+
    "Ý³jS? ã                  ó¸  € ^ RI Ht ^ RIt^ RIt^ RIt^ RIt^ RIHtHt ^ RI	H
t
 ^ RIHtHtHtHt ^ RIt^ RIt^ RIt^RIHt ^RIHtHt ]! 4       '       d   ^ RIHt M^ RIHt ^ RIt^ R	IHtHt ^ R
IH t H!t! ^ RI"H#t# ^RI$H%t% ^RI&H't' ^RI(H)t) ^RI*H+t+ ^RI,H-t- ^RI.H/t/ ^RI0H1t1 ^RI2H3t3H4t4H5t5H6t6H7t7 ^RI8H9t9 ^RI:H;t;H<t<H=t=H>t> ^RI?H@t@ ^RIAHBtBHCtCHDtDHEtEHFtFHGtGHHtHHItIHJtJHKtKHLtLHMtM ^RINHOtOHPtPHQtQHRtRHStSHTtT ^RIUHVtV ^RIWHXtX ^RIYHZtZH[t[H\t\H]t]H^t^H_t_H`t`HataHbtbHctcHdtdHeteHftfHgtgHhthHiti ^RIjHktkHltlHmtm ]X! 4       tnRto];Pà                  ]3Pâ                  ]3Pä                  0];Pæ                  ]3Pâ                  ]3Pè                  ]3Pä                  ]3Pê                  0/tvR R ltwR  R! ltxR" R# ltyR$ R% ltzR& R' lt{R( R) lt|R* R+ lt}R, R- lt~R. R/ ltR0 R1 lt€R2 R3 lt�R4 R5 lt‚R6 R7 ltƒR8 R9 lt„R: R; lt…R< R= lt†R> t‡R? tˆR@ RA lt‰RB RC ltŠRD RE lt‹RF RG ltŒRH RI lt�RJ RK ltŽRL t�RM RN lt�RO RP lt‘RQ RR lt’RS RT lt“RU RV lt”RW RX lt•RY RZ lt–R[ R\ lt—R] t˜R# )^é    )ÚannotationsN)ÚPackageNotFoundErrorÚversion)Úchain)ÚDictÚListÚOptionalÚUnion)Úensure_modelscope_available)Ú
LazyTurtleÚprint_module_tree)Úsnapshot_download)ÚInvalidVersionÚVersion)Ú
AutoConfigÚPretrainedConfig)Úis_flash_attn_2_available)ÚAdapter)ÚExllamaV3Linear)ÚExllamaV3TorchLinear)ÚBaseQuantLinear)ÚExllamaV2Linear)ÚGGUFTorchLinear)ÚQuantizeConfig)ÚFORMATÚMETHODÚMIN_VERSION_WITH_V2ÚBaseQuantizeConfigÚresolve_quant_format)Úinternal_gguf)ÚBACKENDÚPROFILEÚnormalize_backendÚnormalize_profile)Úreplace_exllamav3_placeholders)ÚINTERNAL_HF_GGUF_FILE_KWARGÚget_hf_config_dtypeÚget_hf_gguf_load_kwargsÚ(has_native_transformers_causallm_supportÚload_hf_tokenizerÚnormalize_hf_config_compatÚ&normalize_model_id_or_path_for_hf_ggufÚnormalize_torch_dtype_kwargÚ prepare_remote_model_init_compatÚresolve_trust_remote_codeÚset_hf_config_dtypeÚsuspend_hf_weight_init)Úauto_select_deviceÚexpand_selector_device_familyÚget_kernel_for_backendÚnormalize_device_device_mapÚselect_quant_linearÚselector_device_family)Úsafe_kwargs_call)Úsetup_logger)Ú apply_no_placement_to_device_mapÚ
auto_dtypeÚconvert_gptq_v1_to_v2_formatÚfind_config_seq_lenÚfind_modulesÚget_checkpointsÚget_layers_with_prefixesÚget_module_by_name_prefixÚget_module_nameÚgptqmodel_post_initÚis_embeddings_module_quantizedÚ)load_checkpoint_in_model_then_tie_weightsÚ
make_quantÚno_placement_module_namesÚsimple_dispatch_modelÚvalidate_checkpoint_qweights)ÚDEVICEÚHAS_NPUÚnormalize_deviceÚattn_implementationc               ó$   € V ^8„  d   QhRRRRRR/# )é   Úbackendr!   Úformat_coder   ÚreturnÚNone© )Úformats   "ÚX/home/luca/screen-checking/.venv/lib/python3.14/site-packages/gptqmodel/models/loader.pyÚ__annotate__rW   l   s'   € ÷ ]ñ ]¨wð ]ÀVð ]ÐPTñ ]ó    c                ó²   € \         P                  V 4      pVe   W9   d   R # RP                  R \        VR R7       4       4      p\	        V  RV RV 24      h)Nú, c              3  ó@   "  € T F  pR VP                    2x € K  	  R# 5i)zFORMAT.N©Úname)Ú.0Úitems   & rV   Ú	<genexpr>Ú4_validate_external_backend_format.<locals>.<genexpr>p   s   é € ÐrÑ<q°D˜G D§I¡I ;Õ/Ó<qùs   ‚c                ó   € V P                   # ©Nr\   )r_   s   &rV   Ú<lambda>Ú3_validate_external_backend_format.<locals>.<lambda>p   s   € Ðgk×gpÒgprX   )Úkeyz backend only supports z: actual = )Ú_EXTERNAL_BACKEND_FORMATSÚgetÚjoinÚsortedÚ
ValueError)rP   rQ   Úsupported_formatsÚ	supporteds   &&  rV   Ú!_validate_external_backend_formatrn   l   s[   € Ü1×5Ñ5°gÓ>ÐØÒ  KÔ$DÙØ—	‘	Ñr¼FÐCTÑZpÕ<qÓrÓr€IÜ
˜�yÐ 7¸	°{À+ÈkÈ]Ð[Ó
\Ð\rX   c               ó   € V ^8„  d   QhRR/# )rO   rR   ÚintrT   )rU   s   "rV   rW   rW   t   s   € ÷ 	ñ 	 sñ 	rX   c                óž  € \        V 4      ;'       g    \        P                  pV\        P                  \        P                  39   d   \
        P                  P                  4       # V\        P                  8X  d   \
        P                  P                  4       # V\        P                  8X  d   \
        P                  P                  4       # ^# )zDReturn the visible device count for layer-wise checkpoint placement.)r7   rJ   ÚCPUÚCUDAÚROCMÚtorchÚcudaÚdevice_countÚXPUÚxpuÚNPUÚnpu)ÚdeviceÚfamilys   & rV   Ú_layerwise_device_countr~   t   s�   € ä# FÓ+×9Ð9¬v¯z©z€FØ”&—+‘+œvŸ{™{Ð+Ô+Ü�z‰z×&Ñ&Ó(Ð(Ø”—‘ÔÜ�y‰y×%Ñ%Ó'Ð'Ø”—‘ÔÜ�y‰y×%Ñ%Ó'Ð'ÙrX   c               ó    € V ^8„  d   QhRRRR/# )rO   Únum_gpusrp   rR   ú	list[str]rT   )rU   s   "rV   rW   rW   €   s   € ÷ ñ °ð ¸	ñ rX   c                óâ  € \        V 4      ;'       g    \        P                  pV\        P                  \        P                  39   dN   \
        P                  P                  4       '       d   \        V4       Uu. uF  pRV 2NK
  	  up# \        R4      hV\        P                  8X  dd   \        \
        R4      '       dC   \
        P                  P                  4       '       d   \        V4       Uu. uF  pRV 2NK
  	  up# \        R4      hV\        P                  8X  d6   \        '       d   \        V4       Uu. uF  pRV 2NK
  	  up# \        R4      hR.V,          # u upi u upi u upi )	zCRender layer-wise placement targets using the selected accelerator.zcuda:zCUDA is not availablery   zxpu:zXPU is not availableznpu:zNPU is not availableÚcpu)r7   rJ   rr   rs   rt   ru   rv   Úis_availableÚrangeÚRuntimeErrorrx   Úhasattrry   rz   rK   )r|   r€   r}   Úindexs   &&  rV   Ú_layerwise_device_stringsr‰   €   s  € ä# FÓ+×9Ð9¬v¯z©z€FØ”&—+‘+œvŸ{™{Ð+Ô+Ü�:‰:×"Ñ"×$Ò$Ü16°x´ÓA±¨�e˜E˜7“O±ÑAÐAÜÐ2Ó3Ð3Ø”—‘ÔÜ”5˜%× Ò ¤U§Y¡Y×%;Ñ%;×%=Ò%=Ü05°h´Ó@± u�d˜5˜'“N±Ñ@Ð@ÜÐ1Ó2Ð2Ø”—‘Ôß‹7Ü05°h´Ó@± u�d˜5˜'“N±Ñ@Ð@ÜÐ1Ó2Ð2Øˆ7�XÕÐùò Bùò Aùò As   Á:E"Ã2E'Ä<E,c               ó    € V ^8„  d   QhRRRR/# )rO   Úrequested_device_mapz0Optional[Union[str, Dict[str, Union[str, int]]]]rR   zDict[str, Union[str, int]]rT   )rU   s   "rV   rW   rW   ’   s   € ÷ ñ àJðð  ñrX   c                óº  € \        V 4      ;'       g    \        P                  pR VP                  /pV'       dD   V\        P                  \        P
                  \        P                  \        P                  09  d   V# \        V\        4      '       d   V0M\        VP                  4       4      p\        V4      ^8w  d   V# VP                  4       p\        V\        4      '       d   WSR&   V# \        V\        4      '       d   VR9   d   V#  \        P                   ! V4      pTP&                  e   TP&                  TR&   T#   \"        \$        3 d    Tu # i ; i)r|   Úbase_gpu_id>   rƒ   ÚdiskÚmeta)r7   rJ   rr   Útypers   rt   rx   rz   Ú
isinstanceÚstrÚsetÚvaluesÚlenÚpoprp   ru   r|   r†   rk   rˆ   )r|   r‹   Údevice_familyÚruntime_kwargsÚtargetsÚtargetÚexplicit_devices   &&     rV   Ú_external_runtime_device_kwargsrœ   ’   s0  € ô +¨6Ó2×@Ð@´f·j±j€MØ2:¸M×<NÑ<NÐ1O€Nß =¼¿¹ÄfÇkÁkÔSY×S]ÑS]Ô_e×_iÑ_iÐ8jÔ#jØÐô Ð*¬C×0Ò0ð 
ÑäÐ%×,Ñ,Ó.Ó/ð ô
 ˆ7ƒ|�qÔØÐØ�[‰[‹]€Fä�&œ#×ÒØ(.�}Ñ%ØÐÜ�fœc×"Ò" fÐ0GÔ&GØÐðÜŸ,š, vÓ.ˆð ×ÑÒ(Ø(7×(=Ñ(=ˆ�}Ñ%ØÐøô	 œ*Ð%ô ØÒðús   ÄE ÅEÅEc               ó   € V ^8„  d   QhRR/# )rO   rR   ÚboolrT   )rU   s   "rV   rW   rW   ³   s   € ÷ Gñ G 4ñ GrX   c                 óŽ   € \         P                  P                  R4      p V f   R# V P                  4       P	                  4       R9   # )z=Keep expensive module-tree dumps opt-in during model loading.ÚGPTQMODEL_PRINT_MODULE_TREEF>   Ú1ÚtÚyÚonÚyesÚtrue)ÚosÚenvironrh   ÚstripÚlower)Úraws    rV   Ú_should_print_module_treer¬   ³   s:   € ô �*‰*�.‰.Ð6Ó
7€CØ
‚{ÙØ�9‰9‹;×ÑÓÐ"FÑFÐFrX   c               ó   € V ^8„  d   QhRR/# ©rO   rR   rS   rT   )rU   s   "rV   rW   rW   ¼   s   € ÷ 'ñ ' tñ 'rX   c                óB   € \        4       '       d   \        V R7       R# R# )zCPrint the module tree only when explicitly requested for debugging.©ÚmodelN)r¬   r   r°   s   &rV   Ú_maybe_print_module_treer²   ¼   s   € ô !×"Ò"Ü ×&ñ #rX   c               ó    € V ^8„  d   QhRRRR/# )rO   Úcleanup_originalrž   rR   rT   )rU   s   "rV   rW   rW   Ã   s   € ÷ ñ ¸dð Àtñ rX   c                óP  € \         P                  ! WR 7      p\        V RR4      pV'       dT   V FM  p\        W4      w  rgVf   \        P                  RV4       K,  \         P                  ! WbR 7      ;'       g    TpKO  	  \        V RR4      p\        V4      '       d	   V! V4       V# )©r´   Údefuser_module_pathsNz/Loader: defuser module path `%s` was not found.Úafter_defuser_conversionrT   )ÚdefuserÚconvert_modelÚgetattrrA   ÚlogÚwarnÚcallable)	Úclsr±   r´   Ú	convertedr·   Úmodule_pathÚmoduleÚ_Úafter_conversions	   &&&      rV   Ú_convert_model_with_defuserrÅ   Ã   s™   € Ü×%Ò% eÔO€Iä" 3Ð(>ÀÓCÐßÛ/ˆKÜ1°%ÓE‰IˆFØŠ~Ü—‘ÐJÈKÔXÙÜ×-Ò-¨fÔX×eÐeÐ\eŠIñ 0ô ˜sÐ$>ÀÓEÐô Ð ×!Ò!Ù˜ÔàÐrX   c               ó    € V ^8„  d   QhRRRR/# )rO   Úconfigr   rR   rž   rT   )rU   s   "rV   rW   rW   Ù   s   € ÷ ñ Ð#3ð ¸ñ rX   c                ó  € \        V RR4      '       g   R# \        \        V P                  ^ ,          R4      pVf   R# \        VR4      '       d   \	        \        VR4      4      # \        VR4      '       d   \	        \        VR4      4      # R# )z@Detect whether the resolved HF architecture exposes FA2 kernels.ÚarchitecturesNFÚ_supports_flash_attn_2Ú_supports_flash_attn)r»   ÚtransformersrÉ   r‡   rž   )rÇ   Úmodel_classs   & rV   rÊ   rÊ   Ù   s}   € ô �6˜?¨D×1Ò1Ùäœ,¨×(<Ñ(<¸QÕ(?ÀÓF€KØÒÙäˆ{Ð4×5Ò5Ü”G˜KÐ)AÓBÓCÐCÜˆ{Ð2×3Ò3Ü”G˜KÐ)?Ó@ÓAÐAÙrX   c               ó   € V ^8„  d   QhRR/# )rO   rÇ   r   rT   )rU   s   "rV   rW   rW   ê   s   € ÷ *ñ *Ð,<ñ *rX   c              #  óÌ  "  € V .p\        4       pV'       EdJ   VP                  4       p\        V\        4      '       g   K1  \	        V4      pWB9   d   KD  VP                  V4       Vx € \        V4      P                  4        FÙ  p\        V\        4      '       d   VP                  V4       K,  \        V\        4      '       dC   VP                  4        F,  p\        V\        4      '       g   K  VP                  V4       K.  	  K„  \        V\        \        \         34      '       g   K§  V F,  p\        V\        4      '       g   K  VP                  V4       K.  	  KÛ  	  EKR  R# 5i)z8Yield config and all nested PretrainedConfig nodes once.N)r“   r–   r‘   r   ÚidÚaddÚvarsr”   ÚappendÚdictÚlistÚtuple)rÇ   ÚstackÚvisitedÚcurÚnode_idÚvalueÚsubs   &      rV   Ú_iter_nested_pretrained_configsrÝ   ê   s   é € ð ˆH€EÜ‹e€Gç
ˆ%Ø�i‰i‹kˆÜ˜#Ô/×0Ò0Ùä�S“'ˆØÔÙØ�‰�GÔØŠ	ä˜#“Y×%Ñ%Ö'ˆEÜ˜%Ô!1×2Ò2Ø—‘˜UÖ#Ü˜E¤4×(Ò(Ø Ÿ<™<ž>�CÜ! #Ô'7×8Ô8ØŸ™ SÖ)ó *ô ˜E¤D¬%´Ð#5×6Ô6Û �CÜ! #Ô'7×8Ô8ØŸ™ SÖ)ó !ô (ñ ùs   ‚E$™CE$Ã/5E$Ä)E$ÅE$c               ó$   € V ^8„  d   QhRRRRRR/# )rO   rÇ   r   rM   r’   rR   rS   rT   )rU   s   "rV   rW   rW     s$   € ÷ ñ Ð*:ð ÐQTð ÐY]ñ rX   c                ó\   € \        V 4       F
  p Wn        K  	  R#   \         d     K  i ; i)zCApply attention implementation override to root and nested configs.N)rÝ   Ú_attn_implementationÚ	Exception)rÇ   rM   Ú
sub_configs   && rV   Ú_override_attn_implementationrã     s2   € ô 6°fÖ=ˆ
ð	Ø.AÖ+ó >øô ô 	âð	ús   �œ+ª+c               ó   € V ^8„  d   QhRR/# r®   rT   )rU   s   "rV   rW   rW     s   € ÷ #2ñ #2°Dñ #2rX   c                óÆ  € \        V R4      '       d!   \        V P                  \        4      '       g   R# \	        V P                  RR4      ;'       g    \	        V P                  RR4      p\        V\
        4      '       d4   RVP                  R4       Uu. uF  q"P                  4       NK  	  up9  d   R#  ^ RIH	pH
p \	        T RR4      pTf#    TP                  T P                  4      pYPn        \	        TR	R4      p\        Yc4      '       g   T! RR
7      Tn        R# \	        TRR4      f
   RTn        R# R# u upi   \         d     R# i ; i  \         d     R# i ; i)a­  Disable CUDA graph capture for paged attention by default.

