Ø€Nøô	 ô ØŠðús   ÁA&C ÃCÃCc                ó8   € V ^8„  d   QhR\         P                  /# )r   Úlayer©ÚnnÚModule)r   s   "r   r    r        s   € ÷ Hñ H¤r§y¡yñ Hr!   c                 óJ  € \        V \        4      '       d   V P                  p \        V \        P                  4      '       d;   V P
                  P                  ^,          V P
                  P                  ^ ,          3# \        V \        P                  4      '       d;   V P
                  P                  ^,          V P
                  P                  ^ ,          3# V P
                  P                  ^ ,          \        P                  ! V P
                  P                  R,          4      3# )r^   :r^   NN)Ú
isinstancer   Úmodulern   Ú	EmbeddingÚweightrO   ÚtransformersÚConv1DÚmathÚprod)rl   s   &r   Úget_number_of_rows_and_colsry       sÂ   € ä�%œ×%Ò%Ø—‘ˆä�%œŸ™×&Ò&Ø�|‰|×!Ñ! !Õ$ e§l¡l×&8Ñ&8¸Õ&;Ð;Ð;ä�%œ×,Ñ,×-Ò-à�|‰|×!Ñ! !Õ$ e§l¡l×&8Ñ&8¸Õ&;Ð;Ð;ð �|‰|×!Ñ! !Õ$¤d§i¢i°·±×0BÑ0BÀ2Õ0FÓ&GÐGÐGr!   c                   óô  a € ] tR t^°t o ]V 3R lR l4       tR>V 3R lR lltV 3R lR lt]R 4       tV 3R	 lR
 lt	R t
V 3R lR ltV 3R lR ltR?V 3R lR llt]V 3R lR l4       tR@V 3R lR lltV 3R lR ltV 3R lR lt]P&                  V 3R lR l4       tV 3R lR ltV 3R lR  ltV 3R! lR" ltR@V 3R# lR$ lltV 3R% lR& ltR@V 3R' lR( lltR) tR@V 3R* lR+ lltRAR, ltRBV 3R- lR. llt]P@                  ! 4       V 3R/ lR0 l4       t!]P@                  ! 4       RCR1 l4       t"]P@                  ! 4       RCR2 l4       t#RDV 3R3 lR4 llt$V 3R5 lR6 lt%R7R/V 3R8 lR9 llt&V 3R: lR; lt'R< t(R=t)V t*R# )EÚGPTQc                óN   <€ V ^8„  d   QhRS[ P                  RS[ P                  /# )r   rr   r   rm   )r   Ú__classdict__s   "€r   r    ÚGPTQ.__annotate__²   s#   ø€ ÷ ñ ¡b§i¡ið ±B·I±Iñ r!   c                óÄ   € \        V \        4      '       dJ   V P                  P                  R4      p\        V\        P
                  4      '       d   V# V P                  # V # )zCResolve the dense module view GPTQ should quantize for one wrapper.Úquant_source_module)rq   r   ÚstaterC   rn   ro   rr   )rr   Úquant_sources   & r   Úresolve_module_sourceÚGPTQ.resolve_module_source±   sL   € ô �fœk×*Ò*Ø!Ÿ<™<×+Ñ+Ð,AÓBˆLÜ˜,¬¯	©	×2Ò2Ø#Ð#Ø—=‘=Ð Øˆr!   Nc                óJ   <€ V ^8„  d   QhRS[ P                  RS[S[,          /# )r   rr   Úqcfg)rn   ro   r   r   )r   r}   s   "€r   r    r~   ¼   s,   ø€ ÷ eEñ eE™rŸy™yð eE±¹Õ0Hñ eEr!   c                ó*  € \         P                  ! 4       V n        W0n        V P	                  V4      p\        V4      w  V n        V n        \        V\        4      '       d   W@n
        VP                  V n        Wn        M\        V n        W@n
        R V n        V P                  V n        V P                  V n        V P                  e'   V P                  P                   P#                  R4      pM\%        V P                  RR 4      p\        V\&        4      '       d3   \)        VP#                  R^ 4      ;'       g    ^ 4      p\+        V^ 4      pMR p^ pWPn        W`n        V P.                  '       d$   V ;P                  V P.                  ,          un        \1        V P                  4      p\3        V P                  RV4       VP4                  R8X  d   \6        P8                  ! R4      V n        M\6        P8                  ! V4      V n        V P=                  V P                  4       V'       d   TM	\?        4       V n         V PC                  4        R V n"        R V n#        ^ V n$        V PK                  V P                  R7      V n&        ^ V n'        V P@                  PP                  V n(        R V n)        R V n#        R V n*        / V n+        / V n,        / V n-        ^ V n.        RV n/        R	^ R
^ R^ R^ R^ R^ R^ /V n0        R	^ R^ R^ R^ R^ /V n1        R V n2        R V n3        R V n4        R # )NÚtp_pad_infoÚ_tp_pad_infoÚpad_colsÚtarget_devicer]   Úcpu)ÚnameFÚrequestsÚstaging_requestsÚstaging_hitsÚstaging_missesÚmaterialized_requestsÚmaterialized_hitsÚmaterialized_misses)5r?   ÚRLockrE   Úregion_timerrƒ   ry   rY   Úcolumnsrq   r   rr   r�   Ú_named_moduler   Ú_original_rowsÚ_original_columnsr�   rC   ÚgetattrÚdictr,   rV   r‰   Ú_tp_pad_colsr   Úsetattrr0   r*   r(   Ú_final_hessian_device_hintÚvalidate_moduler   r†   Ú_validate_act_group_aware_shapeÚmodule_copyÚHÚnsamplesÚcreate_quantizerÚ	quantizerÚfwd_counterÚfallbackÚexpected_nsamplesÚ_H_diagÚ_device_hessian_partialsÚ_device_embedding_countsÚ_device_sample_countsÚ_hessian_total_samplesÚ_hessian_dirtyÚ_borrow_workspace_statsÚ_borrow_workspace_totalsÚ_borrow_workspace_last_summaryÚ_borrow_workspace_stage_dtypeÚ!_borrow_workspace_last_chunk_rows)Úselfrr   r†   r–   Úresolved_moduleÚpad_inforŠ   Úmodule_devices   &&&&    r   Ú__init__ÚGPTQ.__init__¼   s¨  € ô —O’OÓ%ˆŒ	Ø(Ôð ×4Ñ4°VÓ<ˆÜ"=¸oÓ"NÑˆŒ	�4”<Ü�fœk×*Ò*Ø)ŒKØŸ™ˆDŒIØ!'Õä"ˆDŒIØ)ŒKØ!%ˆDÔà"Ÿi™iˆÔØ!%§¡ˆÔØ×ÑÒ)Ø×)Ñ)×/Ñ/×3Ñ3°MÓB‰Hä˜tŸ{™{¨N¸DÓAˆHÜ�h¤×%Ò%Ü˜8Ÿ<™<¨
°AÓ6×;Ð;¸!Ó<ˆHÜ˜8 QÓ'‰HàˆHØˆHà$ÔØ$ÔØ××ÐØ�LŠL˜D×-Ñ-Õ-�Lä" 4§;¡;Ó/ˆÜ�—‘˜_¨mÔ<à×Ñ Ô'Ü.3¯lªl¸5Ó.AˆDÕ+ä.3¯lªl¸=Ó.IˆDÔ+à×Ñ˜TŸ[™[Ô)ç ‘D¤nÓ&6ˆŒ	Ø×,Ñ,Ô.àˆÔàˆŒØˆŒà×.Ñ.°D·I±IÐ.Ó>ˆŒð ˆÔàŸ	™	×*Ñ*ˆŒØ26ˆÔà)-ˆŒØ/3ˆŒð KMˆÔ%ØJLˆÔ%Ø>@ˆÔ"Ø+,ˆÔ#Ø$)ˆÔð ˜Ø Ø˜AØ˜aØ# QØ Ø! 1ð(
ˆÔ$ð ˜Ø Ø! 1Ø˜AØ˜að)
ˆÔ%ð LPˆÔ+ØDHˆÔ*Ø@DˆÖ.r!   c                ó   <€ V ^8„  d   QhRR/# ©r   r   Nr$   )r   r}   s   "€r   r    r~   #  s   ø€ ÷ 	ñ 	°ñ 	r!   c                óÞ   € \        V P                  R R4      '       g   R# \        \        V P                  RR4      ;'       g    R4      pV^ 8:  d   \        RV P                   RV R24      hR# )Úact_group_awareFNÚ
group_sizeúQuantization: Module `z<` -> `act_group_aware=True` requires `group_size > 0`, got `ú`.r7   )r›   r†   r,   Ú
ValueErrorr�   )rµ   r¿   s   & r   r¡   Ú$GPTQ._validate_act_group_aware_shape#  sm   € Ü�t—y‘yÐ"3°U×;Ò;Ùäœ §¡¨L¸"Ó=×CÐCÀÓDˆ
Ø˜Œ?ÜØ(¨¯©¨ð 4Ø"�| 2ð'óð ñ r!   c                óÖ   € \        V \        P                  \        P                  \        P                  \        P
                  \        P                  34      '       g   Q R V  R24       hR# )z<We supports only linear and convolutional layers. actual = `Ú`N)rq   rn   rs   ÚLinearÚConv1dÚConv2dru   rv   )rr   s   &r   r    ÚGPTQ.validate_module.  sb   € ä˜&¤2§<¡<´·±¼B¿I¹IÄrÇyÁyÜ#/×#6Ñ#6ð#8÷ 9ò 9ð 	CØ<xÐyð  yAð  ABð  ;Có	Cò 9r!   c                ó&   <€ V ^8„  d   QhRS[ RS[/# )r   r�   r   )r+   r   )r   r}   s   "€r   r    r~   6  s   ø€ ÷ 4ñ 4¡Sð 4©Yñ 4r!   c                ó0   € \        V P                  VR 7      # ))r†   r�   )r   r†   )rµ   r�   s   &&r   r¥   ÚGPTQ.create_quantizer6  s   € Ü˜dŸi™i¨dÔ3Ð3r!   c                ój   € \        V R 4      '       d!   V P                  P                  P                  # R# )rr   )r   r   )Úhasattrrr   rt   rO   )rµ   s   &r   rO   Ú
GPTQ.shape9  s*   € Ü�4˜×"Ò"Ø—;‘;×%Ñ%×+Ñ+Ð+àˆMr!   c                ó4   <€ V ^8„  d   QhRS[ P                  /# ©r   r£   ©r*   rL   )r   r}   s   "€r   r    r~   ?  s   ø€ ÷ ñ ¡e§l¡lñ r!   c                ó¸   € V P                   P                  p\        P                  ! VP                  ^ ,          \        P
                  VP                  R7      pW23# )z%Mock hessian inverse for fast testingrS   )r†   Údamp_percentr*   ÚeyerO   Úfloat32r(   )rµ   r£   ÚdampÚidentitys   &&  r   Úmock_hessian_inverseÚGPTQ.mock_hessian_inverse?  s;   € à�y‰y×%Ñ%ˆä—9’9˜QŸW™W Q�Z¬u¯}©}ÀQÇXÁXÔNˆØˆ~Ðr!   c                ó>   <€ V ^8„  d   QhRS[ RS[P                  RR/# )r   ÚstageÚsource_devicer   N)r+   r*   r(   )r   r}   s   "€r   r    r~   F  s&   ø€ ÷ 	
ñ 	
¡cð 	
¹%¿,¹,ð 	
È4ñ 	
r!   c                óJ   € \         P                  RV P                  VV4       R# )z=Explain when a memory-heavy GPTQ step moves from CUDA to CPU.z“Quantization: Module `%s` -> CUDA OOM during %s on %s; falling back to CPU. Due to this fallback, the calculation may take much longer than normal.N)ÚlogÚwarnr�   )rµ   rÜ   rÝ   s   &&&r   Úlog_cpu_fallbackÚGPTQ.log_cpu_fallbackF  s#   € ô 	�‰ðVà�I‰IØØö	
r!   Tc                ó4   <€ V ^8„  d   QhRS[ P                  /# )r   r(   )r*   r(   )r   r}   s   "€r   r    r~   Q  s   ø€ ÷ ñ ©e¯l©lñ r!   c                ó.  € V'       g+   V P                   P                  P                  P                  pV P                   P                  P                  P	                  WR 7      p\        V P                   \        P                  4      '       d   VP                  4       p\        V P                   \        4      '       d   VP                  ^4      p\        V P                   \        P                  P                  4      '       d   VP                  4       pV P                  '       dd   \        P                   ! VP"                  ^ ,          V P                  3VP$                  VP                  R7      p\        P&                  ! W43^R7      pVP)                  4       # ))Úcopyr(   rS   ©Údim)rr   rt   Údatar(   Útorq   rn   rs   Útr   Úflattenru   Úpytorch_utilsrv   r�   r*   r`   rO   rI   ÚcatÚfloat)rµ   rå   r(   ÚcloneÚpads   &&&  r   Úclone_moduleÚGPTQ.clone_moduleQ  s  € ßØ—[‘[×'Ñ'×,Ñ,×3Ñ3ˆFà—‘×"Ñ"×'Ñ'×*Ñ*°Ð*ÓDˆä�d—k‘k¤2§<¡<×0Ò0Ø—G‘G“IˆEä�d—k‘k¤7×+Ò+Ø—M‘M !Ó$ˆEä�d—k‘k¤<×#=Ñ#=×#DÑ#D×EÒEØ—G‘G“IˆEà××ÐÜ—+’+Ø—‘˜Q• ×!2Ñ!2Ð3Ø—k‘kØ—|‘|ôˆCô
 —I’I˜u˜l°Ô2ˆEà�{‰{‹}Ðr!   c                óT   <€ V ^8„  d   QhRS[ P                  RS[RS[ P                  /# )r   ÚtensorÚlengthr   ©r*   rL   r,   )r   r}   s   "€r   r    r~   k  s/   ø€ ÷ Eñ E¡%§,¡,ð E¹ð EÁÇÁñ Er!   c                ó  € V P                  4       ^ 8X  d   V # \        WP                  R,          4      pW P                  R,          8X  d   V # V P                  V P                  4       ^,
          ^ V4      P	                  4       # )r   r7   )rç   ÚminrO   ÚnarrowÚ
contiguous)rô   rõ   Útrims   && r   Útruncate_last_dimÚGPTQ.truncate_last_dimj  sc   € à�:‰:‹<˜1ÔØˆMä�6Ÿ<™<¨Õ+Ó,ˆØ—<‘< Õ#Ô#ØˆMà�}‰}˜VŸZ™Z›\¨AÕ-¨q°$Ó7×BÑBÓDÐDr!   c                ód   <€ V ^8„  d   QhRS[ P                  RS[ P                  RS[S[,          /# )r   ÚinpÚoutÚbatch_index)r*   rL   r   r,   )r   r}   s   "€r   r    r~   u  s/   ø€ ÷ &'ñ &'™UŸ\™\ð &'±·±ð &'É8ÑTWÍ=ñ &'r!   c                ó¦  € \        V P                  \        P                  4      '       Ed   ??V P	                  V4      w  rEpV^ 8X  g   Vf   R# \
        P                  ! V4      pV P                  ;_uu_ 4        V ;P                  ^,          un        V P                  P                  V4      pVf   WPP                  V&   MVP                  V4       V P                  P                  V^ 4      V,           V P                  V&   V ;P                  V,          un        RV n        RRR4       R# V P	                  V4      w  ršpV	^ 8X  g   V
f   R# \
        P                  ! V4      pV P                  ;_uu_ 4        V ;P                  ^,          un        V P                  P                  V4      pVf   W P                  V&   MVP                  V
4       ?
