+
    "Ý³jÿ  ã                   óB  € ^ RI Ht ^ RIt^ RIHt ^RIHt ^RIHt ^RI	H
t
 ^RIHt ^RIHt ^R	IHt  ^ RIHt ]P&                  ! ]P(                  4       ^ R
IHt ^ RIHtHtHtHt Rt]! 4       tR R lt]P@                  ! 4       R 4       t!R#   ] d    Rt L1i ; i)é    )ÚUnionN)ÚPreTrainedModel)Ú
BaseQModel)ÚTorchLinear)ÚFORMAT)Úresolve_quant_format)Úsetup_logger)Útorch_empty_cache)Úgenerate)Ú_get_classesÚget_model_pathÚload_configÚquantize_modelTFc                ój   € V ^8„  d   QhR\         R\        \        \        3,          R\        R\         /# )é   Úmodel_id_or_pathÚmodelÚgptq_configÚlm_head_name)Ústrr   r   r   Údict)Úformats   "ÚT/home/luca/screen-checking/.venv/lib/python3.14/site-packages/gptqmodel/utils/mlx.pyÚ__annotate__r       s?   € ÷ Iñ I´#ð I¼eÄOÔU_ÐD_Õ>`ð IÔosð Iô  DGñ Ió    c           	      ó~  € \         '       g   \        R 4      hVR,          R9  d   \        R4      h\        VP                  R4      VP                  RVP                  R4      4      4      \        P
                  \        P                  39  d   \        R4      hVP                  R4      e`   \        VR,          4       VR,          P                  4        F2  w  rEV/ 8w  g   K  VR,          R9  g   K  \        RVR,           R	24      h	  VR
,          R9   d   ^@VR
&   \        V 4      w  rd\        V4      p\        V\        4      '       d   VP                  p/ p^p\        P                  \!        VP#                  4       4      4      P%                  R4      P'                  4       p	V	 EFï  w  r«V	P)                  V
 4      P+                  4        \        V\,        4      '       d    \.        P0                  ! VP3                  4       P4                  P7                  4       P9                  R\:        P<                  4      P?                  4       4      Wz R2&   VPA                  4        V^
,          ^ 8X  d   \C        4        V^,          pM�\E        VR4      '       d~   VR,          '       d   W£8X  gi   \.        P0                  ! VPF                  P7                  4       P9                  R\:        P<                  4      P?                  4       4      Wz R2&   V^,          p\E        VR4      '       g   EK  VPH                  f   EK�  \.        P0                  ! VPH                  P7                  4       P9                  R\:        P<                  4      P?                  4       4      Wz R2&   EKò  	  V=\C        4        \K        VR7      w  rÍV! VPM                  V4      4      p\        PO                  R4       VPQ                  \!        VP                  4       4      4       \S        WåVR
,          VR,          R7      w  r\        PO                  R4       W3# )úXMLX not installed. Please install via `pip install gptqmodel[mlx] --no-build-isolation`.ÚbitszModel bits is not in [2,3,4,8]r   ÚmethodÚquant_methodz#Model format is not gptq or gptq_v2ÚdynamiczModel bits z  in dynamic, it not in [2,3,4,8]Ú
group_sizezFormat: Converting to mlx ->Úcpuz.weightÚweightÚtie_word_embeddingsÚbiasz.bias)ÚconfigzStarting MLX quantization...)r"   r   zMLX quantization completed)r   é   é   é   )éÿÿÿÿé   )*ÚMLX_AVAILABLEÚ
ValueErrorr   Úgetr   ÚGPTQÚGPTQ_V2ÚprintÚitemsr   r   Ú