``transformers`` continuous batching defaults ``use_cuda_graph=True`` for flash-attention
paths, but ``flash_attn_with_kvcache`` is not currently CUDA-graph capture-safe. Setting
the model's default ``continuous_batching_config`` keeps paged Flash Attention working for
direct ``generate_batch`` / ``init_continuous_batching`` callers without requiring a monkey
patch.
rÇ   Nrà   rM   ÚpagedÚ|)ÚContinuousBatchingConfigÚGenerationConfigÚgeneration_configÚcontinuous_batching_config)Úuse_cuda_graphrì   )FF)r‡   r‘   rÇ   r   r»   r’   Úsplitr©   rÌ   rè   ré   rá   Úfrom_model_configrê   rë   rì   )r±   ÚattnÚpartrè   ré   Ú
gen_configÚ	cb_configs   &      rV   Ú%_set_paged_attention_safe_cuda_graphsró     s=  € ô �5˜(×#Ò#¬:°e·l±lÔDT×+UÒ+UÙä�5—<‘<Ð!7¸Ó>÷ ð Ä'Ø�‰Ð+¨TóC€Dô �dœC× Ò  GÈTÏZÉZÐX[Ì_Ó3]É_ÀT·J±J¶LÉ_Ñ3]Ô$]ÙðßKô ˜Ð 3°TÓ:€JØÒð	Ø)×;Ñ;¸E¿L¹LÓIˆJð #-Ôä˜
Ð$@À$ÓG€IÜ�i×:Ò:Ù0HÐXfÔ0gˆ
Ö-Ü	�Ð,¨dÓ	3Ò	;Ø#1ˆ	Ö ñ 
<ùò' 4^øô
 ô Úðûô ô 	Úð	ús*   ÂD:Â7D? ÃE Ä?EÅEÅE ÅE c               ó    € V ^8„  d   QhRRRR/# )rO   ÚrotationúOptional[str]rR   rS   rT   )rU   s   "rV   rW   rW   9  s   € ÷ %
ñ %
°ð %
À$ñ %
rX   c                ó°  € V'       g   R# ^RI Hp VR	9  d   \        RV R24      h^ pV P                  4        Fw  w  rE\	        V\
        4      '       g   K  VP                  R4      '       g   K6  VP                  pV! V4      w  rxRVn        W…n	        Ve   VP                  V4       V^,          pKy  	  V^ 8X  d   \        P                  RV R24       R# R# )
a:  Attach online Hadamard transform state to QuantLinear modules for rotation inference.

Rotation (QuaRot/SpinQuant/Hadamard) fuses an orthogonal ``Q`` and an exact
Hadamard ``H`` into the weights. The ``H`` must be re-applied to activations
at inference before the quantized matmul for ``mlp.down_proj``. The per-head
Hadamard fused into ``v_proj``/``o_proj`` cancels inside the attention path,
so ``self_attn.o_proj`` does not require an online transform. Only the
``rotation`` string is persisted; ``had_K``/``K`` are recomputed from the
model dimensions at load time.
N)Úget_hadKzUnsupported rotation mode: `Ú`zmlp.down_projTz
Rotation `zm` requested but no `mlp.down_proj` QuantLinear modules were found; online Hadamard transform was not applied.>   ÚrandomÚhadamard)Ú$quantization.rotation.hadamard_utilsrø   rk   Únamed_modulesr‘   r   ÚendswithÚin_featuresÚonline_full_hadÚKÚ	set_had_Kr¼   r½   )	r±   rõ   rø   Úonline_countr]   rÂ   Úintermediate_sizeÚhad_Kr  s	   &&       rV   Ú_setup_rotation_online_hadr  9  sÖ   € ÷ Ùå?àÐ-Ô-ÜÐ7¸°zÀÐCÓDÐDà€LØ×+Ñ+Ö-‰ˆÜ˜&¤/×2Ò2Ùà�=‰=˜×)Ô)Ø &× 2Ñ 2ÐÙÐ 1Ó2‰HˆEØ%)ˆFÔ"ØŒHØÒ Ø× Ñ  Ô'Ø˜AÕŠLñ .ð �qÔÜ�‰Ø˜˜
ð #Mð Mö	
ñ rX   c               ó    € V ^8„  d   QhRRRR/# )rO   r|   ÚobjectrR   rž   rT   )rU   s   "rV   rW   rW   a  s   € ÷ ñ ¨Vð ¸ñ rX   c                óŽ  € \        V \        \        34      '       d&   \        V 4      ;'       d    \	        V ^ ,          4      # \        V \
        P                  4      '       d   V P                  R9   # \        V \        4      '       d$   V \        P                  \        P                  09   # \        V \        4      '       d   V R9   # R# )zGReturn True when the selected device can run CUDA/ROCm flash attention.F>   Úhiprv   >   r
  rv   Úrocm)r‘   rÖ   rÕ   rž   Ú _is_accelerated_attention_deviceru   r|   r�   rJ   rs   rt   r’   )r|   s   &rV   r  r  a  s’   € ô �&œ5¤$˜-×(Ò(Ü�F‹|×KÐKÔ @ÀÈÅÓ KÐKä�&œ%Ÿ,™,×'Ò'Ø�{‰{˜oÑ-Ð-Ü�&œ&×!Ò!Øœ&Ÿ+™+¤v§{¡{Ð3Ñ3Ð3Ü�&œ#×ÒØÐ0Ñ0Ð0ÙrX   c               ó$   € V ^8„  d   QhRRRRRR/# )rO   Únative_gguf_qspecú5Optional['internal_gguf.GGUFQuantizedCheckpointSpec']Úprofiler"   rR   rT   )rU   s   "rV   rW   rW   p  s(   € ÷ ñ àLðð ðð ñ	rX   c                óÔ   € V ed   V P                   \        P                  P                  8X  d;   V\        P
                  8X  d&   \        P                  R4       \        P                  # V# )z8Resolve user profile intent for native GGUF checkpoints.z3Loader: Q1_0 PROFILE.AUTO resolved to PROFILE.FAST.)	Útensor_qtyper    ÚGGMLQuantizationTypeÚQ1_0r"   ÚAUTOr¼   ÚinfoÚFAST©r  r  s   $$rV   Ú_resolve_native_gguf_profiler  p  sL   € ð 	Ò%Ø×*Ñ*¬m×.PÑ.P×.UÑ.UÔUØ”w—|‘|Ô#ä�‰ÐFÔGÜ�|‰|ÐØ€NrX   c               ó$   € V ^8„  d   QhRRRRRR/# )rO   r  r  r  r"   rR   rž   rT   )rU   s   "rV   rW   rW   �  s(   € ÷ ñ àLðð ðð 
ñ	rX   c                óž   € V RJ;'       dC    V P                   \        P                  P                  8H  ;'       d    V\        P
                  8H  # )z8Fast Bonsai mode stays on the dense HF GGUF import path.N)r  r    r  r  r"   r  r  s   $$rV   Ú"_should_use_dense_native_gguf_pathr  �  sK   € ð 	 Ð%÷ 	$ð 	$Ø×*Ñ*¬m×.PÑ.P×.UÑ.UÑU÷	$ð 	$à”w—|‘|Ñ#ðrX   c               ó   € V ^8„  d   QhRR/# )rO   Úversion_strr’   rT   )rU   s   "rV   rW   rW   �  s   € ÷ Cñ C cñ CrX   c                óT   €  \        V 4      #   \         d    \        R T  24      hi ; i)zInvalid version format: )r   r   rk   )r  s   &rV   Úparse_version_stringr   �  s7   € ðCÜ�{Ó#Ð#øÜô CÜÐ3°K°=ÐAÓBÐBðCús   ‚
 �'c                óâ   € V P                  4       p R F?  pW9   g   K  V P                  V^4      w  r#VP                  4       WP                  4       3u # 	  V '       d   V RR3# \        RV  24      h)ú>=Nz#Unsupported version constraint in: )r"  ú<=Ú>Ú<ú==)r©   rí   rk   )ÚreqÚopÚpkgÚversion_requireds   &   rV   Úparse_requirementr+  –  sm   € Ø
�)‰)‹+€CÛ*ˆØŽ9Ø$'§I¡I¨b°!Ó$4Ñ!ˆCØ—9‘9“; ×$:Ñ$:Ó$<Ð<Ò<ñ +÷ Ø�D˜$ˆÐÜ
Ð:¸3¸%Ð@Ó
AÐArX   c                ó¸   € \        V 4      p\        V4      pVR 8X  d   W48„  # VR8X  d   W48¬  # VR8X  d   W48  # VR8X  d   W48*  # VR8X  d   W48H  # \        RV 24      h)r$  r"  r%  r#  r&  zUnsupported operator: )r   rk   )Úinstalled_versionÚrequired_versionÚoperatorÚ	installedÚrequireds   &&&  rV   Úcompare_versionsr2  ¡  s~   € Ü$Ð%6Ó7€IÜ#Ð$4Ó5€HØ�3„ØÑ#Ð#Ø	�TÔ	ØÑ$Ð$Ø	�SŒØÑ#Ð#Ø	�TÔ	ØÑ$Ð$Ø	�TÔ	ØÑ$Ð$äÐ1°(°Ð<Ó=Ð=rX   c               ó    € V ^8„  d   QhRRRR/# ©rO   Úexcrá   rR   rž   rT   )rU   s   "rV   rW   rW   ²  s   € ÷ Rñ R Ið R°$ñ RrX   c                ó<   € \        V 4      pR V9   ;'       d    RV9   # )z cannot be called on meta tensorsz.item())r’   )r5  Úmessages   & rV   Ú_is_meta_shell_build_errorr8  ²  s%   € ô �#‹h€GØ-°Ñ8×QÐQ¸YÈ'Ñ=QÐQrX   c               ó    € V ^8„  d   QhRRRR/# r4  rT   )rU   s   "rV   rW   rW   ¹  s   € ÷ Sñ S°ið SÀDñ SrX   c                ó®   € \        V \        4      '       g   R # \        \        V RR4      ;'       g    T 4      pRV9   ;'       d    VP	                  R4      # )FÚfilenameÚ Útransformers_modulesz.py)r‘   ÚFileNotFoundErrorr’   r»   rþ   )r5  Úmissing_paths   & rV   Ú,_is_broken_transformers_dynamic_module_errorr@  ¹  sL   € Ü�cÔ,×-Ò-ÙÜ”w˜s J°Ó3×:Ð:°sÓ;€LØ! \Ñ1×RÐR°l×6KÑ6KÈEÓ6RÐRrX   c               ó   € V ^8„  d   QhRR/# )rO   Úmodel_local_pathr’   rT   )rU   s   "rV   rW   rW   À  s   € ÷ Hñ HÐSVñ HrX   c                óÎ  €  V P                   ! V3/ VB #   \         EdD   p\        T4      '       g   h \        \	        TR R4      ;'       g    R4      p\
        P                  P                  T4      p\
        P                  P                  Y4      pT'       dˆ   \
        P                  P                  T4      '       dc   \
        P                  ! \
        P                  P                  T4      RR7       \        P                  ! Yd4       \        P                  RTT4       \!        T4      pRTR&   \        P                  RTT4       T P                   ! T3/ TB u Rp?# Rp?ii ; i)r;  r<  T©Úexist_okzELoader: repaired missing dynamic-module file by copying `%s` -> `%s`.Úforce_downloadzxLoader: detected broken transformers dynamic-module cache while loading `%s`; retrying once with force_download=True: %sN)Úfrom_pretrainedrá   r@  r’   r»   r§   ÚpathÚbasenameri   ÚisfileÚmakedirsÚdirnameÚshutilÚcopy2r¼   r½   rÔ   )ÚloaderrB  Úkwargsr5  r?  Úmissing_nameÚsource_pathÚretry_kwargss   &&,     rV   Ú4_hf_loader_from_pretrained_with_dynamic_module_retryrT  À  s  € ðHØ×%Ò%Ð&6ÑA¸&ÑAÐAøÜõ HÜ;¸C×@Ò@Øäœ7 3¨
°BÓ7×=Ð=¸2Ó>ˆÜ—w‘w×'Ñ'¨Ó5ˆÜ—g‘g—l‘lÐ#3ÓBˆßœBŸG™GŸN™N¨;×7Ò7Ü�KŠKœŸ™Ÿ™¨Ó5ÀÕEÜ�LŠL˜Ô3Ü�H‰HØWØØôô ˜F“|ˆØ)-ˆÐ%Ñ&Ü�‰ð9àØô		
ð ×%Ò%Ð&6ÑG¸,ÑGÕGûð1Hús(   ‚ •E$¡*EÁA1EÂ>BEÅE$ÅE$c               ó    € V ^8„  d   QhRRRR/# )rO   rP   r!   Úqcfgr   rT   )rU   s   "rV   rW   rW   Þ  s   € ÷ ñ ¨Gð ¸>ñ rX   c                ó.  € VP                   \        P                  \        P                  39  d   V# V R \        P