V P                  P                  V^ 4      V	,           V P                  V&   V ;P                  V	,          un        RV n        RRR4       R#   + '       g   i     R# ; i  + '       g   i     R# ; i)r   NT)rq   rr   rn   rs   Úprocess_batchr*   r(   rE   r§   r¬   rC   Úadd_r­   r¤   r¯   r«   )rµ   rÿ   r   r  Útoken_countÚcountsr(   r3   ÚexistingÚbatch_token_sizeÚxtxs   &&&&       r   Ú	add_batchÚGPTQ.add_batchu  sÎ  € Ü�d—k‘k¤2§<¡<×0Ó0Ø�[Ø*.×*<Ñ*<¸SÓ*AÑ'ˆK Ø˜aÔ 6¢>Ùä—,’,˜vÓ&ˆCØ——“Ø× Ò  AÕ%Õ Ø×8Ñ8×<Ñ<¸SÓA�ØÒ#Ø9?×1Ñ1°#Ò6à—M‘M &Ô)Ø26×2LÑ2L×2PÑ2PÐQTÐVWÓ2XÐ[fÕ2f�×*Ñ*¨3Ñ/Ø—’ Õ,•Ø&*�Ô#÷ ñ à(,×(:Ñ(:¸3Ó(?Ñ%Ð˜vØ˜qÔ  C¢KÙä�lŠl˜6Ó"ˆà�Y�Y‹YØ×Ò Õ!Õà×4Ñ4×8Ñ8¸Ó=ˆHØÒØ58×-Ñ-¨cÒ2à—‘˜cÔ"Øà.2×.HÑ.H×.LÑ.LÈSÐRSÓ.TÐWgÕ.gˆD×&Ñ& sÑ+Ø�MŠMÐ-Õ-�MØ"&ˆDÔ÷ ‰Y÷% –ñ ú÷ �YˆYús   Á9B)H+Å7B*H?È+H<	È?I	c                óh   <€ V ^8„  d   QhRS[ P                  RS[ P                  RS[ P                  /# )r   Úinput_dtyper(   r   )r*   rI   r(   )r   r}   s   "€r   r    r~   �  s1   ø€ ÷ ñ ±5·;±;ð ÉÏÉð ÑY^×YdÑYdñ r!   c                ó  € \         P                  ! V4      pV P                  P                  P                  pV\         P
                  8X  d   \         P
                  # V\         P                  \         P                  39  d   \         P
                  # V\         P                  8X  d2   \        V4      '       g   \         P
                  # \         P                  # V\         P                  8X  d   \         P                  # \         P
                  # r.   )	r*   r(   r†   ÚhessianÚstaging_dtyperÖ   Úfloat16ra   rj   )rµ   r  r(   r  s   &&& r   Úpreferred_staging_dtypeÚGPTQ.preferred_staging_dtype�  s¥   € Ü—’˜fÓ%ˆàŸ	™	×)Ñ)×7Ñ7ˆØœEŸM™MÔ)Ü—=‘=Ð àœuŸ}™}¬e¯n©nÐ=Ô=Ü—=‘=Ð àœEŸN™NÔ*Ü,¨V×4Ò4Ü—}‘}Ð$Ü—>‘>Ð!àœEŸM™MÔ)Ü—=‘=Ð ä�}‰}Ðr!   c                óP   <€ V ^8„  d   QhRS[ RS[P                  RS[S[ ,          /# )r   rY   Ústage_dtyper   )r,   r*   rI   r   )r   r}   s   "€r   r    r~   ±  s-   ø€ ÷ ñ ©sð ÁÇÁð ÑQYÑZ]ÕQ^ñ r!   c                ó®  € V^ 8X  d   R# V P                   P                  P                  pVe   \        ^\	        W14      4      # V P                   P                  P
                  pVeo   V P                  \        P                  ! . VR7      P                  4       ,          pV^ 8”  d/   WE,          pV^ 8”  d    \        ^\	        \        V4      V4      4      # ^# R# )r   N©rI   )r†   r  Ú
chunk_sizerV   rø   Úchunk_bytesr—   r*   rô   Úelement_sizer,   )rµ   rY   r  Ú	cfg_chunkÚbytes_budgetÚbytes_per_rowÚ
chunk_rowss   &&&    r   Úresolve_hessian_chunk_sizeÚGPTQ.resolve_hessian_chunk_size±  s«   € Ø�1Œ9Ùà—I‘I×%Ñ%×0Ñ0ˆ	ØÒ Ü�qœ#˜iÓ.Ó/Ð/à—y‘y×(Ñ(×4Ñ4ˆØÒ#Ø ŸL™L¬5¯<ª<¸À+Ô+N×+[Ñ+[Ó+]Õ]ˆMØ˜qÔ Ø)Õ:�
Ø ”>Ü˜q¤#¤c¨*£o°tÓ"<Ó=Ð=Ùár!   c                óT   <€ V ^8„  d   QhRS[ P                  RS[RS[ P                  /# )r   ÚchunkrY   r   rö   )r   r}   s   "€r   r    r~   Å  s6   ø€ ÷ >Lñ >Lá�|‰|ð>Lñ ð>Lñ 
�‰ñ	>Lr!   c           	   #  óô  "  € V^ 8X  d2   VP                  ^ V P                  3\        P                  R7      x € R# VP                  pV P                  VP                  V4      pV P                  pVR;;,          ^,          uu&   \        W4V P                  V4      ;_uu_ 4       w  ppVR;;,          ^,          uu&   V'       d   VR;;,          ^,          uu&   MVR;;,          ^,          uu&   VRV1R3,          pVP                  VP                  VR7      4       V\        P                  8X  dŽ   VR;;,          ^,          uu&   V'       d   VR	;;,          ^,          uu&   MVR
;;,          ^,          uu&    Vx € VP                  R8X  d.   \        P                  P                  V4      P                  4        EM \        V\        P                  V P                  V4      ;_uu_ 4       w  p	p
VR;;,          ^,          uu&   V
'       d   VR	;;,          ^,          uu&   MVR
;;,          ^,          uu&    V	RV1R3,          pVP                  VP                  \        P                  4      4       Vx € VP                  R8X  d.   \        P                  P                  V4      P                  4        RRR4       RRR4       R#   TP                  R8X  d/   \        P                  P                  T4      P                  4        i i ; i  TP                  R8X  d/   \        P                  P                  T4      P                  4        i i ; i  + '       g   i     L¢; i  + '       g   i     R# ; i5i)r   r  NrŽ   r�   r�   r‘   ºNNNr’   r“   r”   Úcuda)Ú	new_zerosr—   r*   rÖ   r(   r  rI   r°   rZ   Úcopy_ré   r0   r%  Úcurrent_streamÚsynchronize)rµ   r"  rY   r(   r  ÚstatsÚstaging_workspaceÚstaging_reusedÚstaging_viewÚfp32_workspaceÚfp32_reusedÚ	fp32_views   &&&         r   Úborrow_materialized_chunk_fp32Ú#GPTQ.borrow_materialized_chunk_fp32Ä  s£  é € ð �1Œ9Ø—/‘/ 1 d§l¡lÐ"3¼5¿=¹=�/ÓIÒIÙà—‘ˆØ×2Ñ2°5·;±;ÀÓGˆà×,Ñ,ˆØˆj×˜QÕÓä˜f°4·<±<À×FÔFñ K
ØØàÐ$×%¨Õ*Ó%ßØ�n×%¨Õ*Ô%àÐ&×'¨1Õ,Ó'à,¨U¨d¨U°A¨XÕ6ˆLØ×Ñ˜uŸx™x¨k˜xÓ:Ô;àœeŸm™mÔ+ØÐ-×.°!Õ3Ó.ß!ØÐ-×.°!Õ3Ô.àÐ/×0°AÕ5Ó0ðHØ&Ò&à—{‘{ fÔ,ÜŸ
™
×1Ñ1°&Ó9×EÑEÔGùä%ØÜ—M‘MØ—L‘LØ÷	ô ñ
Ø"ØàÐ1×2°aÕ7Ó2ß"ØÐ1×2°aÕ7Ô2àÐ3×4¸Õ9Ó4ðLØ$2°5°D°5¸!°8Õ$<˜	Ø!Ÿ™¨¯©¼¿¹Ó(FÔGØ'šà!Ÿ;™;¨&Ô0Ü!ŸJ™J×5Ñ5°fÓ=×IÑIÔK÷+÷5 GÑFøð. —{‘{ fÔ,ÜŸ
™
×1Ñ1°&Ó9×EÑEÕGð -ûð. "Ÿ;™;¨&Ô0Ü!ŸJ™J×5Ñ5°fÓ=×IÑIÕKð 1ú÷)÷ ú÷5 G×FÐFüsq   ‚B!M8Â#B+M$Å+M$Å;K	Å?A.M$Ç-AM	È:?LÉ9>M	Ê7M$Ê?
M8Ë	AL
Ì
M$ÌAMÍM	ÍM!ÍM$Í$M5	Í/	M8c                óN   <€ V ^8„  d   QhRS[ P                  RS[ P                  /# )r   Úmatrixr   rÒ   )r   r}   s   "€r   r    r~     s#   ø€ ÷ ñ ©%¯,©,ð ¹5¿<¹<ñ r!   c           	     ó   € VP                   ^ ,          pV^ 8X  dH   \        P                  ! V P                  V P                  3\        P                  VP
                  R7      # V P                  VP                  VP
                  4      pV P                  W#4      pW0n	        Ve   TMTV n
        Vf[   VP                  \        P                  R7      p\        P                  ! VP                  V4      p?\        VP
                  R7       V# \        P                  ! V P                  V P                  3\        P                  VP
                  R7      p\        ^ W$4       Fs  p\!        WBV,
          4      p	WW‰,            p
V P#                  W©4      ;_uu_ 4       pTpVP%                  \        P                  ! VP                  V4      4       RRR4       Ku  	  \        VP
                  R7       V#   + '       g   i     K   ; i)r   rS   Nr  r;   )rO   r*   r`   r—   rÖ   r(   r  rI   r  r³   r´   ré   rb   ÚTr   Úrangerø   r1  r  )rµ   r4  rY   r  r  Úmat32r	  Ú	xtx_accumÚstartÚ	rows_thisÚsourceÚmaterializedÚmaterialized32s   &&           r   Úcompute_hessian_xtxÚGPTQ.compute_hessian_xtx  sz  € Ø�|‰|˜A�ˆØ�1Œ9Ü—;’; §¡¨d¯l©lÐ;Ä5Ç=Á=ÐY_×YfÑYfÔgÐgà×2Ñ2°6·<±<ÀÇÁÓOˆØ×4Ñ4°TÓGˆ
Ø-8Ô*Ø?IÒ?U±Ð[_ˆÔ.àÒØ—I‘I¤E§M¡M�IÓ2ˆEÜ—,’,˜uŸw™w¨Ó.ˆCØÜ˜cŸj™jÕ)ØˆJä—K’K §¡¨t¯|©|Ð <ÄEÇMÁMÐZ`×ZgÑZgÔhˆ	ä˜1˜dÖ/ˆEÜ˜J¨u­Ó5ˆIØ %Õ"3Ð4ˆFØ×4Ñ4°V×GÔGÈ<Ø!-�Ø—‘œuŸ|š|¨N×,<Ñ,<¸nÓMÔN÷ HÑGñ 0ô 	˜)×*Ñ*Õ+ØÐ÷ H×GÐGús   Æ3G,Ç,G=c                óŒ   <€ V ^8„  d   QhRS[ P                  RS[S[S[S[ P                  ,          S[ P
                  3,          /# )r   rÿ   r   )r*   rL   r   r,   r   r(   )r   r}   s   "€r   r    r~   "  s?   ø€ ÷ d7ñ d7¡§¡ð d7±%¹¹XÁeÇlÁlÕ=SÑUZ×UaÑUaÐ8aÕ2bñ d7r!   c           	     ó�  € \        V4      p\        V P                  \        P                  4      '       Ed+   VP                  R4      P                  \        P                  4      pVP                  4       p\        P                  ! V4      pV^ 8X  d   ^ RV3# \        P                  ! V4      w  rgVP                  4       ^ 8  g    VP                  4       V P                  8¼  d<   \        RV P                   RVP                  4        RVP                  4        R24      h\        P                  ! W0P                  R7      P                  \        P                   R7      pWHV3# \        V P                  \        P"                  \$        P&                  34      '       d&   VP                  RVP(                  R,          4      p	EMQ\        V P                  \        P*                  4      '       Ed	   VP                  VP-                  ^ 4      V P                  P.                  ,          VP-                  ^4      V P                  P.                  ,          VP(                  ^,          ^4      p	\        P0                  ! V P                  P2                  R,           V P                  P4                  R,           V P                  P6                  R,           V P                  P8                  R,           R7      p
V
! V	4      p	MüVP                  VP-                  ^ 4      V P                  P.                  ,          VP-                  ^4      V P                  P.                  ,          VP(                  ^,          VP(                  ^,          4      p	\        P0                  ! V P                  P2                  V P                  P4                  V P                  P6                  V P                  P8                  R7      p
V
! V	4      p	V	P;                  ^^4      P=                  ^ ^4      p	V	P?                  4       p	V P@                  '       dI   V	PC                  V	P(                  ^ ,          V P@                  34      p\        PD                  ! W›3^R	7      p	?\        P                  ! V4      pV	P(                  ^ ,          pV^ 8X  d   ?	^ RV3#  V PG                  V	4      P                  \        P                   R7      pVPI                  4       p?	T P_                  RR7       YÞT3#   \J         Ed2   p\        P                  ! T4      PL                  R