isinstancer   r   ÚlogÚpbÚlistÚnamed_modulesÚtitleÚmanualÚsubtitleÚdrawr   ÚmxÚarrayÚdequantize_weightÚTÚdetachÚtoÚtorchÚfloat16ÚnumpyÚ_empty_gptq_only_weightsr
   Úhasattrr$   r&   r   Ú	from_dictÚinfoÚload_weightsr   )r   r   r   r   Ú_r'   Ú
model_pathÚweightsÚnr6   ÚnameÚmoduleÚmodel_classÚmodel_args_classÚ	mlx_modelÚ
mlx_configs   &&&&            r   Úconvert_gptq_to_mlx_weightsrU       sw  € ß‹=ÜÐsÓtÐtà�6Õ ,Ô.ÜÐ9Ó:Ð:ä˜KŸO™O¨HÓ5°{·±ÀxÐQ\×Q`ÑQ`ÐaoÓQpÓ7qÓrô  |B÷  |Gñ  |Gô  IO÷  IWñ  IWð  {Xô  XÜÐ>Ó?Ð?à‡��yÓ!Ò-Üˆk˜)Õ$Ô%Ø$ YÕ/×5Ñ5Ö7‰IˆAØ˜Ž|Ø˜&•>¨Ö5Ü$ {°6¸&µ>Ð2BÐBbÐ%cÓdÐdñ 8ð �<Õ  HÔ,Ø$&ˆ�LÑ!ä"Ð#3Ó4�M€JÜ˜Ó$€Fä�%œ×$Ò$Ø—‘ˆð €GØ	€AÜ	�‰”�U×(Ñ(Ó*Ó+Ó	,×	2Ñ	2Ð3QÓ	R×	YÑ	YÓ	[€BÜ‰ˆØ
�‰�t�fÓ×#Ñ#Ô%Ü�fœk×*Ò*Ü(*¯ªØ×(Ñ(Ó*×,Ñ,×3Ñ3Ó5×8Ñ8¸ÄÇÁÓN×TÑTÓVó)ˆG�f˜GÐ$Ñ%ð ×+Ñ+Ô-à�2�v˜Œ{ä!Ô#à��F‰Aä�V˜X×&Ò&°Ð7L×0MÔ0MÐRVÔRfÜ(*¯ªØ—‘×$Ñ$Ó&×)Ñ)¨%´·±Ó?×EÑEÓGó)ˆG�f˜GÐ$Ñ%ð ��FˆAä�6˜6×"Õ"Ø�{‰{Õ&Ü*,¯(ª(Ø—K‘K×&Ñ&Ó(×+Ñ+¨E´5·=±=ÓA×GÑGÓIó+�˜& ˜Ô'ñ1 ð8 	ˆÜÔô %1¸Ô$?Ñ!€KÙÐ,×6Ñ6°vÓ>Ó?€Iô ‡H�HÐ+Ô,Ø×Ñœ4 §¡£Ó0Ô1Ü(¨À{ÐS_ÕG`Ø*5°fÕ*=ô?Ñ€Gä‡H�HÐ)Ô*àÐÐr   c                 ó–  € \         '       g   \        R 4      hVP                  RR4      pVf   \        R4      hVP                  RR4      pVP                  RR4       / pVP                  RR4      VR&   R	V9   d   VP                  R	R4      VR	&   R
V9   d   VP                  R
R4      VR
&   RV9   d   VP                  RR4      VR&   RV9   d   VP                  RR4      VR&   VP                  RR4      VR&   RV9   d   VP                  RR4      VR&   VP                  R^@4      VR&   VP                  R^ 4      VR&   R	V9   d   VP                  RR4      VR&   VP                  RR4      e   VP                  R4      VR&   M(VP                  RR4      e   VP                  R4      VR&   RV9   d   VP                  RR4      VR&   RV9   d   VP                  RR4      VR&   RV9   d   VP                  RR4      VR&   RV9   d   VP                  RR4      VR&   RV9   d   VP                  RR4      VR&   \        RRV RVRVRV/VB # )r   ÚpromptNz#MLX requires prompts to be providedÚverboseFÚ	formatterÚ
max_tokensé   ÚsamplerÚlogits_processorsÚmax_kv_sizeÚprompt_cacheÚprefill_step_sizei   Úkv_bitsÚkv_group_sizeÚquantized_kv_startÚprompt_progress_callbackÚtempÚtemperatureÚrepetition_penaltyÚrepetition_context_sizeÚtop_pÚmin_pÚmin_tokens_to_keepr   Ú	tokenizer© )r-   r.   Úpopr   )r   rl   ÚkwargsrW   rX   Úsampling_paramss   &&,   r   Úmlx_generaterq   k   sg  € ç‹=ÜÐsÓtÐtà�Z‰Z˜ $Ó'€FØ‚~ÜÐ>Ó?Ð?à�j‰j˜ EÓ*€GØ
‡J�Jˆ{˜DÔ!à€OØ$*§J¡J¨|¸SÓ$A€O�LÑ!Ø�FÔØ%+§Z¡Z°	¸4Ó%@ˆ˜	Ñ"à˜fÔ$Ø/5¯z©zÐ:MÈtÓ/TˆÐ+Ñ,à˜ÔØ)/¯©°MÀ4Ó)Hˆ˜Ñ&à˜ÔØ*0¯*©*°^ÀTÓ*Jˆ˜Ñ'à+1¯:©:Ð6IÈ3Ó+O€OÐ'Ñ(à�FÔØ%+§Z¡Z°	¸4Ó%@ˆ˜	Ñ"à'-§z¡z°/À2Ó'F€O�OÑ$Ø,2¯J©JÐ7KÈQÓ,O€OÐ(Ñ)à�FÔØ6<·j±jÐA[Ð]aÓ6bˆÐ2Ñ3à‡z�z�&˜$ÓÒ+Ø"(§*¡*¨VÓ"4ˆ˜ÒØ	�‰�M 4Ó	(Ò	4Ø"(§*¡*¨]Ó";ˆ˜Ñà˜vÔ%Ø06·
±
Ð;OÐQUÓ0VˆÐ,Ñ-à  FÔ*Ø5;·Z±ZÐ@YÐ[_Ó5`ˆÐ1Ñ2à�&ÔØ#)§:¡:¨g°tÓ#<ˆ˜Ñ à�&ÔØ#)§:¡:¨g°tÓ#<ˆ˜Ñ à˜vÔ%Ø06·
±
Ð;OÐQUÓ0VˆÐ,Ñ-äÑh˜%Ðh¨9Ðh¸VÐhÈWÐhÐXgÑhÐhr   )"Útypingr   rC   Útransformersr   Úmodelsr   Únn_modules.qlinear.torchr   Úquantizationr   Úquantization.configr   Úloggerr	   r
   Úmlx.coreÚcorer=   Úset_default_devicer#   Úmlx_lmr   Úmlx_lm.utilsr   r   r   r   r-   ÚImportErrorr5   rU   Úinference_moderq   rm   r   r   Ú<module>r€      s�   ðõ ã Ý (å Ý 2Ý !Ý 6Ý  Ý $ð	Ýà×Ò˜"Ÿ&™&Ô!åßVÓVØ€Mñ ƒn€õIðV ×ÒÓñ8ió ò8iøðc ô Ø‚Mðús   ¶6B Â	BÂB