                  \        P                  39   d   V# \        V\        P                  4      '       g   V#  \        WP                   VP                  4      p\        TRR 4      ;'       g    . pY$9   g   \        P                  T9  d   T# \        P!                  RTP"                   R24       \        P                  #   \         d    Tu # i ; i)NÚSUPPORTS_DTYPESzBLoading Quantized Model: Auto fix `dtype` to `torch.float16` for `rù   )Úquant_methodr   ÚAWQÚPAROr!   r  ÚAUTO_TRAINABLEr‘   ru   Údtyper4   rU   rk   r»   Úfloat16r¼   r  Ú__name__)rP   rV  r]  ÚqlinearÚsupported_dtypess   $$$  rV   Ú_coerce_quantized_awq_dtyperb  Þ  sá   € Ø×Ñ¤§¡¬V¯[©[Ð 9Ô9ØˆØ�4œŸ™¤w×'=Ñ'=Ð>Ô>ØˆÜ�eœUŸ[™[×)Ò)ØˆðÜ(¨×2CÑ2CÀTÇ[Á[ÓQˆô ˜wÐ(9¸4Ó@×FÐFÀBÐØÔ ¤E§M¡MÐ9IÔ$IØˆä‡H�HÐQÐRY×RbÑRbÐQcÐcdÐeÔfÜ�=‰=Ðøô ô ØŠðús   Á= D ÄDÄDc               ó   € V ^8„  d   QhRR/# )rO   Úrequirementsz	List[str]rT   )rU   s   "rV   rW   rW   ó  s   € ÷ `ñ `¨iñ `rX   c                ó,  € Vf   R # V FL  p\        V4      w  r4p \        V4      pVf   K#  \        WeV4      '       g   \        V  RV RV RV R24      hKN  	  R #   \         d.    Tf   \        T  RT RT R24      h\        T  RT RT R24      hi ; i)Nz requires version z, but current z version is Ú z requires package z, but z not installed.)r+  r   r2  rk   r   )rÍ   rd  r'  r)  r/  r*  r-  s   &&     rV   Úcheck_versionsrg  ó  sÕ   € ØÒÙÛˆÜ*;¸CÓ*@Ñ'ˆÐ'ð		`Ü '¨£ÐØÒÙÜ#Ð$5È×RÒRÜ  K =Ð0BÀ3À%À~ÐVYÐUZÐZfÐgxÐfyÐyzÐ!{Ó|Ð|ñ Só øô $ô 	`ØÒÜ  K =Ð0BÀ3À%ÀvÈcÈUÐRaÐ!bÓcÐcÜ ˜}Ð,>¸s¸eÀ6È#ÈÈoÐ^Ó_Ð_ð	`ús   ›A¬)AÁ8Bc               ó   € V ^8„  d   QhRR/# )rO   Úmodel_init_kwargsrÔ   rT   )rU   s   "rV   rW   rW     s   € ÷ 7ñ 7¨ñ 7rX   c                óh   € \        \        P                  4      \        R4      8¼  d   WR&   R# WR&   R# )zÿ
Set dtype argument in a version-compatible way for Transformers.
See: https://github.com/huggingface/transformers/releases/tag/v4.56.0

Args:
    model_init_kwargs (dict): kwargs used to initialize model
    torch_dtype: torch dtype (e.g. torch.float16)
z4.56.0r]  Útorch_dtypeN)r   rÌ   Ú__version__)ri  rk  s   &&rV   Úset_dtype_compatrm    s,   € ô Œ|×'Ñ'Ó(¬G°HÓ,=Ô=Ø%0˜'Ó"à+6˜-Ó(rX   c                ó6  € \         P                  P                  V 4      pV'       g&   \         P                  P                  V 4      '       d    \         P                  P	                  V 4      # VP                  \        R 4       R R lp\        \        V VVR7      # )Nc               ó   € V ^8„  d   QhRR/# )rO   Úremovedr�   rT   )rU   s   "rV   rW   Ú*get_model_local_path.<locals>.__annotate__  s   € ÷ cñ c˜iñ crX   c                óP   € \         P                  R RP                  V 4      4       R# )z9Loader: dropping unsupported snapshot_download kwargs: %srZ   N)r¼   Údebugri   )rp  s   &rV   Ú_log_removedÚ*get_model_local_path.<locals>._log_removed  s   € Ü�	‰	ÐMÈtÏyÉyÐY`ÓOaÖbrX   )rP  Ú
on_removed)	r§   rH  ÚisdirÚisabsÚnormpathr–   r&   r8   r   )Úpretrained_model_id_or_pathrP  Úis_localrt  s   &,  rV   Úget_model_local_pathr|    sq   € Ü�w‰w�}‰}Ð8Ó9€Hß”2—7‘7—=‘=Ð!<×=Ò=Ü�w‰w×ÑÐ ;Ó<Ð<Ø
‡J�JÔ*¨DÔ1õcô ÜØ#ØØô	ð rX   c               ó    € V ^8„  d   QhRRRR/# )rO   ri  r   rR   rT   )rU   s   "rV   rW   rW   "  s   € ÷ 6ñ 6°$ð 6¸4ñ 6rX   c                ó   € \        V 4      # rc   )r(   )ri  s   &rV   Ú_get_tokenizer_load_kwargsr  "  s   € Ü"Ð#4Ó5Ð5rX   c               ó$   € V ^8„  d   QhRRRRRR/# )rO   rB  r’   Úhf_gguf_load_kwargsúDict[str, str]rR   rö   rT   )rU   s   "rV   rW   rW   &  s#   € ÷ ñ ¸#ð ÐTbð Ðgtñ rX   c                óÜ   € VP                  R 4      pV'       g   R# \        P                  P                  \	        V 4      V4      p\        P                  P                  V4      '       g   R# V# )Ú	gguf_fileN)rh   r§   rH  ri   r’   rJ  )rB  r�  r„  Úcheckpoint_paths   &&  rV   Ú#_resolve_local_gguf_checkpoint_pathr†  &  sN   € Ø#×'Ñ'¨Ó4€IßÙä—g‘g—l‘l¤3Ð'7Ó#8¸)ÓD€OÜ�7‰7�>‰>˜/×*Ò*ÙØÐrX   c               ó$   € V ^8„  d   QhRRRRRR/# )rO   rB  r’   r�  r‚  rR   zItuple[Optional[str], Optional[internal_gguf.GGUFQuantizedCheckpointSpec]]rT   )rU   s   "rV   rW   rW   1  s'   € ÷ &ñ &Øð&à'ð&ð Oñ&rX   c                óþ   € \         P                  ! 4       '       g   R# \        W4      pVf   R#  \         P                  ! V4      pTf   R# Y#3#   \         d#   p\
        P                  RY$4       Ru R p?# R p?ii ; i)Nz2Loader: failed to inspect GGUF checkpoint `%s`: %s)NN)r    Únative_quantized_loader_enabledr†  Úinspect_quantized_checkpointrá   r¼   rs  )rB  r�  Úgguf_checkpoint_pathÚspecr5  s   &&   rV   Ú)_resolve_native_quantized_gguf_checkpointr�  1  s…   € ô ×8Ò8×:Ò:ØÐä>Ð?OÓeÐØÒ#ØÐðÜ×9Ò9Ð:NÓOˆð
 ‚|ØÐØÐ%Ð%øô ô Ü�	‰	ÐFÐH\ÔbØÕûðús   °A ÁA<ÁA7Á1A<Á7A<c               ó$   € V ^8„  d   QhRRRRRR/# )rO   r±   útorch.nn.Moduler]   r’   rR   ztuple[torch.nn.Module, str]rT   )rU   s   "rV   rW   rW   G  s"   € ÷ ñ ˜ð °cð Ð>Yñ rX   c                óf   € VP                  R 4      w  r#pV'       d   V P                  V4      MT pWT3# )Ú.)Ú
rpartitionÚget_submodule)r±   r]   Úmodule_namerÃ   Ú	attr_namerÂ   s   &&    rV   Ú_resolve_model_slotr–  G  s2   € Ø $§¡°Ó 4Ñ€K�Iß1<ˆU× Ñ  Ô-À%€FØÐÐrX   c               ó$   € V ^8„  d   QhRRRRRR/# )rO   r±   r�  r]   r’   rR   útorch.TensorrT   )rU   s   "rV   rW   rW   M  s&   € ÷ Cñ C _ð C¸Cð CÀLñ CrX   c                óÆ   € \        W4      w  r#W2P                  9   d   VP                  V,          # W2P                  9   d   VP                  V,          # \        R V R24      h)úLoader: model slot `ú` does not exist.)r–  Ú_parametersÚ_buffersÚKeyError)r±   r]   rÂ   r•  s   &&  rV   Ú_lookup_model_slot_tensorrŸ  M  s[   € Ü+¨EÓ8Ñ€FØ×&Ñ&Ô&Ø×!Ñ! )Õ,Ð,Ø—O‘OÔ#Ø�‰˜yÕ)Ð)Ü
Ð)¨$¨Ð/@ÐAÓ
BÐBrX   c               ó(   € V ^8„  d   QhRRRRRRRR/# )	rO   r±   r�  r]   r’   Útensorr˜  rR   rS   rT   )rU   s   "rV   rW   rW   V  s/   € ÷ Cñ C _ð C¸Cð CÈð CÐZ^ñ CrX   c                óX  € \        W4      w  r4VP                  4       pWCP                  9   dÜ   VP                  V,          pVe^   VP                  VP                  8w  g   VP                  VP                  8w  d(   VP                  VP                  VP                  R7      p\        V\        P                  P                  4      '       d   VP                  MRp\        P                  P                  W&R7      VP                  V&   R # WCP                  9   d…   VP                  V,          pVe^   VP                  VP                  8w  g   VP                  VP                  8w  d(   VP                  VP                  VP                  R7      pW#P                  V&   R # \        RV R24      h)N©r|   r]  F)Úrequires_gradrš  r›  )r–  Ú
contiguousrœ  r|   r]  Útor‘   ru   ÚnnÚ	Parameterr¤  r�  rž  )r±   r]   r¡  rÂ   r•  Úcurrentr¤  s   &&&    rV   Ú_assign_model_slot_tensorrª  V  sA  € Ü+¨EÓ8Ñ€FØ×ÑÓ €Fà×&Ñ&Ô&Ø×$Ñ$ YÕ/ˆØÒ F§M¡M°W·^±^Ô$CÀvÇ|Á|ÐW^×WdÑWdÔGdØ—Y‘Y g§n¡n¸G¿M¹M�YÓJˆFÜ1;¸GÄUÇXÁX×EWÑEW×1XÒ1X˜×-Ò-Ð^cˆÜ(-¯©×(:Ñ(:¸6Ð(:Ó(_ˆ×Ñ˜9Ñ%Ùà—O‘OÔ#Ø—/‘/ )Õ,ˆØÒ F§M¡M°W·^±^Ô$CÀvÇ|Á|ÐW^×WdÑWdÔGdØ—Y‘Y g§n¡n¸G¿M¹M�YÓJˆFØ%+�‰˜	Ñ"Ùä
Ð)¨$¨Ð/@ÐAÓ
BÐBrX   c               ó$   € V ^8„  d   QhRRRRRR/# )rO   r±   r�  rÇ   r   rR   údict[str, str]rT   )rU   s   "rV   rW   rW   l  s(   € ÷ @ñ @¨/ð @ÐCSð @ÐXfñ @rX   c                ó(  € ^ RI Hp VP                  P                  VP                  VP
                  4      pW2P
                  Jd'   \        RVP                   RVP                   R24      hV! VP                  4       R7      pVP                  W4      # )r   NzhLoader: native quantized GGUF loading only supports the default tensor processor. Actual processor for `z`: `ú`.)rÇ   )
Ú(transformers.modeling_gguf_pytorch_utilsÚmodeling_gguf_pytorch_utilsÚTENSOR_PROCESSORSrh   Ú
model_typeÚTensorProcessorÚNotImplementedErrorr_  Úto_dictÚget_gguf_hf_weights_map)r±   rÇ   Ú
gguf_utilsÚprocessor_clsÚ	processors   &&   rV   Ú_build_gguf_tensor_key_mappingrº  l  s�   € ÝAà×0Ñ0×4Ñ4°V×5FÑ5FÈ
×HbÑHbÓc€MØ×6Ñ6Ó6Ü!ð%Ø%+×%6Ñ%6Ð$7°t¸M×<RÑ<RÐ;SÐSUðWó
ð 	
ñ
  V§^¡^Ó%5Ô6€IØ×-Ñ-¨eÓ?Ð?rX   c               ó(   € V ^8„  d   QhRRRRRRRR/# )	rO   r±   r�  r‹  r’   Útensor_key_mappingr¬  rR   rS   rT   )rU   s   "rV   rW   rW   z  s2   € ÷ 6ñ 6àð6ð ð6ð 'ð	6ð
 
ñ6rX   c                ó0  € \         P                  ! V4      p\        4       pVP                   EFÅ  pVP	                  VP
                  4      pVf   K%  VP                  R4      w  rxp	V'       d   V P                  V4      MT p
Tp\        V
\        4      '       dð   V	R8X  dé   V'       d   V R2MRp\        P                  ! \        P                  ! VP                  \        P                  RRR7      4      p\!        W4      p\#        VP$                  4      \#        VP$                  4      8w  d<   \'        RV R	\#        VP$                  4       R
\#        VP$                  4       R24      h\)        WV4       VP+                  V4       EK[  \!        W4      p\         P,                  ! VP                  VP.                  VP0                  VP2                  R7      p\)        WV4       VP+                  V4       EKÈ  	  . pV P5                  4        FG  w  pp\        V\        4      '       g   K  V'       d   V R2MRpVV9  g   K6  VP7                  V4       KI  	  V'       d+   \'        RRP9                  \;        V4      4      ,           4      hV P=                  4        R # )Nr‘  Úweightz.qweightÚqweightTÚC)r]  ÚcopyÚorderz)Loader: GGUF qweight shape mismatch for `z`. Expected z, got r£  zELoader: GGUF checkpoint did not populate required quantized weights: rZ   )r    Ú
GGUFReaderr“   Útensorsrh   r]   r’  r“  r‘   r   ru   Ú