8X  Ed   R\O        T4      PQ                  4       9   dç   \R        PU                  R\W        T RR4      4       T	P                  \        P                  ! R4      R7      p?	\        PX                  P[                  4       '       d   \        PX                  P]                  4        \        P                  ! R4      pT PG                  T4      P                  \        P                   R7      pTPI                  4       p? Rp?ELL?	h Rp?ii ; i)r^   Nz/Embedding calibration token IDs must be in [0, z); observed range [z, z].)Ú	minlengthr  )ÚdilationÚpaddingÚstrideræ   r%  úout of memoryz^GPTQ module '%s' fell back to CPU Hessian accumulation due to GPU OOM during batch processing.r�   ú	<unknown>rŒ   r;   r  )Úcontextr7   )r^   ©r   )0r   rq   rr   rn   rs   Úreshaperé   r*   ÚlongÚnumelr(   ÚaminmaxÚitemr—   rÂ   ÚbincountrÖ   rÆ   ru   rv   rO   rÇ   ÚsizeÚgroupsÚUnfoldÚkernel_sizerD  rE  rF  Ú	transposerë   rú   r�   r&  rí   r?  ÚdetachÚRuntimeErrorr0   r+   Úlowerrß   rà   r›   r%  Úis_availableÚempty_cacheÚ _snapshot_borrow_workspace_stats)rµ   rÿ   Ú
inp_deviceÚ	token_idsr  r(   Úmin_idÚmax_idr  Úreshaped_inpÚunfoldrð   Úcanonical_devicer  r	  ÚexcÚreshaped_inp_cpus   &&               r   r  ÚGPTQ.process_batch"  sš  € ô   “_ˆ
ä�d—k‘k¤2§<¡<×0Ó0ØŸ™ B›×*Ñ*¬5¯:©:Ó6ˆIØ#Ÿ/™/Ó+ˆKÜ—\’\ *Ó-ˆFØ˜aÔØ˜$ �Ð&ä"Ÿ]š]¨9Ó5‰NˆFØ�{‰{‹}˜qÔ  F§K¡K£M°T·\±\Ô$AÜ ØEÀdÇlÁlÀ^ð T'Ø'-§{¡{£} o°R¸¿¹»°ÀbðJóð ô —^’^ I¿¹ÔF×IÑIÔPU×P]ÑP]ÐIÓ^ˆFØ¨Ð.Ð.ô
 �d—k‘k¤B§I¡I¬|×/BÑ/BÐ#C×DÒDØŸ;™; r¨3¯9©9°R­=Ó9ŠLä˜$Ÿ+™+¤r§y¡y×1Ó1Ø"Ÿ{™{Ø—H‘H˜Q“K $§+¡+×"4Ñ"4Õ4Ø—H‘H˜Q“K 4§;¡;×#5Ñ#5Õ5Ø—I‘I˜a•LØó	 �ô ŸšØ—K‘K×+Ñ+¨dÕ2Ø!Ÿ[™[×1Ñ1°DÕ8Ø ŸK™K×/Ñ/°$Õ6ØŸ;™;×-Ñ-°Õ4ô	�ñ  & lÓ3‘à"Ÿ{™{Ø—H‘H˜Q“K $§+¡+×"4Ñ"4Õ4Ø—H‘H˜Q“K 4§;¡;×#5Ñ#5Õ5Ø—I‘I˜a•LØ—I‘I˜a•Ló	 �ô ŸšØ—K‘K×+Ñ+Ø!Ÿ[™[×1Ñ1Ø ŸK™K×/Ñ/ØŸ;™;×-Ñ-ô	�ñ  & lÓ3�Ø'×1Ñ1°!°QÓ7×?Ñ?ÀÀ1ÓEˆLð $×.Ñ.Ó0ˆØ××ÐØ×(Ñ(¨,×*<Ñ*<¸QÕ*?À×ARÑARÐ)SÓTˆCÜ Ÿ9š9 lÐ%8¸aÔ@ˆLØÜ Ÿ<š<¨
Ó3Ðà'×-Ñ-¨aÕ0Ðà˜qÔ ØØ�dÐ,Ð,Ð,ð	Ø×*Ñ*¨<Ó8×;Ñ;Ä%Ç-Á-Ð;ÓPˆCð, —*‘*“,ˆCØà×-Ñ-°oÐ-ÔFØÐ&6Ð6Ð6øô3 õ 	ä—’˜ZÓ(×-Ñ-°Õ7Ø#¤s¨3£x§~¡~Ó'7Ô7ä—‘ØtÜ˜D &¨+Ó6ôð $0§?¡?¼%¿,º,ÀuÓ:M ?Ó#NÐ Ø Ü—:‘:×*Ñ*×,Ò,Ü—J‘J×*Ñ*Ô,Ü#(§<¢<°Ó#6Ð Ø×.Ñ.Ð/?Ó@×CÑCÌ%Ï-É-ÐCÓX�Ø—j‘j“l�Þ$à Øûð'	ús   Ò2/T ÔYÔD$Y Ø>Y Ù Yc                ó^   <€ V ^8„  d   QhRS[ S[P                  ,          RS[P                  /# )r   Ú	requestedr   ©r   r*   r(   )r   r}   s   "€r   r    r~   ˆ  s+   ø€ ÷ 'ñ '±xÁÇÁÕ7Mð 'ÑRW×R^ÑR^ñ 'r!   c                ó¶  € V P                   ;_uu_ 4        Ve!   \        P                  ! V4      uuR R R 4       # \        V RR 4      pVe!   \        P                  ! V4      uuR R R 4       # V P                  '       dM   \        \        V P                  P                  4       4      4      p\        P                  ! V4      uuR R R 4       # V P                  '       dM   \        \        V P                  P                  4       4      4      p\        P                  ! V4      uuR R R 4       # \        P                  ! R4      uuR R R 4       #   + '       g   i     R # ; i)NrŸ   rŒ   )	rE   r*   r(   r›   r«   ÚnextÚiterÚkeysr¬   )rµ   rg  ÚhintÚpartial_devices   &&  r   Ú_select_hessian_target_deviceÚ"GPTQ._select_hessian_target_deviceˆ  sæ   € Ø�Y�Y‹YØÒ$Ü—|’| IÓ.÷ ŠYô ˜4Ð!=¸tÓDˆDØÒÜ—|’| DÓ)÷ ŠYð ×,×,Ð,Ü!%¤d¨4×+HÑ+H×+MÑ+MÓ+OÓ&PÓ!Q�Ü—|’| NÓ3÷ ŠYð ×,×,Ð,Ü!%¤d¨4×+HÑ+H×+MÑ+MÓ+OÓ&PÓ!Q�Ü—|’| NÓ3÷ ŠYô  —<’< Ó&÷! �Y�Y‹Yús0   •Eº&EÁ+EÁ=AEÃ	EÃAEÄ'EÅE	c                óH   <€ V ^8„  d   QhRS[ S[P                  ,          RR/# )r   r‹   r   Nrh  )r   r}   s   "€r   r    r~   ›  s)   ø€ ÷ mñ m¹ÁÇÁÕ8Nð mÐZ^ñ mr!   c                óD  € V P                   ;_uu_ 4        V P                  V4      p\        V P                  \        P
                  4      '       Ed‡   V P                  '       gt   V P                  ef   V P                  P                  V8w  d"   V P                  P                  VR7      V n        R V n
        V P                  V n        W n         R R R 4       R # V P                  p\        V P                  P!                  4       4      pW4,           pV P                  e'   V P                  P"                  V P$                  38w  d5   \&        P(                  ! V P$                  \&        P*                  VR7      p^ pM+V P                  P                  V\&        P*                  R7      pV^ 8X  d   VP-                  4        MºV'       d,   VP/                  \1        V4      \1        V4      ,          4       MVP-                  4        \3        V P4                  \6        R7       FX  pV P4                  V,          pVP9                  VP                  V\&        P*                  R7      R\1        V4      ,          R7       KZ  	  W`n        R V n
        WPn        WPn        RV n        W n        V P4                  P;                  4        V P                  P;                  4         R R R 4       R # V P                  '       gV   V P                  eH   V P                  P                  V8w  d"   V P                  P                  VR7      V n
         R R R 4       R # \        V P                  P!                  4       4      pV P                  R J;'       dQ    V P                  P"                  V P$                  V P$                  38H  ;'       d    V P                  P                  V8H  p	V	'       dG   V P                  P<                  \&        P*                  8X  d   V P                  p
V
P-                  4        MH\?        V4       \&        P(                  ! V P$                  V P$                  3\&        P*                  VR7      p
V^ 8X  dZ   W n
        ^ V n        RV n        W n        V P@                  P;                  4        V P                  P;                  4         R R R 4       R # \3        V P@                  \6        R7       Fž  pV P@                  V,          pVP                  V
P                  8w  g    VP<                  \&        P*                  8w  d>    V
P9                  VP                  V
P                  \&        P*                  R7      4       K�  V
P9                  V4       K   	  V
P/                  R\1        V4      ,          4       W n
        WPn        RV n        V
P                  V n        V P@                  P;                  4        V P                  P;                  4        ?
R R R 4       R #   \B         Ed    \D        PG                  RT PH                   R	24       \J        PL                  ! R
4        T
P9                  TP                  T
P                  \&        P*                  R7      4        EKª    \B         dz    \D        PG                  RT PH                   R24       \J        PL                  ! R4       T
P9                  TP                  T
P                  \&        P*                  R7      4         EK-  i ; ii ; i  + '       g   i     R # ; i)Nr;   rS   ©r(   rI   )r>   ç       @)ÚalphaFrÀ   z"` -> Retry partial.to 1/2 in 0.25sg      Ð?z"` -> Retry partial.to 2/2 in 0.75sg      è?)'rE   ro  rq   rr   rn   rs   r¯   rª   r(   ré   r£   r®   r¤   rŸ   Úsumr­   ÚvaluesrO   r—   r*   r`   rÖ   Úzero_Úmul_rî   Úsortedr¬   r8   r  ÚclearrI   r   r«   rc   rß   rà   r�   ÚtimeÚsleep)rµ   r‹   r(   Úprevious_samplesÚnew_samplesÚtotal_samplesÚdiagrn  r  Úreuse_bufferÚresult_accumÚpartials   &&          r   Úmaterialize_global_hessianÚGPTQ.materialize_global_hessian›  s;  € Ø�Y�Y‹Yð ×7Ñ7¸ÓFˆFÜ˜$Ÿ+™+¤r§|¡|×4Ó4Ø×*×*Ð*¨t¯|©|Ò/GØ—|‘|×*Ñ*¨fÔ4Ø'+§|¡|§¡¸f Ó'E˜œØ!�D”FØ$(×$?Ñ$?�D”MØ6<Ô3Ø÷ ‰Yð $(×#>Ñ#>Ð Ü! $×"<Ñ"<×"CÑ"CÓ"EÓF�Ø 0Õ >�à—<‘<Ò'¨4¯<©<×+=Ñ+=À$Ç,Á,ÀÔ+PÜ Ÿ;š; t§|¡|¼5¿=¹=ÐQWÔX�DØ'(Ñ$àŸ<™<Ÿ?™?°&ÄÇÁ˜?ÓN�Dà  AÔ%Ø—J‘J•Lç'ØŸ	™	¤%Ð(8Ó"9¼EÀ-Ó<PÕ"PÕQàŸ
™
œÜ*0°×1NÑ1NÔTi×*j˜Ø!%×!>Ñ!>¸~Õ!N˜ØŸ	™	 &§)¡)°6ÄÇÁ )Ó"OÐWZÔ]bÐcpÓ]qÕWq˜	Örñ +kð  $”Ø�”Ø -”Ø.;Ô+Ø&+�Ô#Ø28Ô/Ø×-Ñ-×3Ñ3Ô5Ø×*Ñ*×0Ñ0Ô2Ø÷U ‰YðX ×&×&Ð&¨4¯6©6Ò+=Ø—6‘6—=‘= FÔ*Ø!ŸV™VŸY™Y¨f˜YÓ5�D”FØ÷_ ‰Yôb   × :Ñ :× AÑ AÓ CÓDˆMð —‘˜dÐ"÷ ,ð ,Ø—F‘F—L‘L T§\¡\°4·<±<Ð$@Ñ@÷,ð ,à—F‘F—M‘M VÑ+ð ÷  §¡§¡´·±Ô =Ø#Ÿv™v�Ø×"Ñ"Õ$ä˜6Ô"Ü$Ÿ{š{Ø—\‘\ 4§<¡<Ð0ÜŸ-™-Ø!ô �ð  Ô!Ø%”Ø !�”Ø&+�Ô#Ø28Ô/Ø×-Ñ-×3Ñ3Ô5Ø×*Ñ*×0Ñ0Ô2Ø÷[ ‰Yôb #)¨×)FÑ)FÔLa×"b�Ø×7Ñ7¸ÕG�Ø—>‘> \×%8Ñ%8Ô8¸G¿M¹MÌUÏ]É]Ô<Zð
kØ$×)Ñ)¨'¯*©*¸L×<OÑ<OÔW\×WdÑWd¨*Ó*eÖfð !×%Ñ% gÖ.ñ# #cð& ×Ñ˜c¤E¨-Ó$8Õ8Ô9à!ŒFØ)ŒMØ"'ˆDÔØ.:×.AÑ.AˆDÔ+Ø×)Ñ)×/Ñ/Ô1Ø×&Ñ&×,Ñ,Ô.Ø÷Y ‰Yøôp %õ kÜŸ™Ð#9¸$¿)¹)¸ÐDfÐ!gÔhÜŸ
š
 4Ô(ðkØ(×-Ñ-¨g¯j©jÀ×@SÑ@SÔ[`×[hÑ[h¨jÓ.i×jÐjøÜ(ô kÜŸH™HÐ'=¸d¿i¹i¸[ÐHjÐ%kÔlÜ ŸJšJ tÔ,Ø(×-Ñ-¨g¯j©jÀ×@SÑ@SÔ[`×[hÑ[h¨jÓ.i×jÑjðkúðkú÷q �YˆYús“   •A\Á$A)\ÃC)\ÇD\Ë\Ë*A\Ì?7\Í76\Î.!\ÏC$\Ò>A+\Ô*:W?Õ$B\×?A\	Ù:ZÙ?\ÚA>\Ü\	Ü\Ü\Ü\	Ü\Ü\	c                ó"   <€ V ^8„  d   QhRRRR/# )r   Úleaderr{   r   Nr$   )r   r}   s   "€r   r    r~   
  s   ø€ ÷ !:ñ !:¨ð !:°Dñ !:r!   c                óŒ  € WJ d   R# VP                   V P                   8w  dA   \        RVP                   RVP                    RV P                   RV P                    R2	4      hVP                  4        VP                  ;_uu_ 4        VP
                  pVP                  pVP                  pVf   VP                  R4      pV P                  \        V P                  RR4      4      pVP                  4       P                  V\        P                  RR7      pRRR4       V P                  ;_uu_ 4        XV n        XV n        XV n        V P                   P#                  4        V P$                  P#                  4        R	V n        VP(                  V n        RRR4       R#   + '       g   i     L�; i  + '       g   i     R# ; i)
a  Replace this task's Hessian statistics with a private copy of ``leader``'s.