from_numpyÚnpÚarrayÚdataÚuint8rŸ  rÖ   Úshaper†   rª  rÑ   Údequantize_to_torchÚtensor_typer|   r]  rý   rÓ   ri   rj   Útie_weights)r±   r‹  r¼  ÚreaderÚloadedr¡  Útarget_namer”  rÃ   r•  Útarget_moduleÚresolved_target_nameÚpackedÚexpectedÚ	referenceÚweightsÚmissing_qweightsrÂ   Úqweight_names   $$$                rV   Ú*_load_quantized_gguf_checkpoint_into_modelrÙ  z  s-  € ô ×%Ò%Ð&:Ó;€FÜ“u€Fà—.•.ˆØ(×,Ñ,¨V¯[©[Ó9ˆØÒÙà$/×$:Ñ$:¸3Ó$?Ñ!ˆ˜	ß<G˜×+Ñ+¨KÔ8ÈUˆØ*Ðä�m¤_×5Ò5¸)ÀxÔ:Oß?J k ]°(Ñ#;ÐPYÐ Ü×%Ò%¤b§h¢h¨v¯{©{Ä"Ç(Á(ÐQUÐ]`Ô&aÓbˆFÜ0°ÓMˆHÜ�V—\‘\Ó"¤e¨H¯N©NÓ&;Ô;Ü"Ø?Ð@TÐ?Uð V Ü % h§n¡nÓ 5Ð6°f¼UÀ6Ç<Á<Ó=PÐ<QÐQRðTóð ô & eÀ6ÔJØ�J‰JÐ+Ô,Úä-¨eÓJˆ	Ü×3Ò3Ø�K‰KØ×ÑØ×#Ñ#Ø—/‘/ô	
ˆô 	" %¸wÔGØ�
‰
Ð'×(ñ= !ð@ ÐØ$×2Ñ2Ö4Ñˆ�VÜ˜&¤/×2Ò2Ùß3>˜+˜ hÑ/ÀIˆØ˜vÖ%Ø×#Ñ# LÖ1ñ  5÷ ÜØSØ�i‰iœÐ/Ó0Ó1õ2ó
ð 	
ð
 
×ÑÖrX   c                óÎ   € \         \        P                  \        P                  R RRR3R R ll4       pWn        \         RR\        P                  RRR 3R R ll4       pW n        V # )FÚautoNc               ó8   € V ^8„  d   QhRRRRRRRRR	R
RRRRRR/# )rO   rz  r’   Úquantize_configr   rP   úUnion[str, BACKEND]r  zUnion[str, int, PROFILE]Útrust_remote_coderž   r]  ú[str | torch.dtype]Ú
device_mapú0Optional[Union[str, Dict[str, Union[int, str]]]]r|   úOptional[Union[str, int]]rT   )rU   s   "rV   rW   Ú!ModelLoader.<locals>.__annotate__µ  se   € ÷ |ñ |à),ð|ð 0ð|ð )ð	|ð
 .ð|ð  $ð|ð 'ð|ð Ið|ð .ñ|rX   c	                óà  € ^ RI p
V
P                  P                  4        \        VV	V P                   R2R7      p\        V	V P                   R2VR7      p\        V4      p\        V4      p\        V	4      p\        V	4      pVP                  \        R4       VP                  RV4      p\        V3/ VB p\        WåR7      pW\R&   \        P                  ! V3/ VBVB p\         P"                  ! VP$                  4       \'        WõR7       \)        Wï4       V	P+                  RR4      pVe-   VR	8w  d&   \,        P/                  R
V R24       \1        VV4       \3        W‡4      p\5        VV4      pV P6                  '       d   V P6                  pM=Ve#   VR	8X  g   \9        WjP:                  4      '       g   \=        V\?        V4      RR7      p\9        WjP:                  4      '       d   \A        V4      V8w  d   \C        Wö4       \E        V3RVRV/\G        V	4      B p\I        VV4      w  pp\K        VVR7      pVEfî   Ve¿   \M        VVR7      '       dC   V\N        PP                  8w  d-   \,        P/                  RVPR                  VVPT                  4       Mi\        V	4      pVP                  RR4       \,        P/                  RVV P                  VPR                  4       V PV                  ! R<RVRVRVRVRVRVRV/VB # \        V4      pV'       d   TMR	VR&   \Y        VV4       VP[                  V4       Ve�   VP\                  \^        P`                  Pb                  8X  dt   VR=9   dm   \e        V4      '       d\   VP$                  R8X  g   \g        V4      '       d:   \i        4       '       d*   RV\j        &   \,        P/                  RVPR                  4       \m        V Pn                  V3RV/VB pVVn8        \s        VR7       RpV ! VVRVVVVR7      p\u        VPv                  4       V# \x        Pz                  ! 4       pRR/pVe   \9        V\|        4      '       g   \        R4      hVP�                  4        VP‚                  e   Vf   Ve   \        R 4      hVP„                  V P†                  9  d)   \‰        V  R!V P†                   R"VP„                   R#24      h\‹        V4      pV PŒ                  '       d   V'       g   V'       g   \‰        V R$24      h\�        W P�                  4       \“        4       ;_uu_ 4        VP‚                  f   \5        RR4      VnA        M\•        VP‚                  4      VnA        Ve#   VR	8X  g   \9        WjP:                  4      '       g   \=        WòP‚                  RR7      pTVR&   \Y        WÆ4       V P–                  VR%&   V P™                  WRR&7       ^R'IMHNp \Ÿ        VR(4      '       d   VP                   R)8X  d   RVnQ        V P¤                  '       g    RVnQ        \,        P§                  V  R*24       VP¢                  '       dý   \¨        Pª                  ! V4      p V! V Pn                  3RV/VB p\­        V VRR+7       \        V4      p V P[                  V4       V Vn8        \s        VR7       \®        P°                  ! VVP²                  V \¨        Pª                  ! \µ        V R,R4      4      \¨        Pª                  ! V P·                  VR-7      4      VR.7      pVf   \¹        R/V R024      h\,        P/                  R1V4       M›\,        P/                  R54       \m        V Pn                  V3RV/VBVB p\µ        VRR4      VJ d   \¨        Pª                  ! V4      VnY        \­        V VRR+7       \        V4      p#V#P[                  V4       V#Vn8        \s        VR7       RpRRR4       XP²                  P½                  4       p$. R>Op%\¿        V$V%4      p&V&e	   V&Vn`        M\,        P§                  R64       R7Vn`        VPÃ                  4        Xe   VPÃ                  4       MR V ! VVRVVVVR7      p\u        VPv                  4       \µ        VR8R4      p'V'eO   \µ        VR9R4      ;'       g    \Å        V4      p(V'PÇ                  R:\x        Pz                  ! 4       V,
          V(R;7       V#   \¸         dÛ   p!\»        T!4      '       g   h \,        P§                  R2TT!4       \,        P/                  R34       TP©                  4       p"T"P                  RR4       RT"R4&   \m        T Pn                  T3RT/T"BTB p\µ        TRR4      TJ d   \¨        Pª                  ! T4      TnY        \­        T TRR+7       T"Tn8        \s        TR7       Rp Rp!?!ELåRp!?!ii ; i  + '       g   i     ELö; i)?r   Nz.from_pretrained©Úapi_name©rç  Úexplicit_dtypeÚtokenizer_trust_remote_code©rß  rß  rM   rÛ  z5Loader: overriding attn_implementation in config to `rù   F©rÇ   r|   Úquant_inferenceÚmodel_configr  zLLoader: PROFILE.%s uses dense GGUF import for `%s`; backend `%s` is ignored.zmLoader: detected native quantized GGUF checkpoint `%s`; redirecting `%s` to from_quantized() with PROFILE.%s.Úmodel_id_or_pathrá  r|   rP   r]  Úqwen3Úflash_attention_2zDLoader: Auto enabling flash_attention_2 for dense Bonsai PROFILE.%s.rÇ   r°   )Úturtle_modelÚ	quantizedrÝ  Ú	tokenizerrß  rB  r<  rƒ   zJ`quantize_config` must be passed and be an instance of BaseQuantizeConfig.zäPassing device and device_map is not allowed when QuantizeConfig.device is set. Non-quantized model is always loaded as cpu. Please set QuantizeConfig.device for accelerator used in quantization or do not set for auto-selection.z only supports desc_act=z", but quantize_config.desc_act is r‘  úW requires trust_remote_code=True. Please set trust_remote_code=True to load this model.Ú
_fast_init©rB  Úload_quantized_model)Úbuild_shell_modelÚ
hidden_actÚxieluzO doesn't support offload_to_disk, set quantize_config.offload_to_disk to False.r¶   Úmodule_tree)Útarget_model)rB  rÇ   ri  rü  Úhf_conversion_map_reversedrý  zLoader: can't open model path `z` for offload_to_disk.z;Loader: using checkpoint-backed lazy turtle source for `%s`ziLoader: meta-device shell build failed for `%s`; falling back to direct CPU load without turtle_model: %szULoader: loading model directly to CPU (meta shell unsupported; turtle_model disabled)Úlow_cpu_mem_usagezMLoader: loading model directly to CPU (not using meta device or turtle_model)zMModel: can't get model's sequence length from model config, will set to 4096.é   Úquant_region_timerrB  Ú
model_load)ÚsourcerT   >   NrÛ  ©Úmax_position_embeddingsÚ
seq_lengthÚn_positionsÚmultimodal_max_length)dÚtorch._dynamoÚ_dynamoÚdisabler,   r_  r-   r#   r$   r(   rÔ   r–   r&   r|  r/   r   rG  r¹   Úreplace_fused_blocksr²  r+   r.   rh   r¼   r  rã   r5   r2   Úrequire_dtyper‘   r]  r;   r7   r'   r0   r*   r  r�  r  r  r!   r  r]   rÛ   Úfrom_quantizedrm  Úupdater  r    r  r  r  rÊ   r   ÚATTN_IMPLEMENTATIONrT  rO  Ú_model_init_kwargsr²   ró   r±   ÚtimeÚperf_counterr   ÚAttributeErrorÚcalculate_bits_per_weightr|   Údesc_actÚsupports_desc_actrk   r)   Úrequire_trust_remote_coderg  Úrequire_pkgsr1   rL   Úrequire_fast_initÚbefore_model_loadÚutils.hfrù  r‡   rú  Úoffload_to_diskÚsupport_offload_to_diskr½   rÁ  ÚdeepcopyrÅ   r   Úmaybe_createrÇ   r»   Ú"resolve_hf_conversion_map_reversedr†   r8  rµ  r=   ÚseqlenÚevalr’   Úrecord))r¿   rz  rÝ  rP   r  rß  r]  rá  r|   ri  ru   r�  Ú"model_init_kwargs_without_internalrê  rB  rÇ   Ú
atten_implÚresolved_devicerô  r‹  r  Úeffective_profileÚredirect_kwargsÚhf_model_init_kwargsr±   rò  ÚinstanceÚ
load_startÚcpu_device_mapÚnative_supportrù  Úshell_configÚshell_model_init_kwargsr5  Úfallback_init_kwargsÚdirect_model_init_kwargsrî  Úseq_len_keysÚconfig_seq_lenÚtimerÚsource_labels)   &&&&&&&&&,                               rV   rG  Ú$ModelLoader.<locals>.from_pretrained´  sÑ	  € ó 	Ø�‰×ÑÔä&LØ'ØØŸ™�~Ð%5Ð6ô'
Ð#ô ,ØØŸ™�~Ð%5Ð6Ø ô
ˆô
 $ GÓ,ˆÜ# GÓ,ˆÜ5Ð6GÓHÐÜ-1Ð2CÓ-DÐ*Ø*×.Ñ.Ô/JÈDÔQà&H×&LÑ&LÐMjÐl}Ó&~Ð#Ü/Ð0KÑrÐOqÑrÐÜ5Ð6FÔlÐàBSÐ+>Ñ?ä×+Ò+Ð,<ÑzÐ@bÐzÐfyÑzˆä×$Ò$ V×%6Ñ%6Ô7ä" 6ÕOÜ(Ð)9ÔBà&×*Ñ*Ð+@À$ÓGˆ
àÒ! j°FÔ&:Ü�H‰HÐLÈZÈLÐXYÐZÔ[Ü)¨&°*Ô=ä5°fÓIˆÜ,¨_¸gÓFˆà××ÐØ×%Ñ%‰EØŠ]˜e vœo´ZÀÇ{Á{×5SÒ5SÜØÜ-¨oÓ>Ø %ôˆEô �eŸ[™[×)Ò)Ô.AÀ&Ó.IÈUÔ.Rä Ô.ä%Øñ
àð
ð :ð
ô )Ð):Ó;ñ	
ˆ	ô 3\ØØó3
Ñ/ÐÐ/ô 9Ø/Øô
Ðð