Both modules must consume the identical input tensor (see ``:in=<tag>`` in
``module_tree``). The copy is independent because :meth:`quantize` mutates
``self.H`` in place and :meth:`free` drops it.
Nz!GPTQ: cannot share Hessian from `z` (z columns) with `z
 columns).r‹   T)r(   rI   rå   F)r—   rÂ   r�   r…  rE   r£   r¤   r§   Úcreate_Hro  r›   rr   rV  ré   r*   rÖ   r«   r{  r­   r¯   r(   rŸ   )rµ   rˆ  r<  r¤   r§   r‹   rå   s   &&     r   Úadopt_hessian_fromÚGPTQ.adopt_hessian_from
  s_  € ð ‹>ÙØ�>‰>˜TŸ\™\Ô)ÜØ3°F·K±K°=ÀÀFÇNÁNÐCSð TØŸ™˜ 3 t§|¡| n°Jð@óð ð
 	×)Ñ)Ô+Ø�[�[‹[Ø—X‘XˆFØ—‘ˆHØ ×,Ñ,ˆKØŠ~ØŸ™¨Ó.�à ×>Ñ>¼wÀtÇ{Á{ÐTcÐeiÓ?jÓkˆMØ—=‘=“?×%Ñ%¨]Ä%Ç-Á-ÐVZÐ%Ó[ˆD÷ ð �Y�Y‹YØˆDŒFØ$ˆDŒMØ*ˆDÔØ×)Ñ)×/Ñ/Ô1Ø×&Ñ&×,Ñ,Ô.Ø"'ˆDÔØ.2¯k©kˆDÔ+÷ ‰Y÷ �[ú÷ �YˆYús   ÂBFÄ3A"F2ÆF/	Æ2G	c                ó^   <€ V ^8„  d   QhRS[ S[P                  ,          RS[P                  /# )r   r‹   r   )r   r*   r(   rL   )r   r}   s   "€r   r    r~   -  s+   ø€ ÷ ñ ©h±u·|±|Õ.Dð ÑPU×P\ÑP\ñ r!   c                ó  € V P                  VR 7       \        V P                  \        P                  4      '       d   V P
                  f   Q hV P
                  # V P                  f   V P                  V4      V n        V P                  # )©r‹   )r…  rq   rr   rn   rs   rª   r£   rŠ  ©rµ   r‹   s   &&r   Úfinalize_hessianÚGPTQ.finalize_hessian-  sd   € Ø×'Ñ'°mÐ'ÔDÜ�d—k‘k¤2§<¡<×0Ò0Ø—<‘<Ò+Ð+Ð+Ø—<‘<ÐØ�6‰6Š>Ø—]‘] =Ó1ˆDŒFØ�v‰vˆr!   c                óš   € \         P                  ! V P                  V P                  3\         P                  V P	                  V4      R 7      # )rS   )r*   r`   r—   rÖ   ro  r�  s   &&r   rŠ  ÚGPTQ.create_H6  s8   € Ü�{Š{˜DŸL™L¨$¯,©,Ð7¼u¿}¹}Ø"&×"DÑ"DÀ]Ó"SôUð 	Ur!   c                óP   <€ V ^8„  d   QhRS[ RS[RS[S[P                  ,          /# )r   ÚstrategyÚ	blocksizer‹   )r	   r,   r   r*   r(   )r   r}   s   "€r   r    r~   :  s<   ø€ ÷ PNñ PNá"ðPNñ ðPNñ  ¡§¡Õ-ñ	PNr!   c           	     ób  € ^V P                   P                  ,          ^,
          pRpV P                   P                  R8w  d   V P                   P                  MV P                  p\        P                  ! 4       p\        V P                  RR4      p^ p	Rp
\        V\        4      '       d   VP                  p	VP                  p
VfE   V P                  e   V P                  P                  MV P                  P                  P                  pV P                  VR7      p\         P"                  ! V4      p. p. p\%        ^ V P                  V4       EFº  p\'        Wö,           V P                  4      pVRVV13,          p\        V\        4      '       d!   \)        VV P                   VV	V
R7      w  pppEM†\+        VV P                  VR7      w  ppV\,        P.                  8X  Ed    VP'                  ^R	R
7      P0                  pVP3                  ^R	R
7      P0                  pVV,           R,          p\         P4                  ! VV,
          V,          RR7      p\         P6                  ! VVR,          4      p\         P8                  ! VV,
          V,          V,           4      p\         P4                  ! V^ V4      p\         P8                  ! VVV,          ,
          4      p\         P4                  ! V^ V4      pVV,
          V,          pEM V\,        P:                  8X  Ed#   VP=                  ^R	R
7      p\         P2                  ! VV,
          P?                  4       ^R	R
7      P0                  p\         P4                  ! VRR7      p^V,          V,          p\         P6                  ! VVR,          4      p\         P8                  ! VV,
          V,          V,           4      p\         P4                  ! V^ V4      p\         P8                  ! VVV,          ,
          4      p\         P4                  ! V^ V4      pVV,
          V,          pEMèV\,        P@                  8X  Ed-   VPC                  ^R	R
7      P0                  p\         P2                  ! VV,
          P?                  4       ^R	R
7      P0                  p\         P4                  ! VRR7      p^V,          V,          p\         P6                  ! VVR,          4      p\         P8                  ! VV,
          V,          V,           4      p\         P4                  ! V^ V4      p\         P8                  ! VVV,          ,
          4      p\         P4                  ! V^ V4      pVV,
          V,          pEM¦V\,        PD                  8X  Ed	   VP=                  ^R	R
7      pVPG                  ^R	RR7      p\         P4                  ! VRR7      pVVV,          ,
          p VVV,          ,           p!\         P4                  ! V!V ,
          V,          RR7      p\         P8                  ! V ) V,          4      p\         P4                  ! V^ V4      p\         P8                  ! VV,          V,           4      p\         P4                  ! V^ V4      pVV,
          V,          pMˆV\,        PH                  8X  df   V PJ                  PM                  VR	R7       V PJ                  PO                  V4      pV PJ                  PP                  pV PJ                  PR                  pM\U        RV 24      hVe   VV,          pVV,          pVVRVV13&   VPW                  4       ^8”  d   TMVPY                  ^4      p"VPW                  4       ^8”  d   TMVPY                  ^4      p#V"PZ                  ^,          ^8”  d   V"P=                  ^R	R
7      p"V#PZ                  ^,          ^8”  d   V#P=                  ^R	R
7      p#VP]                  V"4       VP]                  V#4       EK½  	  \         P^                  ! V^R7      p\         P^                  ! V^R7      pV P`                  '       d?   V Pb                  p$VRRV$13,          pV Pe                  VV$4      pV Pe                  VV$4      pMV P                  p$VR8w  d   TMV P                  p%\         Pf                  ! V$VP                  \         Ph                  R7      V%,          p&\        V P                  \j        Pl                  4      '       d   VPo                  4       pVPZ                  V P                  P                  PZ                  8w  d^   VPq                  V P                  P                  PZ                  4      Ps                  V P                  P                  Pt                  4      pM/VPs                  V P                  P                  Pt                  4      pVPs                  V P                  P                  Pv                  P                  RR7      pWÀP                  P                  Pv                  ,
          P?                  4       P=                  4       Py                  4       p'\        P                  ! 4       V,
          p(RVPz                   RV'R 2p)Rp*RV n        VVVV&V(V)V*V P|                  3# )zÎApply a lightweight quantization fallback using the requested strategy.

``target_device`` is the device the weight clone is quantized on; when
None it is taken from ``self.H`` if present, else the weight.
g      @ÚsmoothNgš™™™™™é?r;   r$  )ÚstepsÚ	maxshrink)r¿   T)rç   Úkeepdimrt  g:Œ0âŽyE>©rø   F)rç   rœ  Úunbiased©rt   zUnsupported fallback strategy: ræ   rs  ©r(   Únon_blockingz	fallback(z): z.7fç        r7   )?r†   Úbitsr¿   r—   r|  r›   r¨   rq   r
   rš  r›  r£   r(   rr   rt   rñ   r*   Ú
empty_liker7  rø   r   r   r	   ÚMIDPOINTrw  rV   ÚclampÚ	full_likeÚroundÚMEANÚmeanÚabsÚMEDIANÚmedianÚSTDCLIPÚstdÚRTNr¦   Úfind_paramsÚquantizeÚscaleÚzerorÂ   rç   Ú	unsqueezerO   Úappendrí   r�   rš   rü   ÚarangeÚint32ru   rv   rê   rK  ré   rI   rè   rO  Úvaluer¤   )+rµ   r–  r—  r‹   ÚmaxqÚsigmaÚeffective_group_sizeÚ
start_timeÚsmooth_methodÚ	mse_stepsÚmse_maxshrinkÚWÚQÚscale_chunksÚzero_chunksr:  ÚendÚblockÚdequantr³  r´  Ú	block_modÚscale_factorÚw_minÚw_maxÚmidÚzero_midÚqrª  Úmax_devr­  r¯  ÚloÚhiÚscale_blockÚ
zero_blockÚ
valid_colsr¿   Úg_idxÚmean_abs_errÚdurationÚavg_lossr×   s+   &&&&                                       r   Ú_fallback_quantizeÚGPTQ._fallback_quantize:  s"  € ð �D—I‘I—N‘NÕ" QÕ&ˆØˆØ7;·y±y×7KÑ7KÈrÔ7Q˜tŸy™y×3Ò3ÐW[×WcÑWcÐÜ—Y’Y“[ˆ
Ü §¡¨x¸Ó>ˆØˆ	ØˆÜ�m¤Y×/Ò/Ø%×+Ñ+ˆIØ)×3Ñ3ˆMàÒ Ø-1¯V©VÒ-?˜DŸF™FŸMšMÀTÇ[Á[×EWÑEW×E^ÑE^ˆMØ×Ñ ]ÐÓ3ˆÜ×Ò˜QÓˆØˆØˆä˜1˜dŸl™lÐ,@×AˆEÜ�eÕ2°D·L±LÓAˆCØ�a˜˜s˜�l•OˆEä˜-¬×3Ò3Ü'8ØØ—I‘IØØ#Ø+ô(Ñ$�˜¢ô +7ØØ—M‘MØ3ô+Ñ'�	˜<ð
 Ô/×8Ñ8Õ8Ø%ŸM™M¨a¸˜MÓ>×EÑE�EØ%ŸM™M¨a¸˜MÓ>×EÑE�EØ  5�=¨CÕ/�CÜ!ŸKšK¨°­¸$Õ(>ÀDÔI�EÜ$Ÿš¨u°d¸SµjÓA�HÜŸš Y°¥_¸Õ$=ÀÕ$HÓI�AÜŸš A q¨$Ó/�AÜ Ÿ;š; x°3¸µ;Õ'?Ó@�DÜ Ÿ;š; t¨Q°Ó5�DØ  4�x¨5Õ0’GØÔ!1×!6Ñ!6Õ6Ø$Ÿ>™>¨a¸˜>Ó>�DÜ#Ÿiši¨°TÕ)9×(>Ñ(>Ó(@ÀaÐQUÔV×]Ñ]�GÜ#Ÿkšk¨'°tÔ<�GØ �[¨DÕ0�EÜ$Ÿš¨u°d¸SµjÓA�HÜŸš Y°Õ%5¸Õ$>ÀÕ$IÓJ�AÜŸš A q¨$Ó/�AÜ Ÿ;š; x°4¸%µ<Õ'@ÓA�DÜ Ÿ;š; t¨Q°Ó5�DØ  4�x¨5Õ0’GØÔ!1×!8Ñ!8Õ8Ø&×-Ñ-°!¸TÐ-ÓB×IÑI�FÜ#Ÿiši¨°VÕ);×(@Ñ(@Ó(BÈÐSWÔX×_Ñ_�GÜ#Ÿkšk¨'°tÔ<�GØ �[¨DÕ0�EÜ$Ÿš¨u°d¸SµjÓA�HÜŸš Y°Õ%7¸5Õ$@À8Õ$KÓL�AÜŸš A q¨$Ó/�AÜ Ÿ;š; x°6¸Eµ>Õ'BÓC�DÜ Ÿ;š; t¨Q°Ó5�DØ  4�x¨5Õ0’GØÔ!1×!9Ñ!9Õ9Ø$Ÿ>™>¨a¸˜>Ó>�DØ#Ÿ-™-¨A°tÀe˜-ÓL�CÜŸ+š+ c¨tÔ4�CØ ¨¥Õ+�BØ ¨¥Õ+�BÜ!ŸKšK¨¨b­°DÕ(8¸dÔC�EÜ Ÿ;š;¨ s¨U¥{Ó3�DÜ Ÿ;š; t¨Q°Ó5�DÜŸš I°Õ$5¸Õ$<Ó=�AÜŸš A q¨$Ó/�AØ  4�x¨5Õ0‘GØÔ!1×!5Ñ!5Ô5Ø—N‘N×.Ñ.¨yÀÐ.ÔFØ"Ÿn™n×5Ñ5°iÓ@�GØ ŸN™N×0Ñ0�EØŸ>™>×.Ñ.‘Dä$Ð'FÀxÀjÐ%QÓRÐRàÒ+Ø! LÕ0�EØ%¨Õ4�Gà%ˆAˆa��s�ˆl‰Oà#(§9¡9£;°¤?™%¸¿¹ÈÓ8JˆKØ!%§¡£¨a¤™°T·^±^ÀAÓ5FˆJØ× Ñ  Õ# aÔ'Ø)×.Ñ.°1¸dÐ.ÓC�Ø×Ñ Õ" QÔ&Ø'Ÿ_™_°¸D˜_ÓA�
Ø×Ñ Ô,Ø×Ñ˜z×*ñi Bôl —	’	˜,¨AÔ.ˆÜ�yŠy˜¨!Ô,ˆà××ÐØ×/Ñ/ˆJØ�!�[�j�[�.Õ!ˆAØ×*Ñ*¨5°*Ó=ˆEØ×)Ñ)¨$°
Ó;‰DàŸ™ˆJà-AÀRÔ-GÑ)ÈTÏ\É\ˆ
Ü—’˜Z°·±ÄÇÁÔLÐPZÕZˆä�d—k‘k¤<×#6Ñ#6×7Ò7Ø—‘“ˆAà�7‰7�d—k‘k×(Ñ(×.Ñ.Ô.Ø—	‘	˜$Ÿ+™+×,Ñ,×2Ñ2Ó3×6Ñ6°t·{±{×7IÑ7I×7OÑ7OÓP‰Aà—‘�T—[‘[×'Ñ'×-Ñ-Ó.ˆAà�D‰D˜Ÿ™×*Ñ*×/Ñ/×6Ñ6ÀUˆDÓKˆØŸK™K×.Ñ.×3Ñ3Õ3×8Ñ8Ó:×?Ñ?ÓA×FÑFÓHˆÜ—9’9“; Õ+ˆØ˜xŸ~™~Ð.¨c°,¸sÐ1CÐDˆØˆàˆŒØ�%˜˜u h°¸$ÀÇÁÐMÐMr!   c                ó(   € V P                  WW4WV4      # r.   )Úhf_quantize)rµ   r—  ÚpercdampÚdamp_auto_incrementr¿   ÚactorderÚstatic_groupss   &&&&&&&r   ÚfasterquantÚGPTQ.fasterquantÍ  s   € ð ×Ñ 	Ð5HÐV^ÓnÐnr!   c                ó0   <€ V ^8„  d   QhRS[ S[,          /# )r   r¾   )r   r   )r   r}   s   "€r   r    r~   Ù  s   ø€ ÷ Dñ Dñ &¡d�^ñDr!   c                ól  € W@P                   n        W P                   n        W0P                   n        WPP                   n        Ve   WpP                   n        V P                   P                  WW4       W`P                   n        V P                  VR7      w  r‰r«rÍrïW€P                  P                  n        WšW¼WÞ3# )N©r—  )r†   r¿   rÔ   rÞ  Údesc_actr¾   Ú_resolve_activation_orderingrà  r²  rr   rt   rè   )rµ   r—  rÝ  rÞ  r¿   rß  rà  r¾   rÂ  r³  r´  rÕ  r×  rØ  rÔ   r¤   s   &&&&&&&&        r   rÜ  ÚGPTQ.hf_quantizeÙ  s�   € ð  *�	‰	ÔØ!)�	‰	ÔØ(;�	‰	Ô%Ø%�	‰	ÔØÒ&Ø(7�I‰IÔ%Ø�	‰	×.Ñ.¨xÔIØ"/�	‰	ÔØNRÏmÉmÐfoÈmÓNpÑKˆ�4 °LØ"#�‰×ÑÔØ˜E¨XÐCÐCr!   c                ó4   <€ V ^8„  d   QhRS[ P                  /# rÑ   rÒ   )r   r}   s   "€r   r    r~   ð  s   ø€ ÷ dGñ dG¡§¡ñ dGr!   c                óŒ
  € \        V R R4      pVe   VP                  R4      M\        P                  ! 4       pT;_uu_ 4        \	        4       '       d9   \
        P                  RV P                   R\        VP                  4       24        VP                  4       pVP                  4       p\        P                  ! VP                  4       4      P                  4       p\         P"                  ! V4      '       d   VR8X  d   RpVR,          p^pV P$                  P&                  p	Rp
^ pW¸8:  EdX   V^ 8X  d   TpM™V\         P(                  ! RV^,
          4      ,          p\        P*                  ! W],           VR	7      pV^8X  d)   \
        P-                  R
V P                   RVR R24       M'\
        P-                  R
V P                   RVR R24       VP/                  V4       \        P0                  ! V4      pV P$                  P&                  pRpRpRp^ Tu;8  d	   ^8  Ed   M EM VP3                  Wþ,          4       VP4                  P6                  R8X  d   \9        V4      pMU\        P:                  P=                  V4      p\        P:                  P=                  \        P>                  ! V4      RR7      p?VP/                  V4       Tp	V'       d,   \
        P-                  R
V P                   RVR RVR R24       VV	3\	        4       '       d#   \
        P                  RV P                   24       uuRRR4       # V'       d0   Ve   TMTp\
        P-                  R
V P                   RVR R24       V^,          pEK^  \
        PG                  R
V P                   RXR R24       V
e   \
        PI                  R V
 24       R!\	        4       '       d#   \
        P                  RV P                   24       uuRRR4       #   \        P@                  PB                   dÒ   pTp
TP/                  T4       T P$                  PD                  ^ 8w  dq   T'       gD   RpTp\
        P-                  R
T P                   RTR RT P$                  PD                  R R24       YðP$                  PD                  ,          pTp Rp?EK¯  \
        P-                  R
T P                   RTR R24        Rp?EK¯  Rp?ii ; i  \	        4       '       d$   \
        P                  RT P                   24       i i ; i  + '       g   i     R# ; i)"r–   NÚhessian_inversezGPTQ: hessian_inverse begin z shape=r¢  ç      ð?ç�íµ ÷Æ°>g      $@r�  rÀ   z'` -> Applying Hessian diagonal floor (+z.2ez#) to recover positive definiteness.z+` -> Increasing Hessian diagonal floor to +Ú.FÚnpuT)Úupperz.` -> Damp recovery succeeded at `damp_percent=z.5fz` (started at ú).zGPTQ: hessian_inverse end z-` -> Starting damp recovery at `damp_percent=z`, increment step `rÁ   z0` -> Hessian Cholesky failed with `damp_percent=z#` and no auto increment configured.z7` -> Damp recovery failed after reaching `damp_percent=zg` -> Hessian remained non positive-definite after diagonal floor attempts. Last `damp_percent` tried = zHessian failure detail: )Nrì  )%r›   ÚmeasureÚ
contextlibÚnullcontextr%   rß   Úinfor�   ÚtuplerO   Údiagonalrï   r*   rV   r«  rO  rw   Úisfiniter†   rÔ   Úpowr¦  rà   r'  rª  r  r(   r0   r   ÚlinalgÚcholeskyÚcholesky_inverseÚ_CÚ_LinAlgErrorrÞ  ÚerrorÚdebug)rµ   r£   ÚtimerÚtimer_cmÚ	diag_viewÚ	orig_diagÚbase_abs_maxÚ
floor_baseÚmax_floor_attemptsÚ	used_dampÚ
last_errorÚattemptÚcurrent_diagÚfloor_incrementrª  r×   Údamp_recovery_startedÚrecovery_initial_dampÚrecovery_last_dampÚHinv_resultÚH2ÚeÚ
final_damps   &&                     r   rë  ÚGPTQ.hessian_inverseï  sÝ  € ä˜˜n¨dÓ3ˆð Ò ð �M‰MÐ+Ô,ä×'Ò'Ó)ð 	÷
 ŠXÜ#×%Ò%Ü—‘Ð7¸¿	¹	°{À'Ì%ÐPQ×PWÑPWË.ÐIYÐZÔ[ðZGàŸJ™J›L�	Ø%ŸO™OÓ-�	ô  %Ÿyšy¨¯©«Ó9×>Ñ>Ó@�Ü—}’} \×2Ò2°lÀcÔ6IØ#&�LØ)¨DÕ0�
Ø%&Ð"Ø ŸI™I×2Ñ2�	Ø!�
à�ØÕ3Ø !”|Ø'0™à*4´t·x²xÀÀgÐPQÅkÓ7RÕ*R˜Ü',§{¢{°9Õ3NÐTcÔ'd˜Ø" aœ<ÜŸH™HØ"8¸¿¹¸ÐCjÐkzÐ{~Ðjð  @cð  !dõeô  ŸH™HØ"8¸¿¹¸ÐCnÐo~ð  @Cð  oDð  DEð  !FôGð —O‘O LÔ1Ü Ÿ:š: lÓ3�DØŸ9™9×1Ñ1�Dà,1Ð)Ø,0Ð)Ø)-Ð&à˜dž, QŸ,™,ð &Ø%ŸN™N¨4­;Ô7Ø Ÿx™xŸ}™}°Ô5Ü.IÈ!Ó.L¡ä%*§\¡\×%:Ñ%:¸1Ó%= Ü.3¯l©l×.CÑ.CÄE×DZÒDZÐ[]ÓD^ÐfjÐ.CÓ.k Ø$&Ø%ŸO™O¨LÔ9Ø(,˜Iß4Ü #§¡Ø&<¸T¿Y¹Y¸KÐGuÐvzÐ{~Ðuð @3Ø3HÈÐ2MÈRð%Qô!"ð $/°	Ð#9ôB (×)Ò)Ü—H‘HÐ9¸$¿)¹)¸ÐEÔF÷{ ŠX÷\ -Ø;MÒ;YÑ%7Ð_c˜
ÜŸ™Ø4°T·Y±Y°KÐ?vð  xBð  CFð  wGð  GIð  Jôð ˜q•L“Gä—	‘	Ø,¨T¯Y©Y¨Kð  8_ð  `dð  ehð  _ið  ijð  kôlàÒ)Ü—I‘IÐ 8¸¸ÐEÔFØ ä'×)Ò)Ü—H‘HÐ9¸$¿)¹)¸ÐEÔF÷{ ŠXøôx  %Ÿx™x×4Ñ4ô &Ø)*˜JØ%ŸO™O¨LÔ9Ø#Ÿy™y×<Ñ<ÀÔAß'<Ø<@Ð$9Ø<@Ð$9Ü$'§H¡HØ*@ÀÇÁÀð L9Ø9=¸c¸
ÐBUÐVZ×V_ÑV_×VsÑVsÐtwÐUxÐxzð)|ô%&ð !%¯	©	×(EÑ(EÕ E Ø59× 2ä #§¡Ø&<¸T¿Y¹Y¸KÐGwÐx|ð  ~Að  xBð  Beð  %fô!gæ %ûð!&ûô@ (×)Ò)Ü—H‘HÐ9¸$¿)¹)¸ÐEÕFð *ú÷y �XˆXús~   ÁA	T2ÂA-S:Ã:D/S:È*B-PË.PÌ2T2ÍS:ÍA?S:Ï
2T2ÐS7Ð&5S2ÑA!S2Ò=S:Ó'S2Ó+S:Ó2S7Ó7S:Ó:5T/Ô/T2Ô2U	c                ó\  € ?\         P                   ! 4       p\        V P                  RR4      pVf!   V P                  P                  P                  p\
        P                  ! V4      pV P                  VR7      pV P                  VR7      pTpRpRpV P                  P                  '       d   EMV P                  P                  '       d;   \
        P                  ! VRR7      p	\
        P                  ! V	4      pVRV	3,          pM¬V P                  P                  '       d‘   \        W@P                  P                  RR7      w  r«\        VV P                  P                  VR	7      p\!        W¨V P                  P                  4      p	\#        W�P$                  4      p	\'        V	4      pVRV	3,          p\
        P(                  ! V4      p. p. pV P                  P                  R8w  d   V P                  P                  MV P$                  p\+        ^ V P$                  V4       F¶  p\-        VV,           V P$                  4      pVRVV13,          pV P.                  P1                  VRR
7       V P.                  P3                  V4      VRVV13&   VP5                  V P.                  P6                  4       VP5                  V P.                  P8                  4       K¸  	  \
        P:                  ! V^R7      p\
        P:                  ! V^R7      p\
        P<                  ! V P$                  \
        P>                  VP                  R7      V,          pV P                  P                  '       d'   V P                  P                  '       g   VV,          pMTpVe   VRV3,          pVeß   V P$                  V,          pV'       dÄ   \'        V4      PA                  VP                  R7      pVRRV13,          PC                  ^V4      pVRRV13,          PC                  ^VPA                  VP                  R7      4      p\
        P:                  ! VVRVR13,          3^R7      p\
        P:                  ! VVRVR13,          3^R7      p\E        VPE                  4       PG                  4       R4      pVR,          p\
        PH                  ! VVR7      V P                  PJ                  VPM                  4       ,          ,           pW\,
          PO                  4       PQ                  ^ R7      pV PR                  '       dE   R\