 Ó"Ø Ò,Ü5Ø&7Ø-÷ó ð ¤'§,¡,Ô.ÜŸ™ØjØ-×2Ñ2Ø0Ø#ŸM™Mô	øô '+Ð+<Ó&=�OØ#×'Ñ'Ð(EÀtÔLÜ—H‘Hð HØ,ØŸ™Ø)×.Ñ.ô	ð ×-Ò-ñ 	Ø)Dð	à#-ð	ð  &ð	ð !(ð		ð
 $ð	ð +<ð	ð 5Pð	ð *ñ	ð 	ô $(Ð(JÓ#KÐ ß?I±ÈvÐ  Ñ.ÜÐ1°5Ô9Ø ×'Ñ'Ð(;Ô<à!Ò-Ø%×2Ñ2´m×6XÑ6X×6]Ñ6]Ô]Ø .Ô0Ü4°_×EÒEØ×&Ñ&¨'Ô1Ô5KÈF×5SÒ5SÜ-×/Ò/à<OÐ$Ô%8Ñ9Ü—‘Ð_Ðar×awÑawÔxäHØ—
‘
Ø ñð ðð 'ñ	ˆEð (<ˆEÔ$Ü$¨5Õ1àˆLáØØ)ØØ /Ø#Ø"3Ø!1ôˆHô 2°(·.±.ÔAàˆOä×&Ò&Ó(ˆ
ð ˜e˜ˆàÒ"¬*°_ÔFX×*YÒ*YÜ Ð!mÓnÐnà×1Ñ1Ô3à×!Ñ!Ò-ØÒ! ZÒ%;Ü$ð  &Ló  Mð  Mà×#Ñ#¨3×+@Ñ+@Ô@Ü ˜uÐ$<¸S×=RÑ=RÐ<Sð T@Ø@O×@XÑ@XÐ?YÐYZð\ó ]ð ]ô BÐBRÓSˆØ×(×(Ð(×1BÏ>ÜØ.Ð/ð  0Gð  Hóð ô 	�s×,Ñ,Ô-ä#×%Õ%à×%Ñ%Ò-Ü);¸DÀ$Ó)G�Õ&ä)9¸/×:PÑ:PÓ)Q�Ô&àŠ} ¨¤´zÀ%ÏÉ×7UÒ7Uä"¨&×9OÑ9OÐafÔg�ð @NÐ.¨|Ñ<ÜÐ?ÔGØ?B×?TÑ?TÐ.¨|Ñ<ð ×!Ñ! #Ð_dÐ!ÔeÝ4ô �v˜|×,Ò,°×1BÑ1BÀgÔ1MØ27�Ô/ð ×.×.Ð.Ø27�Ô/Ü—‘˜C˜5Ð oÐpÔqà×.×.Ð.Ü#Ÿ}š}¨VÓ4�ð5Ù-¨c¯j©jÑtÀÐtÐQsÑt�Eô8 0°°UÈUÕSÜ.2Ð3UÓ.VÐ+Ø+×2Ñ2Ð3FÔGØ/F�EÔ,Ü,°5Õ9Ü#-×#:Ò#:Ø)9Ø$Ÿ|™|Ø*AÜ$(§M¢M´'¸#¸}ÈdÓ2SÓ$TÜ37·=²=Ø×BÑBÐPUÐBÓVó4ð &+ô	$�Lð $Ò+Ü*Ø=Ð>NÐ=OÐOeÐfóð ô —H‘HØUØ(õô
 —‘ÐhÔiÜLØ—J‘JØ$ñð "ðð 9ð	ð
 *ñ�ô ˜5 (¨DÓ1°VÓ;Ü#'§=¢=°Ó#8�E”LÜ+¨C°ÈÕOÜ+/Ð0RÓ+SÐ(Ø(×/Ñ/Ð0CÔDØ+C�Ô(Ü(¨uÕ5à#�÷Q &ðT —|‘|×+Ñ+Ó-ˆÚhˆÜ,¨\¸<ÓHˆØÒ%Ø)ˆE�Lä�H‰HÐdÔeØˆEŒLà�
‰
ŒØ+Ò7ˆ×ÑÔ¸TøáØØ%ØØ+ØØ/Ø-ô
ˆô 	.¨h¯n©nÔ=ä˜Ð"6¸Ó=ˆØÒÜ" 8Ð-?ÀÓF×jÐjÌ#ÐNiÓJjˆLØ�L‰L˜¤t×'8Ò'8Ó':¸ZÕ'GÐP\ˆLÔ]àˆøôG $ô (Ü5°c×:Ò:Øä—H‘Hð DØ(Øôô
 —H‘HÐtÔuØ+M×+RÑ+RÓ+TÐ(Ø(×,Ñ,¨\¸4Ô@Ø@EÐ(Ð)<Ñ=ÜPØŸ
™
Ø(ñð  &ðð /ð	ð
 .ñ�Eô ˜u h°Ó5¸Ó?Ü'+§}¢}°VÓ'<˜œÜ/°°UÈUÕSØ/C�EÔ,Ü,°5Õ9Ø#'–Lûð3(ú÷G &×%Ð%úsE   Ô;C+gØ'0gÙgÙ/c4ÚE+gã4gã?Cgçgçgçgçg-	c               ó4   € V ^8„  d   QhRRRRRRRRR	R
RRRR/# )rO   rï  rö   rá  râ  r|   rã  rP   rÞ  ÚadapterzOptional[Adapter]r]  rà  rß  rž   rT   )rU   s   "rV   rW   rä  ö  sZ   € ÷ Rñ Rà+ðRð IðRð .ð	Rð
 )ðRð 'ðRð 'ðRð  $ñRrX   c                ó�1  ajakalam€ ^ RI p	V	P                  P                  4        \        VVV P                   R2R7      p\        VV P                   R2VR7      p\        V4      p
\        V4      pVP                  \        R4       VP                  RV4      pTp\        V\        4      '       d   TMRpVfa   Rp\        V\        4      '       d   RV9   d   TpM4\        W9P                  4      '       d   VP                  e   \        V4      pVe   RV/pTpVf   TMRp\        VV4      p\        V4      p\!        W44      pVf   \#        V4      p\%        V3/ VB p\'        VVR7      p\)        V4      p V P*                  '       d   V'       g   V'       g   \-        V R	24      h\/        W P0                  4       VP                  R
R4      pVP                  RR4      pVP                  RR4      pVP                  RR4      pVP                  RR4      pVP                  RR4      pVP                  RR4      pVP                  RR4      pVP                  RR4      pVP                  RR4      pR
VRVRVRVRVRVRVRVRRRVRV/p\2        P4                  ! V3RV/VBV
B p\6        P8                  ! VP:                  4       \=        VVR7       \?        VV4       V P@                  '       d   V P@                  pVe#   VR8X  g   \        WiPB                  4      '       g   \E        V\G        V4      RR7      p\        WiPB                  4      '       d   \I        V4      V8w  d   \K        VV4       \M        VV
4      w  pp V e2   \O        V PP                  \R        PT                  V PV                  R7      p!M\N        P4                  ! V3/ VBVB p!V!PY                  4       p"\[        V!P\                  V!P^                  4      p#\        VV"R7      pV em   V P`                  \b        Pd                  Pf                  8X  dD   V\h        Pj                  \h        Pl                  \h        Pn                  09  d   \-        RV R24      hV et   V P`                  \b        Pd                  Pp                  8X  dK   V\h        Pj                  8X  d   \h        Pl                  pM$V\h        Pl                  8w  d   \-        RV R24      hV#\r        Pt                  8X  Ed>   V\h        Pj                  \h        Pv                  \h        Px                  39  d   \{        R4      hV\h        Pj                  8X  dp   V	P|                  P                  4       '       d@   \G        V4      \€        P‚                  \€        P„                  39   d   \h        Pv                  pM\h        Px                  pV\h        Pv                  8X  de   V	P|                  P                  4       '       g   \-        R 4      h\G        V4      \€        P‚                  \€        P„                  39  d   \-        R!4      hMUV#\r        P†                  8X  dA   V\h        Pj                  \h        P†                  39  d   \{        R"4      h\h        P†                  pV"\R        Pˆ                  8X  dG   V#\r        PŠ                  \r        PŒ                  39   d"   \Ž        P‘                  R#4       V	P’                  p\•        VV!VR$7      pVe   VV!nK        V!P™                  4        \›        V3R%VRV/V
B omV\h        Pœ                  8X  g   V\h        Pž                  8X  Ed   Rp$\¡        VV#4       V\h        Pœ                  8X  d<   ^R&IQHRp%HSp&HTp'HUp( V'! R{R'VRV/VB okV&! Sk4      SknV        V%! Sk4      p)V)e   V)Skn        T(p$M�V\h        Pž                  8X  dm   ^R(IWHXp*HYp+ \        V4      p,\µ        W=4      P·                  4        F  w  p-p.V,P¹                  V-V.4       K  	  V*! R{R'VRVR)V	P’                  /V,B w  okp/V/SknV        T+p$V ! SkRV!SmRRVVR*7      p0V$V0n]        \½        V0P¾                  4       V0# V#\r        PÀ                  8X  d}   V!PÂ                  \R        Pˆ                  8X  d   \h        PÄ                  M\h        PÆ                  p1V1.p2VV29  d2   V\h        Pj                  8w  d   \{        R+V1PÈ                   R,V R24      hT1pV#\r        PÊ                  8X  dz   V!PÂ                  \R        Pˆ                  8X  d   \h        PÌ                  M\h        PÎ                  p3VV38w  d2   V\h        Pj                  8w  d   \{        R-V3PÈ                   R,V R24      hT3pV\h        PÎ                  \h        PÌ                  39   d   ^R.IhHip4Hjp5 V4RJ d   \-        V54      h\        V4      pV e   Rp6Tp7MdV#\r        Pt                  8X  d   R'.p8MR/V!PÖ                   R0V!PØ                   R12R'.p8R2.p9\Û        R{R3VR4V9R5V8/VB w  p6p:p;R6V:9   d   \-        R74      hT:p7V#V!nn        Rp<\ß        4       ;_uu_ 4        V Pá                  V VRR87       VPâ                  '       dh   \å        \æ        VPâ                  ^ ,          R4      p=V=e?   \é        V=R94      '       d   \å        V=R94      p>M$\é        V=R:4      '       d   \å        V=R:4      p>MRp>MRp>/ p?X>'       dj   \G        V4      \€        P‚                  \€        P„                  39   d<   Ve   VV?\ê        &   M-\í        4       '       d   \ê        R;/p?\Ž        P‘                  R<4       \ï        V?V4        V Pð                  Pò                  ! V3RV/V?B okE\        T SkRRK7       T7Skn‰        T e   E\        SkT4      p<T EP                  4       pHE\        SkTH4      w  pIpJE\        Sk4      pKT EP                  .T EP                  Sk4      ,           pLSkEP!                  4       pMSkEP#                  4       pNTMe   E\%        SkXM4      MRpOXNe   E\%        SkXN4      MRpPE\'        TXOTPRL7      w  pQpRT EP)                  TT!4      pSE\+        XKEP-                  4       4       EF�  olXQ'       d
   SlXO8X  d   K  XR'       d
   SlXP8X  d   K&  T!EP                  '       d   SlT EP                  8X  d   KM  E\.        ;QJ d    Tl3RM lXH 4       F  '       g   K   RM	  RM! Tl3RM lXH 4       4      '       dƒ   E\.        ;QJ d    Tl3RN lXL 4       F  '       g   K   RM	  RM! Tl3RN lXL 4       4      '       gB   E\0        ;QJ d    Tl3RO lXS 4       F  '       d   K   RM	  RM! Tl3RO lXS 4       4      '       d1   SlT EP                  Jd   \Ž        P‘                  RPSl RQ24       XKSl EK@  T EP3                  SkSlXKSl,          T!4      '       d   EKe  \Ž        P‘                  RPSl RQ24       XKSl EK„  	  T!EP4                  '       d¬   T#\r        EP6                  \r        EP8                  39  d   E\;        RRT# R24      h\        TT!P^                  R7      pT\h        Pj                  8X  d   \h        EP<                  pT\h        EP<                  \h        EP>                  39  d   E\;        RST R24      hT#\r        Pt                  8X  d¥   \        T!EP@                  \        4      '       d   T!EP@                  '       g   \-        RT4      hT\h        Px                  8X  d   E\B        ME\D        pTE\G        SkE\+        T!EP@                  EP-                  4       4      T!EP@                  TTRU7       TTpUME\I        SkT!XKTT EP                  TTT6RV7      pURRR4       E\K        SkT7T#4       \        T\        4      '       d   TR|9  d   \-        RW4      hR}RX RY llpV\Ž        P‘                  RZT 24       Tf*   E\        SkXH4      w  pIpJE\M        T4      pWXV! SkTTIXLTW4      pM#\        T4      p\Ž        P‘                  R[T 24       \        T4      pXE\O        SkT4      pTTX8w  d/   E\Q        E\S        Sk4      4      pY\Ž        P‘                  R\TY 24       \        XX4      pZTZEPU                  \        EPW                  E\S        Sk4      R]4      4       \Ž        P‘                  R^T 24       T RJ p[T#\r        EP6                  \r        EPX                  \r        EPZ                  39   Ed   E\]        SkTT7TRRR_7       Rp[T#\r        EP6                  8X  då   T!EP^                  '       gT   T!EPa                  4       '       g=   T!EPc                  4       '       g&   \-        R`\r        EP6                   RaE\d         24      hE\.        ;QJ d+    Rb SkEPg                  4        4       F  '       g   K   RM 	  RM! Rb SkEPg                  4        4       4      '       d&   E\i        SkT!XURc7      ok\r        EP8                  T!nn        T\h        EPj                  \h        EPl                  39   d   T6'       d   \-        Rd4      hT\h        PÆ                  \h        PÄ                  39   d¯   XUE\n        8X  g   T#\r        PÀ                  8X  dŽ   T6'       d   \-        Re4      hT\h        PÆ                  8X  d)   YiP’                  T	EPp                  39  d   \-        Rf4      hT\h        PÄ                  8X  d)   YiP’                  T	EPp                  39  d   \-        Rg4      hT\h        EPr                  \h        EPt                  \h        EPv                  39   d<   T6'       d   \-        Rh4      hYiP’                  T	EPp                  39  d   \-        Ri4      hT\h        PÎ                  \h        PÌ                  39   d.   ^RjI¼H½p\ T\! SkT!XUTT7TT!EP|                  T!EP^                  X[Rk7	      okX['       dV   T\h        EPj                  \h        EPl                  \h        PÎ                  \h        PÌ                  39  d   E\]        SkTT7TRl7       T e   E\        SkXZ4      okE\�        SkTT<Rm7       ME\        SkXZ4      okT#\r        Pt                  8X  d%   T\h        Px                  8X  d   E\B        ME\D        p]MXE\ƒ        T!EP„                  T!EP†                  T!PØ                  T!EP|                  T!EP^                  TT#T"TT!EPˆ                  T6Rn7      p]SkP¬                  EP‹                  4       p^. R~Op_E\�        T^T_4      p`T`e	   X`SknÇ        M\Ž        EP‘                  Ro4       RpSknÇ        T#\r        Pt                  8w  d   E\“        SkT!EP|                  T!Rq7      okSkEP•                  4        T\h        EP–                  8X  d¬   ^ RIÌpa ^ RrIÍHÎpb ^ RsIÏHÐpcHÑpd ^RtIÒHÓpeHÔoj XaEP¯                  4       ;_uu_ 4       pgXe! TSkT!EP‹                  4       T EP                  4      w  phpiXd! TgThRRw7       Xc! TiTgRx,           Ry7       SmEP±                  Tg4       Xb! Tg4      w  okpJTjTkTm3Rz lT nÙ        RRR4       T ! SkRT!SmX]RTTR*7      p0E\µ        T0P¾                  T!EP4                  4       \½        T0P¾                  4       T0#   \ô         Ed©   p@\        T@4      pAT'       Ed�   R=XA9   Ed‡   R>XA9   Ed   RpBR?XA9   d+   XAP÷                  R?4      pC\ù        TC4      ^8¼  d
   XC^,          pBXB'       d»   XBPû                  R@4      '       d¤   \ü        Pþ                  EP                  XB4      '       g~   \ü        EP                  ! \ü        Pþ                  EP                  XB4      RRA7       E\        TBRBRCRD7      ;_uu_ 4       pDTDEP	                  RE4       RRR4       M  + '       g   i     M; i\å        TRFR4      ;'       g    / pETEEP                  RG4      pF\        TF\        4      '       d&   ^ RHI†H‡pG TG! XF\        \å        TRIR4      4      RRJ7       T Pð                  Pò                  ! T3RT/T?B ok Rp@?@ELðh Rp@?@ii ; i  + '       g   i     ELå; i  E\ª         d   pfE\­        Tf4      ! RuRv4      hRpf?fii ; i  + '       g   i     ELL; i)r   Nz.from_quantizedræ  rè  rê  Ú:r<  rë  rõ  Ú	cache_dirrF  FÚresume_downloadÚproxiesÚlocal_files_onlyÚuse_auth_tokenÚrevisionÚ	subfolderÚ_commit_hashrM   Ú%_raise_exceptions_for_missing_entriesrß  rÛ  Trì  )ÚbitsÚmethodÚlm_head)rY  zpNative Q1_0 GGUF checkpoints support BACKEND.AUTO, BACKEND.GGUF_TORCH, or BACKEND.GGUF_TRITON. Actual backend: `r®  zTNative Q2_0 GGUF checkpoints currently require BACKEND.GGUF_TORCH. Actual backend: `zRFORMAT.EXL3 requires BACKEND.AUTO, BACKEND.EXL3_EXLLAMA_V3, or BACKEND.EXL3_TORCH.z$EXL3 CUDA loading requires CUDA/HIP.z-EXL3 CUDA loading requires a CUDA/HIP device.zBFORMAT.BITSANDBYTES requires BACKEND.AUTO or BACKEND.BITSANDBYTES.z<Loading Quantized Model: Auto fix `dtype` to `torch.float16`)rP   rV  r]  rî  )Úget_vllm_deviceÚget_vllm_model_configÚload_model_by_vllmÚvllm_generater±   )Úload_model_by_sglangÚsglang_generater]  )ró  rÝ  rô  Úqlinear_kernelrø  rß  rB  z/FORMAT.MARLIN requires BACKEND.AUTO or BACKEND.z: actual = `z0FORMAT.BITBLAS requires BACKEND.AUTO or BACKEND.)ÚBITBLAS_AVAILABLEÚBITBLAS_INSTALL_HINTzgptq_model-zbit-Úgz.safetensorsrï  Ú
extensionsÚpossible_model_basenamesz.binzoLoading of .bin files are not allowed due to safety. Please convert your model to safetensor or pytorch format.r÷  rÊ   rË   rñ  z&Loader: Auto enabling flash attention2r=  zNo such file or directoryÚ'z__configuration_llama.pyrD  Úwzutf-8)Úencodingz<from transformers.models.llama.configuration_llama import *
Úauto_mapÚAutoModelForCausalLM)Úget_class_from_dynamic_moduleÚ_name_or_path)rF  r¶   )Ú	model_dirÚinput_embed_nameÚoutput_embed_namec              3  óF   <"  € T F  pSP                  V4      x € K  	  R # 5irc   ©Ú
startswith)r^   Úprefixr]   s   & €rV   r`   Ú6ModelLoader.<locals>.from_quantized.<locals>.<genexpr>˜  s   øé € ÐUÑAT°v˜4Ÿ?™?¨6×2Ð2ÓATùó   ƒ!c              3  óF   <"  € T F  pSP                  V4      x € K  	  R # 5irc   r_  )r^   Úignore_moduler]   s   & €rV   r`   rb  ˜  s4   øé € ð  ]añ  R`ð  ANÐ]a×]lÑ]lÐmz×]{Ð]{ó  R`ùrc  c              3  ó`   <"  € T F#  q F  pSP                  V4      '       * x € K  	  K%  	  R # 5irc   )rþ   )r^   Úsublistre  r]   s   &  €rV   r`   rb  ˜  s1   øé € ð hÙH\¸WÒryÐan˜DŸM™M¨-Ó8×8Ñ8ÑryÑ8ÓH\ùs   ƒ+.z
The layer z is not quantized.z@`rotation` is only supported for GPTQ/GPTQ_V2 checkpoints, got `zN`rotation` is only supported with `gptq_torch` or `gptq_triton` backend, got `zGEXL3 checkpoints require `quantization_config.tensor_storage` metadata.)r±   Úmodule_namesÚtensor_storageÚ
module_cls)rV  Úquant_resultrP   Úlm_head_namer|   r]  Ú
is_shardedziIf passing a string for `device_map`, please choose 'auto', 'balanced', 'balanced_low_0' or 'sequential'.c               ó(   € V ^8„  d   QhRRRRRRRR/# )rO   ÚlayerszList[torch.nn.Module]Úignore_modulesr€   zOptional[int]rR   r‚  rT   )rU   s   "rV   rW   Ú9ModelLoader.<locals>.from_quantized.<locals>.__annotate__Ø  s:   € ÷ i	ñ i	ð .ði	ð !6ð	i	ð
 (ði	ð ñi	rX   c                ó  aa a!a"€ Vf   \        V4      pV^8  d=   \        V4      ;'       g    \        P                  p\	        RVP
                   R24      h\        \        V4      4      p/ o V P                  4        UUu/ uF  w  rxW‡bK	  	  uppo"\        W4      o!V V!V"3R lp	 V P                  4       p
V	! W¦^ ,          4       / p\        V4       F7  w  rÍWlV,          ,          pV	! WÞ4       S"P                  V4      pVf   K3  WëV&   K9  	  V F|  pV
e	   VV
J d   K  S"P                  V4      pVf   K&  TpRpV'       d5   VV9   d   VV,          pM#RV9  d   MVP!                  R^4      ^ ,          pK<  Vf
   VR,          pV	! VV4       K~  	  \#        V RR4      pVf*   R F#  p\%        V V4      '       g   K  \#        V V4      p M	  VeN   V	! VVR,          4       \'        \#        \#        V RR4      R	R
4      4      pV
e   V'       d   V	! VV^ ,          4       \)        V P+                  4       V P-                  4       4       UaUu. uFJ  w  op\.        ;QJ d    V3R lS  4       F  '       g   K   RM	  R
M! V3R lS  4       4      '       d   KH  SNKL  	  ppp\1        S"P3                  4       4      pV'       d†   VR,          pV Fv  pTpV'       d.   VV9  d'   RV9  d   RpK  VP!                  R^4      ^ ,          pK5  V'       d   S P5                  VS!V,          4       K]  \        P7                  RV R24       Kx  	  \        S P9                  4       4       F›  w  ppV'       g   K  S  Uu0 uF1  pVV8w  g   K  VP;                  V R24      '       g   K(  S V,          kK3  	  ppV'       g   KW  \=        V4      ^8”  g
   VV9  g   Kp  \        P7                  RV R24       S P?                  VR4       K�  	  \)        V P+                  4       V P-                  4       4       UaUu. uFJ  w  op\.        ;QJ d    V3R lS  4       F  '       g   K   RM	  R
M! V3R lS  4       4      '       d   KH  SNKL  	  pppV'       d†   VR,          pV Fv  pTpV'       d.   VV9  d'   RV9  d   RpK  VP!                  R^4      ^ ,          pK5  V'       d   S P5                  VS!V,          4       K]  \        P7                  RV R24       Kx  	  \        P7                  RV R\=        S 4       R\        S P9                  4       4      R,           24       S # u uppi   \         d    \        P                  R4       Rp
 EL¿i ; iu uppi u upi u uppi )u«  
Build a deterministic alternating device_map for multi-GPU loading.
Designed for quantized GPTQ models.