        PP                  ! VV,          4      ,          V PR                  ,          PG                  4       pMRpV PT                  '       d'   VRRV PV                  13,          pVRV PV                   pVPY                  4       P[                  V P                  P                  P\                  4      PA                  V P                  P                  P^                  4      pVPA                  V P                  P                  P                  RR7      pRV n0        V Pb                  Pe                  4        VVVV\         P                   ! 4       V,
          VV P                  PJ                  V PR                  3# )zAQuantize an embedding using its token-frequency diagonal Hessian.r‹   Nr�  r;   T©Ú
descendingr$  ©Úreturn_values©Úprecomputed_valuesrŸ  ræ   rS   rì  rí  r�  g      à?éÿÉš;Fr   r7   )3r|  r›   rr   rt   r(   r*   r‘  rñ   r†   rà  ræ  Úargsortr¾   r   r¿   r   r   r   r—   r   r¤  r7  rø   r¦   r±  r²  r¶  r³  r´  rí   r·  r¸  ré   Úindex_selectrV   rO  r¦  rÔ   rª  Úsquarerv  r¤   r�   rš   rê   rK  rO   rI   r£   r«   r{  ) rµ   r—  r:  r‹   r�  Úoriginal_weightrt   Úinverse_permutationÚgroup_permutationÚpermutationÚlocal_permsÚlocal_valuesÚ	quantizedÚscalesr`   r¿   Ú	start_colÚend_colrÆ  r³  r´  Úbase_group_indexrÕ  Úfull_group_countÚinverse_groupsÚ
full_scaleÚ	full_zeroÚabs_maxÚfloorÚeffective_diagÚcolumn_errorrØ  s    &&                              r   Ú_quantize_embeddingÚGPTQ._quantize_embeddingV  sD  € ð Ü—	’	“ˆÜ §¡¨_¸dÓCˆØÒ Ø ŸK™K×.Ñ.×5Ñ5ˆMÜŸš ]Ó3ˆà×$Ñ$°=Ð$ÓAˆØ×+Ñ+°=Ð+ÓAˆØ ˆØ"ÐØ Ðà�9‰9×"×"Ð"ÙØ�Y‰Y××ÐÜŸ-š-¨¸Ô>ˆKÜ"'§-¢-°Ó"<ÐØ˜A˜{˜NÕ+‰FØ�Y‰Y×&×&Ð&Ü(;Ø—i‘i×*Ñ*¸$ô)Ñ%ˆKô !4ØØ—	‘	×$Ñ$Ø#/ô!Ðô
 -¨[ÈTÏYÉY×MaÑMaÓbˆKÜ/°¿\¹\ÓJˆKÜ"-¨kÓ":ÐØ˜A˜{˜NÕ+ˆFä×$Ò$ VÓ,ˆ	ØˆØˆØ-1¯Y©Y×-AÑ-AÀRÔ-G�T—Y‘Y×)Ò)ÈTÏ\É\ˆ
ä˜q $§,¡,°
Ö;ˆIÜ˜) jÕ0°$·,±,Ó?ˆGØ˜1˜i¨Ð/Ð/Õ0ˆEØ�N‰N×&Ñ& u°TÐ&Ô:Ø.2¯n©n×.EÑ.EÀeÓ.LˆI�a˜ 7Ð*Ð*Ñ+Ø�M‰M˜$Ÿ.™.×.Ñ.Ô/Ø�L‰L˜Ÿ™×,Ñ,Ö-ñ <ô —	’	˜& aÔ(ˆÜ�yŠy˜ AÔ&ˆÜ Ÿ<š<¨¯©¼E¿K¹KÐPY×P`ÑP`ÔaÐeoÕoÐØ�9‰9××Ð d§i¡i×&=×&=Ð&=Ø$Ð%8Õ9‰Eà$ˆEàÒ*Ø! !Ð%8Ð"8Õ9ˆIàÒ(Ø#Ÿ|™|¨zÕ9ÐßÜ!,Ð->Ó!?×!BÑ!BÈ%Ï,É,Ð!BÓ!W�Ø" 1Ð&7Ð'7Ð&7Ð#7Õ8×EÑEÀaÈÓX�
Ø  Ð$5Ð%5Ð$5Ð!5Õ6×CÑCÀAÀ~×GXÑGXÐ`d×`kÑ`kÐGXÓGlÓm�	ÜŸ	š	 :¨u°QÐ8HÑ8IÐ5IÕ/JÐ"KÐQRÔS�Ü—y’y )¨T°!Ð5EÑ5FÐ2FÕ-GÐ!HÈaÔP�ä�d—h‘h“j—o‘oÓ'¨Ó-ˆØ˜$•ˆÜŸš T¨uÔ5¸¿	¹	×8NÑ8NÐQU×QZÑQZÓQ\Õ8\Õ\ˆØ'Õ3×;Ñ;Ó=×AÑAÀaÐAÓHˆØ�=�=ˆ=ØœeŸiši¨¸Õ(EÓFÕFÈÏÉÕV×\Ñ\Ó^‰Hà ˆHà××ÐØ! !Ð%< d×&<Ñ&<Ð%<Ð"<Õ=ˆIØÐ1˜4×1Ñ1Ð2ˆEà—K‘K“M×)Ñ)¨$¯+©+×*<Ñ*<×*BÑ*BÓC×FÑFÀtÇ{Á{×GYÑGY×G_ÑG_Ó`ˆ	Ø—L‘L¨¯©×(:Ñ(:×(AÑ(AÐPU�LÓVˆ	àˆŒØ×%Ñ%×+Ñ+Ô-àØØØÜ�IŠI‹K˜%ÕØØ�I‰I×"Ñ"Ø�M‰Mð	
ð 		
r!   c                óŽ2  € \        V P                  \        P                  4      '       d   V P	                  VR 7      # \
        P
                  ! 4       p\        V P                  RR4      p\        P                  ! V P                  P                  P                  P                  4      pRp^RIHpHpHp V! V P                  4      p	V! V P                  \!        V P"                  4      V P$                  4      p
V! V P                  V P$                  4      w  r¼VRJpV
'       Ed   Rp\'        \        V P                  RR4      4      pVe   V'       d   RV 2MRp\(        P+                  RV P,                   R	V	P.                   R
V P"                   RV V RV P$                   R24       V P0                  ;_uu_ 4        V P3                  V4      pV P4                  P7                  4        V P8                  P7                  4        RV n        RRR4       V P=                  W‘XR7      # RpV P?                  VR7       V P@                  PB                  '       d   V PD                  V n#        V PH                  f(   V PK                  V PL                  P                  R7      pM2V PH                  PO                  V PL                  P                  R7      pV =$V PP                  PS                  VRR7       V'       d<   \        PT                  ! V PL                  4      ^ 8H  p^V PL                  VV3&   ^ VRV3&   . p. p^pV P@                  PV                  '       dÔ   ^ RI,p. p\[        ^ V P\                  V P@                  P^                  4       Fž  pVPa                  V PP                  4      pVPS                  VRVVV P@                  P^                  ,           13,          RR7       VPc                  VPd                  4       VPc                  VPf                  4       VPc                  V4       K   	  V P@                  Ph                  '       dƒ   V'       d{   \        Pj                  ! \        PT                  ! V PL                  4      RR7      p VRV3,          pV PL                  V,          RV3,          V n&        \        Pj                  ! V4      pMðV P@                  Pt                  '       dÕ   V'       dÍ   \        PT                  ! V PL                  4      p\w        VV P@                  P^                  RR7      w  p p!\y        VV P@                  P^                  V!R7      p"?!\{        V V"V P@                  P^                  4      p#\}        V#V P\                  4      p# VRV#3,          pV PL                  V#,          RV#3,          V n&        V'       d!    V PG                  V PL                  4      w  p$p%MRRp%p$\        P~                  ! V4      p&\        P~                  ! V4      p'V P@                  PB                  '       Ed›   \[        ^ V P\                  V4       EF|  p(\�        V(V,           V P\                  4      p)V)V(,
          p*VRV(V)13,          p+\        P~                  ! V+4      p,V P@                  P^                  R-8w  Ed&   V P@                  PV                  '       gê   \ƒ        \[        V(V)V P@                  P^                  4      4      p-V- F¸  p.\�        V.V P@                  P^                  ,           V P\                  4      p/V.V/8  g   K=  V PP                  PS                  VRV.V/13,          RR7       VPc                  V PP                  Pd                  4       VPc                  V PP                  Pf                  4       V^,          pKº  	  Mi\[        V*4       FZ  pV(V,           p0V P@                  Ph                  '       d
   XV0,          p0XV0V P@                  P^                  ,          ,          V n(        K\  	  \…        V4      ^ 8”  EdI   \…        V4      ^ 8”  Ed8   VR-,          p1VR-,          p2V1P‡                  4       ^8X  d   V1P‰                  R-^4      p1V2P‡                  4       ^8X  d   V2P‰                  R-^4      p2^V P@                  PŠ                  ,          ^,
          p3V P@                  PŒ                  '       dL   V1\        PŽ                  ! \        P�                  ! V+V1,          4      V3^,          ) V3^,          4      ,          p,EM�\        PŽ                  ! \        P�                  ! V+V1,          4      V2,           ^ V34      p4V1V4V2,
          ,          p,EMC\[        V*4       FM  pV+RV3,          p5V PP                  P“                  V5P•                  ^4      4      P—                  4       p6V6V,RV3&   KO  	  EMå^V P@                  PŠ                  ,          ^,
          p3\™        V PP                  R4      '       EdI   \™        V PP                  R4      '       Ed,   V PP                  Pd                  p1V PP                  Pf                  p2V1P‡                  4       ^8X  d   V1P‰                  R-^4      p1V2P‡                  4       ^8X  d   V2P‰                  R-^4      p2V P@                  PŒ                  '       dK   V1\        PŽ                  ! \        P�                  ! V+V1,          4      V3^,          ) V3^,          4      ,          p,M§\        PŽ                  ! \        P�                  ! V+V1,          4      V2,           ^ V34      p4V1V4V2,
          ,          p,M\\[        V*4       FM  pV+RV3,          p5V PP                  P“                  V5P•                  ^4      4      P—                  4       p6V6V,RV3&   KO  	  V,V'RV(V)13&   EK  	  EMÿTp7V$fN   V P@                  P^                  '       d2   V P@                  P^                  ^ 8”  d   V P@                  P^                  p7\[        ^ V P\                  V74       EF�  p(\�        V(V7,           V P\                  4      p)V)V(,
          p*VRV(V)13,          P›                  4       p+\        P~                  ! V+4      p,V$e   \        P~                  ! V+4      MRp8V$e   \        P~                  ! V+4      MRp9V$e   V$V(V)1V(V)13,          p:\[        V*4       EF|  pV+RV3,          p5V$e   X:VV3,          p;V P@                  P^                  R-8w  Edr   V P@                  PV                  '       gÿ   V(V,           V P@                  P^                  ,          ^ 8X  dR   V PP                  PS                  VRV(V,           V(V,           V P@                  P^                  ,           13,          RR7       V(V,           V P@                  P^                  ,          V,
          R-8X  dT   VPc                  V PP                  Pd                  4       VPc                  V PP                  Pf                  4       V^,          pMWV(V,           p0V P@                  Ph                  '       d
   XV0,          p0XV0V P@                  P^                  ,          ,          V n(        V PP                  P“                  V5P•                  ^4      4      P—                  4       p6V6V,RV3&   V$f   EKñ  V5V6,
          ^,          X;^,          ,          V9RV3&   V5V6,
          V;,          p<V+RVR13;;,          V<P•                  ^4      P�                  X:VVR13,          P•                  ^ 4      4      ,          uu&   V<V8RV3&   EK  	  V,V'RV(V)13&   V$eF   V9^,          V&RV(V)13&   VRV)R13;;,          V8P�                  V$V(V)1V)R13,          4      ,          uu&   ?+?,?8?9V$f   EK�  ?:EK“  	  V$Eep   ?$V P"                  ^ 8w  dõ   \        Pž                  ! V&4      P¡                  4       V P"                  ,          p=\¢        P¤                  ! V=4      '       d£   \§        R\        Pž                  ! V&4      P¡                  4       4       V'       dT   \(        P©                  R V P,                   R!V P,                   R"24       RV P@                  n!        V P“                  VR 7      # \«        R V P,                   R#24      hM‚V'       d%   \(        P+                  RV P,                   R$24       M#\(        P+                  R%V P,                   R&24       V'       d   V	P.                   R'2MR(p=MV'       d   V	P.                   R'2MR(p=?&V =&?V P@                  P^                  R-8w  d   V P@                  P^                  MV P\                  p>V P@                  PV                  '       dP   V P@                  Ph                  '       d4   \[        V P\                  4       Uu. uF  pXV,          V>,          NK  	  p?pM+\[        V P\                  4       Uu. uF  pVV>,          NK  	  p?p\        P¬                  ! V?\        P®                  V'P                  R)7      p?V P@                  Ph                  '       d<   V'       d4   XPO                  V'P                  R7      pV'RV3,          p'V?V,          p???MÜV P@                  Pt                  '       dÁ   V'       d¹   \±        X#4      PO                  V'P                  R7      p@V'RV@3,          p'\±        X"4      pAVAP³                  4       pB\…        VB4      pCVB Uu. uF  pVV,          NK  	  pDpVDPµ                  VXCR 4       TDpXB Uu. uF  pVV,          NK  	  pEpVEPµ                  VXCR 4       TEp?#?@?"?A?B? V P¶                  '       d   V P¸                  pFV'RRVF13,          p'V?RVF p?\        V P                  \º        P¼                  4      '       d   V'P¿                  4       p'V'PÀ                  V P                  P                  PÀ                  8w  d^   V'PÃ                  V P                  P                  PÀ                  4      PO                  V P                  P                  PÄ                  4      p'M/V'PO                  V P                  P                  PÄ                  4      p'V. 8X  dK   VPc                  V PP                  Pd                  4       VPc                  V PP                  Pf                  4       \        PÆ                  ! V^R*7      p\        PÆ                  ! V^R*7      pV P¶                  '       d1   V P¸                  pFV PÉ                  VVF4      pV PÉ                  VVF4      pV'       d3   V'P                  V8w  d"   \(        P©                  R+V P,                  V4       V'PO                  VRR,7      p'\
        P
                  ! 4       V,
          pGV'VVV?VGV=V%V P"                  3#   + '       g   i     ELê; i  \l         Ed	   pT PL                  P                  Pn                  R8w  g   R\'        T4      Pq                  4       9  d   h T Ps                  RT PL                  P                  4       Rp\        P                  ! R4      pTPO                  TR7      pTPO                  TR7      RT3,          pT PL                  PO                  TR7      T,          RT3,          T n&        T PP                  PS                  TRR7        Rp?ELeRp?ii ; i  \l         Ed	   pT PL                  P                  Pn                  R8w  g   R\'        T4      Pq                  4       9  d   h T Ps                  RT PL                  P                  4       Rp\        P                  ! R4      pT#PO                  TR7      p#TPO                  TR7      RT#3,          pT PL                  PO                  TR7      T#,          RT#3,          T n&        T PP                  PS                  TRR7        Rp?ELuRp?ii ; i  \l         dü   pT PL                  P                  Pn                  R8w  g   R\'        T4      Pq                  4       9  d   h T Ps                  RT PL                  P                  4       Rp\        P                  ! R4      pT PL                  PO                  TR7      T n&        TPO                  TR7      pT PP                  PS                  TRR7       T PG                  T PL                  4      w  p$p% Rp?ELRRp?ii ; iu upi u upi u upi u upi ).rå  r‹   NF)Úresolve_fallback_strategyÚresolve_thresholdÚshould_use_fallbackÚ	thresholdz, threshold_raw=Ú rÀ   z` -> Using `z"` fallback quantization (observed z samples, threshold=z, max_total=rñ  r�  Tr;   rŸ  r$  r  r%  rG  zHessian permutationrŒ   r  r  zact-group Hessian permutationzHessian inverser¢  r³  r´  zLosses sum item:z,Quantization: Failed due to `NaN` loss for `z$`, use mock quantization retry for `rÅ   zI`, please try increasing calibration data samples or enable fallback=TruezJ` -> using fail safe mode. Please check if calibration data is sufficient.zQuantization: `z5` is not activated due to model inference logic (MoE)z	 fallbackr  rS   ræ   z^Quantization: Module `%s` -> CPU fallback complete; moving final quantized weights back to %s.r   r7   )erq   rr   rn   rs   r3  r|  r›   r*   r(   rt   rè   Úutils.fallbackr6  r7  r8  r¨   rî   r¤   r©   r+   rß   rà   r�   r¹  rE   ro  r«   r{  r­   r¯   rÙ  r‘  r†   Úmock_quantizationrÙ   rë  r¢   rñ   r£   ré   r¦   r±  r�  rà  rå   r7  r—   r¿   Údeepcopyr¶  r³  r´  ræ  r  rW  r0   rX  rá   r¾   r   r   r   r   Ú
zeros_likerø   ÚlistÚlenrç   Úviewr£  Úsymr¦  r¨  r²  rµ  rë   rÎ   rï   rb   rv  rO  rw   ÚisnanÚprintrõ  rÂ   rô   r¸  r   ÚtolistÚextendr�   rš   ru   rv   rê   rO   rK  rI   rí   rü   )Hrµ   r—  r:  r‹   Úresult_deviceÚcpu_fallback_usedr6  r7  r8  Úresolved_strategyÚfallback_requestedÚthreshold_rawÚ
is_percentÚfallback_configuredÚuse_hessianÚthreshold_textÚthreshold_infoÚfallback_devicerÁ  Údeadr³  r´  Únow_idxrå   rR  Úir¦   Úpermrc  Ú
cpu_deviceÚinvpermÚdiag_hr$  r%  Úglobal_permÚ
final_permÚHinvr×   ÚLossesrÂ  Úi1Úi2ÚcountÚW1ÚQ1Úgroup_start_colsÚgroup_startÚ	group_endÚidxÚlatest_scaleÚlatest_zeroÚmaxq_valr&  ÚwrÎ  Úeffective_blockÚErr1ÚLosses1ÚHinv1ÚdÚerr1rØ  r¿   rÕ  Ú	inv_finalÚinv_global_permÚinv_global_perm_listÚreordered_group_countÚ
temp_scaleÚ	temp_zerorÔ  r×  sH   &&                                                                      r   r²  ÚGPTQ.quantize¹  s7  € ô
 �d—k‘k¤2§<¡<×0Ò0Ø×+Ñ+°iÐ+Ó@Ð@ô —	’	“ˆä §¡¨_¸dÓCˆÜŸš T§[¡[×%7Ñ%7×%<Ñ%<×%CÑ%CÓDˆØ!ÐßfÑfá5°d·m±mÓDÐÙ0Ø�M‰MÜ�$—-‘-Ó Ø×"Ñ"ó
Ðñ
 %6°d·m±mÀT×E[ÑE[Ó$\Ñ!ˆØ+°4Ð7ÐçÐØˆKÜ ¤¨¯©¸ÀTÓ!JÓKˆNØCPÒC\×akÐ/°¨Ñ?ÐqsˆNÜ�H‰HØ(¨¯©¨ð 4Ø+×1Ñ1Ð2Ð2TÐUY×UbÑUbÐTcÐcwð  yGð  xHð  IWð  HXð  Xdð  ei÷  e{ñ  e{ð  d|ð  |~ðôð ——“Ø"&×"DÑ"DÀ]Ó"S�Ø×-Ñ-×3Ñ3Ô5Ø×*Ñ*×0Ñ0Ô2Ø&+�Ô#÷	 ð ×*Ñ*Ø!¸Oð +ó ð ð ˆKØ×!Ñ!°Ð!Ô>à�9‰9×&×&Ð&à#'×#<Ñ#<ˆDÔ à×ÑÒ#à×!Ñ!¨¯©¯©Ð!Ó7‰Aà× Ñ ×#Ñ#¨4¯6©6¯=©=Ð#Ó9ˆAØÐ à�‰×"Ñ" 1¨TÐ"Ô2÷ Ü—:’:˜dŸf™fÓ%¨Ñ*ˆDØ!"ˆD�F‰F�4˜�:ÑØˆAˆa�ˆg‰Jð ˆØˆØˆà�9‰9×"×"Ð"ÛàˆFÜ˜1˜dŸl™l¨D¯I©I×,@Ñ,@ÖA�Ø ŸM™M¨$¯.©.Ó9�	Ø×%Ñ% a¨¨1¨q°4·9±9×3GÑ3GÕ/GÐ+HÐ(HÕ&IÐRVÐ%ÔWà—‘˜YŸ_™_Ô-Ø—‘˜IŸN™NÔ+Ø—‘˜iÖ(ñ Bð �9‰9××Ð§+Ü—=’=¤§¢¨D¯F©FÓ!3ÀÔEˆDð;Ø�a˜�g•J�ØŸ™ � a¨ gÕ.�”ô —m’m DÓ)‰Gà�Y‰Y×&×&Ð&¯;Ü—Z’Z §¡Ó'ˆFÜ(;Ø˜Ÿ	™	×,Ñ,¸Dô)Ñ%ˆK˜ô .ØØ—	‘	×$Ñ$Ø#/ôˆKð
 Ü+¨K¸ÀdÇiÁi×FZÑFZÓ[ˆJÜ.¨z¸4¿<¹<ÓHˆJð;Ø�a˜�mÕ$�ØŸ™ 
Õ+¨A¨z¨MÕ:�”÷ ð:Ø!×1Ñ1°$·&±&Ó9‘
�‘dð ˜s�$ˆDä×!Ò! !Ó$ˆÜ×Ò˜QÓˆð �9‰9×&×&Ñ&Ü˜A˜tŸ|™|¨Y×7�Ü˜˜i�¨¯©Ó6�Ø˜R��à�q˜"˜R˜%�x•[�Ü×%Ò% bÓ)�ð —9‘9×'Ñ'¨2Õ-ØŸ9™9×2×2Ð2ä+/´°b¸"¸d¿i¹i×>RÑ>RÓ0SÓ+TÐ(Û+;˜KÜ(+¨K¸$¿)¹)×:NÑ:NÕ,NÐPT×P\ÑP\Ó(]˜IØ*¨YÖ6Ø $§¡× :Ñ :¸1¸QÀÈIÐ@UÐ=UÕ;VÐ_cÐ :Ô dØ %§¡¨T¯^©^×-AÑ-AÔ BØ $§¡¨D¯N©N×,?Ñ,?Ô @Ø '¨1¥¢ò ,<ô "' u¦˜AØ"$ q¥&˜CØ#Ÿy™y×1×1Ð1Ø&*¨3¥i Ø-3°C¸4¿9¹9×;OÑ;OÕ4OÕ-P˜DžNñ	 ".ô ˜5“z A•~¬#¨d«)°a­-à',¨R¥y˜Ø&*¨2¥h˜ð (×+Ñ+Ó-°Ô2Ø+7×+<Ñ+<¸RÀÓ+C˜LØ&Ÿ?™?Ó,°Ô1Ø*5×*:Ñ*:¸2¸qÓ*A˜Kð $%¨¯	©	¯©Õ#6¸Õ#:˜ØŸ9™9Ÿ=Ÿ=˜=à!-´·²Ü %§¢¨B°Õ,=Ó >Ø"*¨a¥-Ð 0Ø (¨A¥ó1õ "šBô ).¯ªÜ %§¢¨B°Õ,=Ó >ÀÕ LØ !Ø (ó)˜Ið
 ".°¸[Õ1HÕ!IšBô "' u¦˜AØ " 1 a 4¥˜AØ $§¡× 7Ñ 7¸¿¹ÀA»Ó G× OÑ OÓ Q˜AØ'(˜B˜q !˜t›Hó ".ð  ! D§I¡I§N¡NÕ2°QÕ6�HÜ˜tŸ~™~¨w×7Ó7¼GÀDÇNÁNÐTZ×<[Ó<[Ø'+§~¡~×';Ñ';˜Ø&*§n¡n×&9Ñ&9˜à'×+Ñ+Ó-°Ô2Ø+7×+<Ñ+<¸RÀÓ+C˜LØ&Ÿ?™?Ó,°Ô1Ø*5×*:Ñ*:¸2¸qÓ*A˜KàŸ9™9Ÿ=Ÿ=˜=Ø!-´·²Ü %§¢¨B°Õ,=Ó >Ø"*¨a¥-Ð 0Ø (¨A¥ó1õ "™Bô ).¯ªÜ %§¢¨B°Õ,=Ó >ÀÕ LØ !Ø (ó)˜Ið
 ".°¸[Õ1HÕ!I™Bô "' u¦˜AØ " 1 a 4¥˜AØ $§¡× 7Ñ 7¸¿¹ÀA»Ó G× OÑ OÓ Q˜AØ'(˜B˜q !˜t›Hñ ".ð