Rules:
  â€¢ Input embedding(s) â†’ GPU 0
  â€¢ Each repeating layer â†’ alternate across GPUs (round-robin)
  â€¢ Output head (lm_head / embed_out):
      â€“ If weight-tied with input embedding â†’ GPU 0
      â€“ Else â†’ last GPU
  â€¢ Ignore modules (e.g., norms, projections) â†’ co-located with last layerâ€™s GPU
Nz,No devices detected for accelerator family `rù   c                óZ   <€ V f   R # SP                  V 4      pVe   SV,          SV&   R # R # rc   )rh   )ÚmodÚ	device_idr]   rá  Údevice_strsÚmod2names   && €€€rV   ÚassignÚWModelLoader.<locals>.from_quantized.<locals>.build_layerwise_device_map.<locals>.assignø  s3   ø€ Ø’;ÙØ—|‘| CÓ(�ØÒ#Ø'2°9Õ'=�J˜tÓ$ñ $rX   zTModel does not implement get_input_embeddings. Skipping input embeddings assignment.r‘  rG  rÇ   Útie_word_embeddingsFc              3  ón   <"  € T F*  pSV8H  ;'       g    SP                  VR ,           4      x € K,  	  R# 5i©r‘  Nr_  ©r^   ÚkÚns   & €rV   r`   ÚZModelLoader.<locals>.from_quantized.<locals>.build_layerwise_device_map.<locals>.<genexpr>G  ó-   øé € ÐOÁJ¸q˜1 ™6×:Ð: Q§\¡\°!°cµ'Ó%:Ô:ÃJùó   ƒ5”!5Tr<  zLoader: unable to map param 'z,' to a module; skipping fallback assignment.zLoader: dropping parent 'z/' from device_map to preserve child placements.c              3  ón   <"  € T F*  pSV8H  ;'       g    SP                  VR ,           4      x € K,  	  R# 5ir|  r_  r}  s   & €rV   r`   r€  k  r�  r‚  zLoader: Built map across z	 GPU(s), z entries. First 8: :Né   Néÿÿÿÿ)Ú	embed_outÚoutput_projectionÚoutput_head) r~   r7   rJ   rr   r†   rÛ   rÕ   r…   rý   r‰   Úget_input_embeddingsr´  r¼   ÚwarningÚ	enumeraterh   Úrsplitr»   r‡   rž   r   Únamed_parametersÚnamed_buffersÚanyr“   r”   Ú
setdefaultr  Úitemsr`  r•   r–   )#r±   r|   ro  rp  r€   r}   Ú
device_idsr  Úmrx  Úin_embÚlayer_name2devidÚiÚlayerÚgpuÚlnamert  r]   ÚownerÚdev_idÚheadÚcandÚtie_cfgrÃ   Úmissingrh  Úfallback_deviceÚ
param_nameÚ
device_strÚ
child_nameÚchild_devicesÚmissing_after_prunerá  rv  rw  s#   &&&&&  `                        @@@rV   Úbuild_layerwise_device_mapÚGModelLoader.<locals>.from_quantized.<locals>.build_layerwise_device_mapØ  s  û€ ð( ÒÜ2°6Ó:�Ø˜!Œ|Ü/°Ó7×EÐE¼6¿:¹:�Ü"Ð%QÐRX×R^ÑR^ÐQ_Ð_`Ð#aÓbÐbäœe H›oÓ.ˆJØ)+ˆJØ).×)<Ñ)<Ô)>Ô?Ñ)>¡ ˜šÑ)>Ò?ˆHä3°FÓEˆK÷>ðØ×3Ñ3Ó5�ñ �6 a�=Ô)ð 02Ðä% fÖ-‘�Ø  X¥Õ.�Ù�uÔ"Ø Ÿ™ UÓ+�ØÔ$Ø.1 UÓ+ñ .ó &�àÒ%¨#°«-Ùà—|‘| CÓ(�Ø’<Ùà�Ø�ßØÐ 0Ô0Ø!1°%Õ!8˜ØØ %Ô'ØØ!ŸL™L¨¨aÓ0°Õ3’Eà’>Ø'¨�^�Fá�s˜FÖ#ñ- &ô6 ˜5 )¨TÓ2ˆDØŠ|ÛM�DÜ˜u d×+Ô+Ü& u¨dÓ3˜Ùñ Nð
 ÒÙ�t˜Z¨�^Ô,äœw¤w¨u°hÀÓ'EÐG\Ð^cÓdÓe�àÒ&ßá˜4 ¨A¥Ô/ô $ E×$:Ñ$:Ó$<¸e×>QÑ>QÓ>SÔTõÙT‘d�a˜ß“sÔOÁJÓO—s—s’sÔOÁJÓO×O÷ �ÑTð ñ ô ˜xŸ™Ó0Ó1ˆLßà",¨R¥.�Û")�JØ&�Eß E°Ô$=Ø eÔ+Ø$&šEà$)§L¡L°°aÓ$8¸Õ$;šEßØ"×-Ñ-¨e°[ÀÕ5QÖRäŸ™Ð#@ÀÀÐLxÐ!yÖzñ #*ô %)¨×)9Ñ)9Ó);Ö$<Ñ ��jßÙñ '1ó!á&0˜
Ø! TÑ)ô +à.8×.CÑ.CÀtÀfÈAÀJ×.Oô +�J˜z×*Ð*Ù&0ð ð !÷
 !‘=¤c¨-Ó&8¸1Ô&<À
ÐR_Ö@_Ü—H‘HÐ8¸¸Ð>mÐnÔoØ—N‘N 4¨Ö.ñ %=ô $ E×$:Ñ$:Ó$<¸e×>QÑ>QÓ>SÔTõ#ÙT‘d�a˜ß“sÔOÁJÓO—s—s’sÔOÁJÓO×O÷ �ÑTð  ñ #÷
 #Ø",¨R¥.�Û"5�Jà&�Eß E°Ô$=Ø eÔ+Ø$&šEà$)§L¡L°°aÓ$8¸Õ$;šEçØ"×-Ñ-¨e°[ÀÕ5QÖRäŸ™Ð#@ÀÀÐLxÐ!yÖzñ #6ô �H‰HÐ0°°
¸)Ü˜“_Ð%Ð%8¼¸j×>NÑ>NÓ>PÓ9QÐRTÕ9UÐ8VðXô Yð Ðùó[ @øô" 'ô Ü—‘ÐrÔsØ“ðüó@ùò4!ùó#sZ   Â U	Â%U ÉU8ÉU8É/U8ÊU8Í+	U>Í9U>ÎU>Ð"VÐ5VÑVÑ,VÕ"U5Õ4U5zLoader: device = z.Loader: honoring explicit device_map request: z-Loader: keeping no-placement modules on CPU: rƒ   zLoader: device_map = )r]  Ú
checkpointrá  Úoffload_state_dictÚoffload_buffersz1Format: Loading of a sym=False model with format=z7 is only supported if produced by gptqmodel version >= c              3  ól   "  € T F*  p\        V\        4      ;'       d    \        VR R4      x € K,  	  R# 5i)ÚREQUIRES_FORMAT_V2FN)r‘   r   r»   )r^   r“  s   & rV   r`   rb  ®  s4   é € ð á,˜ô ˜q¤/Ó2×^Ð^´w¸qÐBVÐX]Ó7^Ô^Û,ùs   ‚4ž4)ÚcfgrN  zSFormat: The loading of sharded checkpoints with Machete is currently not supported.zRFormat: The loading of sharded checkpoints with Marlin is currently not supported.zCMarlin kernel requires dtype=torch.float16 or dtype=torch.bfloat16.zGAWQ Marlin kernel requires dtype=torch.float16 or dtype=torch.bfloat16.zUFormat: The loading of sharded checkpoints with Swordfish is currently not supported.zFSwordfish kernel requires dtype=torch.float16 or dtype=torch.bfloat16.)Úprepare_model_for_bitblas_load)	r±   rV  Úquant_linear_classr]  Úmodel_save_namerá  r  ÚsymÚload_checkpoint_in_model)r]  r¨  rá  )r±   r‹  r¼  )rE  ÚdynamicÚ
group_sizer  r±  rP   rU   rY  r|   Ú
pack_dtyperm  zFcan't get model's sequence length from model config, will set to 4096.r   )Úuse_act_orderrÝ  )Úload)Úsave_configÚ
save_model)Úconvert_gptq_to_mlx_weightsÚmlx_generatezBGPT-QModel load mlx model required dependencies are not installed.zEPlease install via `pip install gptqmodel[mlx] --no-build-isolation`.)Údonate_modelz/config.json)Úconfig_pathc                ó   <€ S! RR SRS/VB # )r±   rô  rT   rT   )rÃ   rP  r»  r±   rô  s   &,€€€rV   rd   Ú5ModelLoader.<locals>.from_quantized.<locals>.<lambda>9  s   ø€ ±<Ñ3kÀeÐ3kÐW`Ð3kÐdjÒ3krX   rT   )rÛ  ÚbalancedÚbalanced_low_0Ú
sequentialrc   r  )Ûr	  r
  Úresetr,   r_  r-   r(   rÔ   r–   r&   r‘   r’   r|   rˆ   r5   r#   r2   r3   r|  r/   r)   r  rk   rg  r  r   rG  r¹   r  r²  r+   r.   r  r]  r;   r7   r'   r0   r�  r   Ú
bits_aliasr   ÚGGUFÚlm_head_quantizedÚexport_quant_methodr   rU   rF  r  r    r  r  r!   r  Ú
GGUF_TORCHÚGGUF_TRITONÚQ2_0r   ÚEXL3ÚEXL3_EXLLAMA_V3Ú
EXL3_TORCHÚ	TypeErrorrv   r„   rJ   rs   rt   ÚBITSANDBYTESrZ  Ú	GEMV_FASTÚLLM_AWQr¼   r  r^  rb  r9  r  r*   ÚVLLMÚSGLANGrn   Ú
utils.vllmrH  rI  rJ  rK  rÇ   Úutils.sglangrL  rM  rœ   r‘  r�  Ú_runtime_generateró   r±   ÚMARLINrY  Ú
AWQ_MARLINÚGPTQ_MARLINr]   ÚBITBLASÚAWQ_BITBLASÚGPTQ_BITBLASÚnn_modules.qlinear.bitblasrO  rP  rE  r´  r?   Úruntime_formatr1   r  rÉ   r»   rÌ   r‡   r  r   rm  rO  Úfrom_configr>  rí   r•   rþ   r§   rH  ÚexistsrK  rL  ÚopenÚwriterh   Ú!transformers.dynamic_module_utilsrY  rÅ   Úcheckpoint_file_namerº  Úextract_layers_noder@   r>   rG  Úget_base_modulesr‰  Úget_output_embeddingsrB   rD   Úsimple_layer_modulesrÕ   Úkeysr�  ÚallÚshould_quantize_modulerõ   ÚGPTQÚGPTQ_V2r´  Ú