 !��!�R˜�U�(”ó 8ðD (ˆOØŠ| §	¡	× 4× 4Ð 4¸¿¹×9MÑ9MÐPQÔ9Qð #'§)¡)×"6Ñ"6�ä˜A˜tŸ|™|¨_×=�Ü˜˜oÕ-¨t¯|©|Ó<�Ø˜R��à�q˜"˜R˜%�x•[×&Ñ&Ó(�Ü×%Ò% bÓ)�Ø/3Ò/?”u×'Ò'¨Ô+ÀT�Ø26Ò2Bœ%×*Ò*¨2Ô.È�àÒ#Ø   B ¨¨2¨ Õ.�Eä˜uŸ�AØ˜1˜a˜4��AØÒ'Ø! ! Q $�K˜à—y‘y×+Ñ+¨rÕ1Ø#Ÿy™y×6×6Ð6Ø " Q¥¨$¯)©)×*>Ñ*>Õ>À!ÔCØ $§¡× :Ñ :¸1¸QÀÀaÅÈBÐQRÍFÐUY×U^ÑU^×UiÑUiÕLiÐ@jÐ=jÕ;kÐtxÐ :Ô yà!# a¥¨D¯I©I×,@Ñ,@Õ @ÀGÕKÈrÔQØ %§¡¨T¯^©^×-AÑ-AÔ BØ $§¡¨D¯N©N×,?Ñ,?Ô @Ø '¨1¥ øà"$ q¥&˜CØ#Ÿy™y×1×1Ð1Ø&*¨3¥i à-3°C¸4¿9¹9×;OÑ;OÕ4OÕ-P˜DœNàŸ™×/Ñ/°·±¸A³Ó?×GÑGÓI�AØ �B�q˜!�t‘HØÕ'Ø)*¨Q­°1­°q¸!µtÕ(;˜  1 ™Ø ! A¥¨�{˜Ø˜1˜a™b˜5Ÿ	 T§^¡^°AÓ%6×%=Ñ%=¸eÀAÀqÁrÀE½l×>TÑ>TÐUVÓ>WÓ%XÕX›	Ø%)˜˜Q ˜Tœ
ñ7 &ð: !��!�R˜�U�(‘ØÒ#Ø'.°¥{�F˜1˜b ˜e˜8Ñ$Ø�a˜™�f—I §¡¨T°"°R°%¸¹°*Õ-=Ó!>Õ>“Ià˜˜D 'ØÕ#Ûña >ðj ÓØØ�}‰} Ô!Ü Ÿ9š9 VÓ,×1Ñ1Ó3°d·m±mÕC�ä—:’:˜h×'Ò'ÜÐ,¬e¯iªi¸Ó.?×.DÑ.DÓ.FÔGß*ÜŸ™Ð#OÐPT×PYÑPYÈ{ÐZ~ð  @D÷  @Iñ  @Ið  Jð  JKð  "Lô  MØ6:˜Ÿ	™	Ô3Ø#Ÿ}™}°y˜}ÓAÐAä(Ð+WÐX\×XaÑXaÐWbð  clð  *mó  nð  nð (÷ 'Ü—H‘HÐ5°d·i±i°[ð  AKð  Lõ  Mä—H‘H˜¨t¯y©y¨kÐ9nÐoÔpßDWÐ/×5Ñ5Ð6°iÑ@Ð]f‘ç@SÐ+×1Ñ1Ð2°)Ñ<ÐYbˆHàØˆFØà-1¯Y©Y×-AÑ-AÀRÔ-G�T—Y‘Y×)Ò)ÈTÏ\É\ˆ
à�9‰9×"×"Ð" t§y¡y×'9×'9Ð'9Ü49¸$¿,¹,Ô4GÓHÑ4G¨q�T˜!•W 
×*Ð*Ñ4GˆEÐHˆEä.3°D·L±LÔ.AÓBÑ.A¨�Q˜*—_�_Ñ.AˆEÐBä—’˜U¬%¯+©+¸a¿h¹hÔGˆà�9‰9××Ð§+Ø—j‘j¨¯©�jÓ1ˆGØ�!�W�*•ˆAØ˜'•NˆEØ‘gà�Y‰Y×&×&Ð&¯;Ü# JÓ/×2Ñ2¸!¿(¹(Ð2ÓCˆIØ�!�Y�,•ˆAÜ)¨+Ó6ˆOØ#2×#9Ñ#9Ó#;Ð Ü$'Ð(<Ó$=Ð!Ù,@ÓAÑ,@ q˜% Ÿ(˜(Ñ,@ˆJÐAØ×Ñ˜eÐ$9Ð$:Ð;Ô<ØˆEÙ*>Ó?Ñ*> Q˜˜aŸ˜Ñ*>ˆIÐ?Ø×Ñ˜TÐ"7Ð"8Ð9Ô:ØˆDØ˜I {°OÐEYÐ[fà××ÐØ×/Ñ/ˆJØ�!�[�j�[�.Õ!ˆAØ˜+˜:Ð&ˆEä�d—k‘k¤<×#6Ñ#6×7Ò7Ø—‘“ˆAà�7‰7�d—k‘k×(Ñ(×.Ñ.Ô.Ø—	‘	˜$Ÿ+™+×,Ñ,×2Ñ2Ó3×6Ñ6°t·{±{×7IÑ7I×7OÑ7OÓP‰Aà—‘�T—[‘[×'Ñ'×-Ñ-Ó.ˆAà�BŒ;Ø�L‰L˜Ÿ™×-Ñ-Ô.Ø�K‰K˜Ÿ™×+Ñ+Ô,ä—	’	˜% QÔ'ˆÜ�yŠy˜ 1Ô%ˆà××ÐØ×/Ñ/ˆJØ×*Ñ*¨5°*Ó=ˆEØ×)Ñ)¨$°
Ó;ˆDç §¡¨]Ô!:Ü�H‰HØpØ—	‘	Øôð �D‰D˜°EˆDÓ:ˆä—9’9“; Õ&ˆà�%˜˜u h°¸$ÀÇÁÐMÐM÷a —�ûôn  õ 
;Ø—6‘6—=‘=×%Ñ%¨Ô/°?Ì#ÈcË(Ï.É.ÓJZÔ3ZØà×%Ñ%Ð&;¸T¿V¹V¿]¹]ÔKØ$(Ð!Ü"Ÿ\š\¨%Ó0�
Ø—w‘w j�wÓ1�Ø—D‘D 
�DÓ+¨A¨t¨GÕ4�ØŸ™Ÿ™¨*˜Ó5°dÕ;¸A¸t¸GÕD�”Ø—‘×*Ñ*¨1°TÐ*×:Ò:ûð
;ûô:  õ 
;Ø—6‘6—=‘=×%Ñ%¨Ô/°?Ì#ÈcË(Ï.É.ÓJZÔ3ZØà×%Ñ%Ð&EÀtÇvÁvÇ}Á}ÔUØ$(Ð!Ü"Ÿ\š\¨%Ó0�
Ø'Ÿ]™]°*˜]Ó=�
Ø—D‘D 
�DÓ+¨A¨z¨MÕ:�ØŸ™Ÿ™¨*˜Ó5°jÕAÀ!ÀZÀ-ÕP�”Ø—‘×*Ñ*¨1°TÐ*×:Ò:ûð
;ûô  ô :Ø—6‘6—=‘=×%Ñ%¨Ô/°?Ì#ÈcË(Ï.É.ÓJZÔ3ZØð ×%Ñ%Ð&7¸¿¹¿¹ÔGØ$(Ð!Ü"Ÿ\š\¨%Ó0�
ØŸ™Ÿ™¨*˜Ó5�”Ø—D‘D 
�DÓ+�Ø—‘×*Ñ*¨1°TÐ*Ô:Ø!×1Ñ1°$·&±&Ó9‘
�–dûð:üò^ IùâBùò  Bùò @s�   Æ:AAW(Ñ:,AW< Ö,A\ Ö6A`* ÁGAd3ÁHAd8ÁL%Ad=ÁMAeÁW(AW9	ÁW<A\ÁXC=A\Á\A\Á\A`'Á\C=A`"Á`"A`'Á`*Ad0Á`5C0Ad+Ád+Ad0c                ó<   <€ V ^8„  d   QhRS[ RS[S[S[3,          /# )r   Úresetr   )r   r   r+   r,   )r   r}   s   "€r   r    r~   Q  s$   ø€ ÷ ñ ±Tð ÁdÉ3ÑPSÈ8Ånñ r!   c                ó„   € \        V P                  4      pV'       d#   V P                   F  p^ V P                  V&   K  	  V# rJ  )rœ   r°   )rµ   rx  r*  r>   s   &&  r   Úborrow_materialized_chunk_statsÚ$GPTQ.borrow_materialized_chunk_statsQ  s:   € Ü�T×1Ñ1Ó2ˆßØ×3Ô3�Ø45�×,Ñ,¨SÓ1ñ 4àˆr!   c                ó$   <€ V ^8„  d   QhRS[ RR/# )r   rI  r   N)r+   )r   r}   s   "€r   r    r~   X  s   ø€ ÷ !3ñ !3¹3ð !3À4ñ !3r!   c               óB  € V P                  R R7      p\        VP                  R^ 4      ;'       g    ^ 4      pV^ 8X  d   R# \        VP                  R^ 4      ;'       g    ^ 4      p\        VP                  R^ 4      ;'       g    ^ 4      p\        VP                  R^ 4      ;'       g    ^ 4      p\        VP                  R^ 4      ;'       g    ^ 4      pV P                  pV P                  p	V	e   \        V	4      MRp
V'       d	   WC,          MR	pR
VRVRVRVRVRVRVRV
RV/	pWÀn        V P                  pVR;;,          V,          uu&   VR;;,          V,          uu&   VR;;,          V,          uu&   VR;;,          V,          uu&   VR;;,          V,          uu&   R# )T)rx  rŽ   Nr“   r”   r�   r‘   ún/ar¢  rI  r  r  Úhit_rate)rz  r,   rC   r´   r³   r+   r²   r±   )rµ   rI  r*  Útotal_requestsr“   r”   r�   r‘   r  r  Ústage_dtype_strr  ÚsummaryÚtotalss   &$            r   r[  Ú%GPTQ._snapshot_borrow_workspace_statsX  sŠ  € Ø×4Ñ4¸4Ð4Ó@ˆÜ˜UŸY™Y z°1Ó5×:Ð:¸Ó;ˆØ˜QÔÙä §	¡	Ð*=¸qÓ A× FÐ FÀQÓGÐÜ! %§)¡)Ð,AÀ1Ó"E×"JÐ"JÈÓKÐÜ˜5Ÿ9™9 ^°QÓ7×<Ð<¸1Ó=ˆÜ˜UŸY™YÐ'7¸Ó;×@Ð@¸qÓAˆØ×;Ñ;ˆ
Ø×8Ñ8ˆØ.9Ò.Eœ#˜kÔ*È5ˆß9GÐ$Ö5ÈSˆð �wØ˜ØÐ!2Ø!Ð#6Ø˜LØ˜nØ˜*Ø˜_Ø˜ð

ˆð /6Ô+à×.Ñ.ˆØˆz×˜nÕ,ÓØÐ"×#Ð'8Õ8Ó#ØÐ$×%Ð)<Õ<Ó%Øˆ~× ,Õ.ÓØÐ×  NÕ2Õ r!   rx  c                ó*   <€ V ^8„  d   QhRS[ RS[RR/# )r   rI  rx  r   N)r+   r   )r   r}   s   "€r   r    r~   {  s"   ø€ ÷ ')ñ ')©cð ')¹$ð ')È$ñ ')r!   c               óŠ  € V P                   p\        VP                  R ^ 4      ;'       g    ^ 4      pV^ 8X  d   V'       d   V P                  4        R# \        VP                  R^ 4      ;'       g    ^ 4      p\        VP                  R^ 4      ;'       g    ^ 4      pV'       d	   WT,          MRpV P                  ;'       g    / p\        VP                  R ^ 4      ;'       g    ^ 4      p	\        VP                  R^ 4      ;'       g    ^ 4      p
\        VP                  R^ 4      ;'       g    ^ 4      p\        VP                  RR4      ;'       g    R4      pVP                  RR4      pVP                  RR4      p\        P                  R	V\        V R
R4      VVV	V
VVVVVV4       V'       d   V P                  4        R# R# )rŽ   Nr“   r”   r¢  r  r  r~  r  z­GPTQ workspace cache [%s]: module=%s rows=%s staging_dtype=%s requests=%d hits=%d misses=%d hit_rate=%.2f total_requests=%d total_hits=%d total_misses=%d total_hit_rate=%.2fr�   rH  )	r±   r,   rC   Úreset_workspace_statsr²   rî   rß   rõ  r›   )rµ   rI  rx  rƒ  r€  Ú
total_hitsÚtotal_missesÚtotal_hit_rateÚlastÚlast_requestsÚ	last_hitsÚlast_missesÚlast_hit_rateÚ
rows_labelr  s   &$$            r   Úlog_workspace_statsÚGPTQ.log_workspace_stats{  sŒ  € Ø×.Ñ.ˆÜ˜VŸZ™Z¨
°AÓ6×;Ð;¸!Ó<ˆØ˜QÔßØ×*Ñ*Ô,Ùä˜Ÿ™Ð$7¸Ó;×@Ð@¸qÓAˆ
Ü˜6Ÿ:™:Ð&;¸QÓ?×DÐDÀ1ÓEˆß8F˜Ö4ÈCˆà×2Ñ2×8Ð8°bˆÜ˜DŸH™H Z°Ó3×8Ð8°qÓ9ˆÜ˜Ÿ™Ð!4°aÓ8×=Ð=¸AÓ>ˆ	Ü˜$Ÿ(™(Ð#8¸!Ó<×AÐAÀÓBˆÜ˜dŸh™h z°3Ó7×>Ð>¸3Ó?ˆØ—X‘X˜l¨EÓ2ˆ
Ø—h‘h˜°Ó6ˆä�‰ð@ð Ü�D˜& +Ó.ØØØØØØØØØØô	
÷$ Ø×&Ñ&Ö(ñ r!   c                ó   <€ V ^8„  d   QhRR/# r¼   r$   )r   r}   s   "€r   r    r~   ¤  s   ø€ ÷ 6ñ 6 tñ 6r!   c                ó¸   € V P                    F  p^ V P                   V&   K  	  V P                   F  p^ V P                  V&   K  	  RV n        RV n        RV n        R# )r   N)r°   r±   r²   r³   r´   )rµ   r>   s   & r   r‡  ÚGPTQ.reset_workspace_stats¤  sZ   € Ø×/Ô/ˆCØ01ˆD×(Ñ(¨Ó-ñ 0à×0Ô0ˆCØ12ˆD×)Ñ)¨#Ó.ñ 1à.2ˆÔ+Ø-1ˆÔ*Ø15ˆÖ.r!   c                ó  € V P                   ;_uu_ 4        V P                  P                  4        V P                  P                  4        R V n        RRR4       \        V R4      '       d   V =\        V R4      '       d   V =V P                  P                  4        V =	\        V R4      '       d   V =
V P                  e'   V P                  P                  P                  RR4       \        V RR4      pVe   V =R# R#   + '       g   i     L³; i)FNr£   rª   r¢   rˆ   rr   )rE   r«   r{  r­   r¯   rÎ   r£   rª   r¬   r¦   r¢   r˜   r�   rU   r›   rr   )rµ   Útargets   & r   ÚfreeÚ	GPTQ.free­  sÜ   € ð �Y�Y‹YØ×)Ñ)×/Ñ/Ô1Ø×&Ñ&×,Ñ,Ô.Ø"'ˆDÔ÷ ô
 �4˜×ÒØ�Ü�4˜×#Ò#Ø�Ø×%Ñ%×+Ñ+Ô-ØˆNÜ�4˜×'Ò'ØÐ à×ÑÒ)Ø×Ñ×$Ñ$×(Ñ(¨¸Ô=ä˜˜x¨Ó.ˆØÒØ’ñ ÷% �Yús   •<C<Ã<D	) r£   rª   r´   r²   r³   r°   r±   r¬   r«   r­   rŸ   r¯   r®   r˜   rš   r™   r�   r‰   r—   r©   r¨   r§   rë  rE   rr   r¢   r�   r¤   r†   r¦   r–   rY   )NN)TNr.   )é€   ç{®Gáz„?çú~j¼t“X?r7   FF)rš  r›  rœ  r7   FFN)rš  )F)+Ú__name__Ú
__module__Ú__qualname__Ú__firstlineno__Ústaticmethodrƒ   r¹   r¡   r    r¥   rO   rÙ   rá   rñ   rü   r
  r  r  ró  Úcontextmanagerr1  r?  r  ro  r…  r‹  r‘  rŠ  rÙ  rá  rÜ  r*   Úinference_moderë  r3  r²  rz  r[  r‘  r‡  r˜  Ú__static_attributes__Ú__classdictcell__)r}   s   @r   r{   r{   °   sÎ  ø‡ € Ø÷ó ð÷eEò eE÷N	ð 	ð ñCó ðC÷4ð 4ò÷ð ÷	
ð 	
÷ò ð2 ÷Eó ðE÷&'ò &'÷Pð ÷(ð ð& ×Ñ÷>Ló ð>L÷@ð ÷:d7ð d7÷L'ð '÷&mò m÷^!:ð !:÷Fò òU÷PNò PNôf	o÷Dò Dð, ×ÒÓ÷dGó ðdGðL ×ÒÓó`
ó ð`
ðD ×ÒÓóUNó ðUN÷nò ÷!3ð !3ðF')À÷ ')ò ')÷R6ð 6÷ð r!   r{   c                óÆ  € V ^8„  d   Qh/ ^ \         9   dA   \        \        \        \        \
        ,          3,          \        P                  3,          ;R&   ^\         9   dA   \        \        \        \        \
        ,          3,          \        P                  3,          ;R&   ^\         9   d7   \        \        \        \        \
        ,          3,          \        3,          ;R&   # )r   rT   rB   r_   )Ú__conditional_annotations__r   r   r+   r   r,   r*   rL   r?   r@   r   )r   s   "r   r    r       s—   € × Ñ ÷d EÒ D”$”uœS¤(¬3¥-Ð/Õ0´%·,±,Ð>Õ?Ñ Dñe ÷f GÒ F”$”uœS¤(¬3¥-Ð/Õ0´)·.±.Ð@ÕAÑ Fñg ÷h @Ò ?”Tœ%¤¤X¬c¥]Ð 2Õ3´TÐ9Õ:Ñ ?òi r!   )?r§  ró  rw   r?   r|  Útypingr   r   r   r*   Útorch.nnrn   ru   Útorch.nn.modules.convr   Úlooper.named_moduler   Úquantizationr   Úquantization.configr	   r
   Úutils.devicer   Ú	utils.envr   Úutils.loggerr   Úutils.torchr   Úfallback_smoothr   r   Úgarr   r   r   r   r   Ú
npu_linalgr   r¦   r   r   rß   r%   r@   rD   rT   rB   r_   r4   r8   r<   rF   rP   r¢  rZ   rj   ry   r{   Ú__all__r    )r§  s   @r   Ú<module>r¶     sã   øð÷ Ð Û Û Û ß (Ñ (ã Ý Û Ý )å -Ý )ß =Ý %Ý  Ý 'Ý $ß <÷õ õ 4ß ,ñ ƒn€õ-ð #ŸšÓ)Ð ð CEÐ Ó DØDFÐ Ó FØ=?Ð Ó ?õõ
<õ%õ
õð& ×Ñô.ó ð.õ4õ.H÷ Sñ Sðp0 ˆ(‚r!   