GPTQ_TORCHÚGPTQ_TRITONri  r   r   r%   rF   rI   r~   r:   rj   rG   r  ÚfromkeysÚGEMMÚ	PAROQUANTrE   r±  Úis_quantized_by_gptaqÚis_quantized_by_foemr   Úmodulesr<   ÚGPTQ_MACHETEÚAWQ_MACHETEr   Úbfloat16ÚGPTQ_SWORDFISHÚAWQ_SWORDFISHÚ	SWORDFISHÚutils.bitblasr®  r  rH   rÙ  r6   Úruntime_bitsr³  rµ  rµ  r=   r"  r½   rC   r#  ÚMLXÚtempfileÚmlx_lmr·  Úmlx_lm.utilsr¸  r¹  Ú	utils.mlxrº  r»  ÚModuleNotFoundErrorr�   ÚTemporaryDirectoryÚsave_pretrainedÚgenerater  )nr¿   rï  rá  r|   rP   r9  r]  rß  rP  ru   r�  Úkwargs_without_internalrê  r‹   Úexplicit_device_mapr›   Únormalized_devicerB  r.  r<  rF  r=  r>  r?  r@  rA  rB  Úcommit_hashrM   Úcached_file_kwargsrÇ   r‹  r  rV  rÇ  rQ   Úruntime_generaterH  rI  rJ  rK  Úruntime_devicerL  rM  Úsglang_kwargsrf   rÛ   Ú	hf_configr+  Úexpected_marlin_backendÚexpected_marlin_backendsÚexpected_backendrO  rP  rm  r°  rS  rR  Úresolved_archive_fileÚtrue_model_basenameÚgguf_tensor_key_mappingrÍ   Úsupports_flash_attnÚargsr5  r?  Úmissing_fileÚpartsÚfprW  Ú	class_refrY  rå  ro  rÃ   rõ  rp  Úinput_embeddingsÚoutput_embeddingsr\  r]  Úinput_embed_quantizedÚoutput_embed_quantizedrè  Úexl3_module_clsÚpreload_qlinear_kernelr¦  r€   Úoriginal_device_mapÚcpu_modulesÚdispatch_device_mapr²  r®  rN  rî  r3  r4  rÿ  r·  r¸  r¹  rº  Ú	exceptionÚtemp_dirÚmlx_weightsÚ
mlx_configr»  r±   r]   rô  sn   &&&&&&&&,                                                                                                 @@@@rV   r  Ú#ModelLoader.<locals>.from_quantizedõ  s#  û€ ó 	Ø�‰×ÑÔÜAØØØŸ™�~ _Ð5ô
Ðô
 ,ØØŸ™�~ _Ð5Ø ô
ˆô
 6°fÓ=ÐÜ"& v£,ÐØ×#Ñ#Ô$?ÀÔFØ&=×&AÑ&AÐB_ÐarÓ&sÐ#Ø)ÐÜ6@ÐAUÔW[×6\Ò6\Ñ2ÐbfÐàÒ'Ø"ˆOÜ˜&¤#×&Ò&¨3°&¬=Ø"(‘Ü˜F§L¡L×1Ò1°f·l±lÒ6NÜ"% f£+�àÒ*Ø')¨?Ð&;Ð#Ø':Ð$ð ';Ò&B™FÈÐÜ,Ð->Ð@TÓUˆô $ GÓ,ˆÜ# FÓ4ˆØÒ'ô 3°6Ó:ˆFä/Ð0@Ñ\ÐD[Ñ\ÐÜ5Ð6FÐZkÔlÐÜAÐBRÓSˆà2Ø×(×(Ð(×1BÏ>ÜØ#Ð$Ð${Ð|óð ô 	�s×,Ñ,Ô-ð ,×/Ñ/°¸TÓBˆ	Ø0×4Ñ4Ð5EÀuÓMˆØ1×5Ñ5Ð6GÈÓOˆØ)×-Ñ-¨i¸Ó>ˆØ2×6Ñ6Ð7IÈ5ÓQÐØ0×4Ñ4Ð5EÀtÓLˆØ*×.Ñ.¨z¸4Ó@ˆØ+×/Ñ/°¸RÓ@ˆ	Ø-×1Ñ1°.À$ÓGˆØ5×9Ñ9Ð:OÐQUÓVÐð ˜Ø˜nØ�wØ˜ØÐ 0Ø˜nØ˜Ø˜Ø3°UØ˜KØ!Ð#6ð
Ðô $.×#=Ò#=Øñ$
à/ð$
ð !ð$
ð "ñ	$
ˆô 	×$Ò$ V×%6Ñ%6Ô7ä" 6Ð=NÕOÜ(Ð)9¸6ÔBà××ÐØ×%Ñ%ˆEàŠ=˜E VœO´:¸eÇ[Á[×3QÒ3QäØÜ-¨fÓ5Ø $ôˆEô �eŸ[™[×)Ò)Ô.AÀ&Ó.IÈUÔ.Rä ¨Ô.ä2[ØØó3
Ñ/ÐÐ/ð Ò(Ü!Ø&×1Ñ1Ü—{‘{Ø)×;Ñ;ô‰Dô "×1Ò1Ð2BÑtÐFXÐtÐ\sÑtˆDØ"×6Ñ6Ó8ÐÜ*¨4¯;©;¸¿¹ÓDˆÜ# GÐ:MÔNˆð Ò)Ø!×.Ñ.´-×2TÑ2T×2YÑ2YÔYØ¤§¡¬g×.@Ñ.@Ä'×BUÑBUÐVÔVäð$Ø$+ 9¨Bð0óð ð
 Ò)Ø!×.Ñ.´-×2TÑ2T×2YÑ2YÔYàœ'Ÿ,™,Ô&Ü!×,Ñ,‘ØœG×.Ñ.Ô.Ü ð(Ø(/ y°ð4óð ð
 œ&Ÿ+™+Õ%ØœwŸ|™|¬W×-DÑ-DÄg×FXÑFXÐYÔYÜÐ tÓuÐuØœ'Ÿ,™,Ô&Ø—:‘:×*Ñ*×,Ò,Ô1GÈÓ1OÔTZ×T_ÑT_Ôag×alÑalÐSmÔ1mÜ%×5Ñ5‘Gä%×0Ñ0�GØœ'×1Ñ1Ô1Ø—z‘z×.Ñ.×0Ò0Ü$Ð%KÓLÐLÜ)¨&Ó1¼&¿+¹+ÄvÇ{Á{Ð9SÔSÜ$Ð%TÓUÐUøØœF×/Ñ/Ô/ØœwŸ|™|¬W×-AÑ-AÐBÔBÜÐ dÓeÐeÜ×*Ñ*ˆGà¤&§*¡*Ô,°Ä×AQÑAQÔSY×SaÑSaÐ@bÔ1bä�H‰HÐSÔTØ—M‘MˆEä+°GÀ$ÈeÔTˆð ÒØ"ˆDŒLà×&Ñ&Ô(ä%Øñ
àð
ð :ð
ð "ñ	
ˆ	ð ”g—l‘lÔ" g´·±Õ&?Ø#ÐÜ-¨g°{ÔCàœ'Ÿ,™,Ô&÷ó ñ +ñ Ø*ðà&7ðð .ñ�ñ  5°UÓ;�”Ù!0°Ó!7�Ø!Ò-Ø#1�E”LØ#0Ñ àœGŸN™NÔ*ßPä $Ð%<Ó =�Ü"AÀ&Ó"_×"eÑ"eÖ"g‘J�C˜Ø!×,Ñ,¨S°%Ö8ñ #há#7ñ $Ø*ð$à&7ð$ð  Ÿ-™-ð$ð $ñ	$Ñ ��yð  )�”Ø#2Ð ÙØØØ $Ø#Ø#Ø%)Ø"3Ø!1ô	ˆHð *:ˆHÔ&Ü1°(·.±.ÔAØˆOàœ&Ÿ-™-Ô'à<@×<MÑ<MÔQW×Q[ÑQ[Ô<[¤g×&8Ò&8Ôah×atÑatÐ#Ø(?Ð'@Ð$ØÐ6Ô6¸7ÄgÇlÁlÔ;RÜØEÐF]×FbÑFbÐEcÐcoÐpwÐoxÐxzÐ{óð ð .ˆGð œ&Ÿ.™.Ô(à6:×6GÑ6GÌ6Ï:É:Ô6Uœw×2Ò2Ô[b×[oÑ[oÐØÐ*Ô*¨w¼'¿,¹,Ô/FÜØFÐGW×G\ÑG\ÐF]Ð]iÐjqÐirÐrtÐuóð ð 'ˆGà”w×+Ñ+¬W×-@Ñ-@ÐAÔAß\Ø  EÓ)Ü Ð!5Ó6Ð6äÐ/Ó0ÐØÒ(ØˆJØ2‰OàœfŸk™kÔ)Ø,3¨9Ñ(ð " $§)¡) ¨D°·±Ð0AÀÐCØð,Ð(ð
 )Ð)ˆJô FUñ FØ!1ðFà%ðFð *BðFð %ñ	FÑBˆJÐ-Ð/Bð Ð.Ô.Ü ð Fóð ð 4ˆOà)ˆÔð #'ÐÜ#×%Õ%Ø×!Ñ! #Ð8HÐ_cÐ!Ôdà×#×#Ð#Ü%¤l°F×4HÑ4HÈÕ4KÈTÓR�ØÒ*ä˜{Ð,D×EÒEÜ.5°kÐC[Ó.\Ñ+Ü  Ð.D×EÒEÜ.5°kÐCYÓ.ZÐ+øà*.Ñ'à&*Ð#àˆDß"Ô'=¸fÓ'EÌ&Ï+É+ÔW]×WbÑWbÐIcÔ'cØ&Ò2Ø0C�DÔ,Ò-Ü.×0Ò0Ü/Ð1DÐE�DÜ—H‘HÐEÔFÜ˜T 5Ô)ð*ØŸ
™
×.Ò.ØñØ.?ðØCGñ�õT (¨¨UÀTÕJØ)8ˆEÔ&Ø Ò,Ý*HÈÐPVÓ*WÐ'à"%×"9Ò"9Ó";Ðå0°Ð8KÓL‰IˆF�Aå" 5Ó)ˆGØ!Ÿkšk˜]¨S×-AÒ-AÀ%Ó-HÕHˆNØ$×9Ò9Ó;ÐØ %× ;Ò ;Ó =ÐØK[ÒKg�¨uÐ6FÔGÐmqÐØM^ÒMj¥°Ð7HÔ IÐptÐÝ<ZØ*Ø!1Ø"3ô=Ñ9Ð!Ð#9ð $'×#;Ò#;¸FÀDÓ#IÐ Ý˜WŸ\š\›^×,�ß(¨TÐ5EÔ-EÙß)¨dÐ6GÔ.GÙð —<—<‘< D¨C¯KªKÔ$7Ùç”sÔUÑATÓU—s—s’sÔUÑATÓU×UÒU×Y\ÔY\ô  ]añ  R`ó  ]a×Y\×Y\ÒY\ô  ]añ  R`ó  ]a÷  Zaò  Za÷  ehô  ehô hÙH\óh÷  eh÷  ehò  ehô hÙH\óh÷ eò eð  3§;¢;Ó.ÜŸ™ :¨d¨VÐ3EÐ!FÔGØ ˜Úà×1Ò1°%¸¸wÀt½}Èd×SÕSÜ—H‘H˜z¨$¨Ð/AÐBÔCØ ›ñ+ -ð. �}�}‰}Ø¤v§{¢{´F·N²NÐ&CÔCÝ-ØZÐ[fÐZgÐgiÐjóð ô ,¨GÀ$Ç+Á+ÔN�ØœgŸl™lÔ*Ü%×0Ò0�GØ¤7×#5Ò#5´w×7JÒ7JÐ"KÔKÝ-ØhÐipÐhqÐqsÐtóð ð œfŸk™kÔ)Ü! $×"5Ò"5´t×<Ò<ÀD×DW×DWÑDWÜ$Ð%nÓoÐoà:AÄW×EWÑEWÔ:WÖ"6Õ]l�Ý.ØÝ!% d×&9Ò&9×&>Ò&>Ó&@Ó!AØ#'×#6Ò#6Ø.õ	ð *9Ñ&å)3ØØØ!(Ø#Ø!$§¢Ø!ØØ)ô	*Ð&÷W &õl 	% U¨O¸[ÔIäÐ*¬C×0Ò0Ð5Ið Rô 6ô !ð$óð ÷i	ôV 	�‰Ð$ V HÐ-Ô.ØÒ&Ý0°Ð8KÓL‰IˆF�AÝ.¨vÓ6ˆHÙ3°E¸6À6È>Ð[cÓd‰JäÐ1Ó2ˆJÜ�H‰HÐEÀjÀ\ÐRÔSÜ" :Ó.Ðõ 6°e¸ZÓHˆ
ØÐ,Ô,Ý Õ!:¸5Ó!AÓBˆKÜ�H‰HÐDÀ[ÀMÐRÔSô #Ð#6Ó7ÐØ×"Ò"¤4§=¢=Õ1JÈ5Ó1QÐSXÓ#YÔZÜ�‰Ð(¨¨Ð5Ô6à#4¸Ð#<Ð àœ6Ÿ;š;¬¯ª´V×5EÒ5EÐFÕFÝ5ØØà*Ø%Ø#'Ø $õð (-Ð$àœfŸkškÔ)à—x—x‘x¨×(BÒ(B×(DÒ(DÈT×MfÒMf×MhÒMhÜ$ØKÌFÏKÊKÈ=ð  YPõ  Qdð  Peð  fóð ÷ ”3ñ à"Ÿ]š]œ_ó—3—3’3ñ à"Ÿ]š]œ_ó÷ ò õ 9ØØ Ø'=ô�Eô +1¯.ª.�DÔ'à”w×+Ò+¬W×-@Ò-@ÐAÔAßÜ Øióð ð ”w×*Ñ*¬G×,>Ñ,>Ð?Ô?Ø&­/Ô9¸[ÌFÏMÉMÔ=YßÜ Øhóð ð œ'×-Ñ-Ô-°%ÇÁÈuÏ~Ê~Ð?^Ô2^Ü Ð!fÓgÐgØœ'×,Ñ,Ô,°¿}¹}ÈeÏnÊnÐ>]Ô1]Ü Ð!jÓkÐkà”w×-Ò-¬w×/DÒ/DÄg×FWÒFWÐXÔXßÜ Økóð ð Ÿ]™]¨E¯NªNÐ;Ô;Ü Ð!iÓjÐjà”w×+Ñ+¬W×-@Ñ-@ÐAÔAÝFñ 3ØØØ#9ØØ /Ø%ØŸšØ—H’HØ)Aô
ˆE÷ $¨Ü× Ò Ü×ÒÜ× Ñ Ü×Ñð	8
ô )
õ 6ØØà*Ø%õð Ò(Ý)¨%Ð1DÓEˆEÝ6ØØ%9Ø#:öõ *¨%Ð1DÓEˆEàœ&Ÿ+™+Ô%Ø5<Ä×@RÑ@RÔ5RÖ1ÕXg‰Nå0Ø×&Ò&ØŸšØŸ?™?ØŸšØ—H’HØØ"Ø0ØØŸ?š?Ø%ôˆNð —|‘|×+Ò+Ó-ˆÚhˆÝ,¨\¸<ÓHˆØÒ%Ø)ˆE�Lä�HŠHÐ]Ô^ØˆEŒLàœ&Ÿ+™+Ô%å'¨¸T¿]º]Ð\`ÔaˆEà�
Š
Œà”g—k’kÔ!Ûð	Ý'ß@çQð ×,Ò,×.Ô.°(Ù*EÐFVÐX]Ð_c×_kÒ_kÓ_mÐor×ozÒozÓ*{Ñ'�˜Zá˜8 [¸tÕDÙ˜J°H¸~Õ4MÕNØ×)Ò)¨(Ô3á ›>‘��qæk�”÷ /ñ ØØØ ØØ)Ø!%Ø/Ø-ô	
ˆõ 	# 8§>¡>°4·=²=ÔAÜ-¨h¯n©nÔ=Øˆøôo %õ &ô  # 3›x�ç%Ð%Ø.°,Õ>Ø3°|ÕCà#'�LØ˜lÔ*Ø ,× 2Ñ 2°3Ó 7˜Ü˜u›:¨œ?Ø+0°­8˜L÷ %Ø(×1Ñ1Ð2L×MÒMÜ "§¡§¢¨|× <Ò <äŸ›¤B§G¡G§O¢O°LÓ$AÈDÕQÝ! ,°¸g×FÕFÈ"ØŸHšHÐ%dÔe÷ G×F×FÐFúô  ' v¨z¸4Ó@×FÐFÀB�HØ (§¢Ð-CÓ D�IÜ! )¬S×1Ò1ÝcÙ5Ø%Ü¤¨°ÀÓ DÓEØ+/õð  ŸJ™J×2Ò2ØñØ2CðØGKñ–Eð ûðM&ú÷; &×%Ð%ûõl 'ô Ý˜9”oØXØ[óð ûðú÷ /×.Ð.ús0  éBAa9ë+A5Aa9í!A[î DAa9òAa9ò:Aa9óAa9ó1Aa9ôAa9ôAa9ô2Aa9õAa9õAa9õ3Aa9öAAa9÷&3Aa9øC3Aa9üBAa9ÁWAb ÁXA/Ab4Á[Aa6Á[AAa1Á\-Aa1Á]%Aa1Á]*AAa1Á^8A_Á_	Aa1Á_A_$Á_Aa1Á_;A/Aa1Áa*Aa9Áa0Aa1Áa1Aa6Áa6Aa9Áa9Ab
	ÁbAb1ÁbAb,Áb,Ab1Áb4Ac	)Úclassmethodr!   r  r"   rG  r  )r¿   rG  r  s   &  rV   ÚModelLoaderr+  ³  sw   € Üô
 ,3¯<©<Ü07·±Ø&+Ø)/ØKOØ04ö|ó ð|ð|	 *Ôäð LPØ04Ü+2¯<©<Ø)-Ø)/Ø&+öRó ðRðh (Ôà€JrX   )™Ú
__future__r   rÁ  r§   rM  r  Úimportlib.metadatar   r   Ú	itertoolsr   Útypingr   r   r	   r
   ÚnumpyrÆ  ru   rÌ   Úutils.modelscoper   Úutils.structurer   r   Ú
modelscoper   Úhuggingface_hubr¹   Úpackaging.versionr   r   r   r   Útransformers.utilsr   Úadapter.adapterr   Únn_modules.exllamav3r   Únn_modules.exllamav3_torchr   Únn_modules.qlinearr   Únn_modules.qlinear.exllamav2r   Únn_modules.qlinear.ggufr   Úquantizationr   Úquantization.configr   r   r   r   r   Úutilsr    Úutils.backendr!   r"   r#   r$   Úutils.exllamav3r%   r  r&   r'   r(   r)   r*   r+   r,   r-   r.   r/   r0   r1   Úutils.importerr2   r3   r4   r5   r6   r7   Úutils.inspectr8   Úutils.loggerr9   Úutils.modelr:   r;   r<   r=   r>   r?   r@   rA   rB   rC   rD   rE   rF   rG   rH   rI   Ú_constrJ   rK   rL   r¼   r  rÒ  rì  rñ  rÓ  rí  r×  rg   rn   r~   r‰   rœ   r¬   r²   rÅ   rÊ   rÝ   rã   ró   r  r  r  r  r   r+  r2  r8  r@  rT  rb  rg  rm  r|  r  r†  r�  r–  rŸ  rª  rº  rÙ  r+  rT   rX   rV   Ú<module>rG     sÓ  ðõ #ã Û 	Û Û ß <Ý ß .Ó .ã Û Û å :ß ;ñ × Ò Þ,å1ã ß 5ß 5Ý 8å %Ý 2Ý =Ý 0Ý :Ý 5Ý )ß oÕ oÝ !ß RÓ RÝ <÷÷ ÷ ó ÷÷ õ -Ý '÷÷ ÷ ÷ ó ÷$ 6Ñ 5ñ ƒn€à+Ð ð ‡L�LØ�‰Ø�‰ðð ‡N�NØ�‰Ø�‰Ø�‰Ø�‰ð	ðÐ õ]õ	õõ$õBGõ'õõ,õ"*õ<õ#2õL%
õPõõ"õCòBò>õ"RõSõHõ<õ*`õ"7òõ 6õõ&õ,õCõCõ,@õ6ôrYrX   