6iM6ddlZddlmZddlmZddlmZmZmZm Z m Z m Z m Z m Z mZmZmZmZmZmZmZmZmZmZmZmZmZmZddlmZddlmZdd l m!Z!dd l"m#Z#dd l$m%Z%dd l&m'Z'dd l(m)Z)ddl*m+Z+ddl,m-Z-ddl.m/Z/ddl0m1Z1ddl2m3Z3ddl4m5Z5ddl6m7Z7ddl8m9Z9ddl:m;Z;ddlm?Z?ddl@mAZAddlBmCZCddlDmEZEddlFmGZGide#de'de)d e5d!ed"e=d#e?d$e3d%e-d&e7d'e9d(e+d)e/d*eEd+e%d,eGd-eCe1e!e;eAd.ZHidede de d%e d ed!ed"ed#ed$ed'ed(e d)e d&ed*ed+e d,ed-eeeeed.ZIejeKZLGd/d0ZMGd1d2ZNd3eOfd4ZPd5eOfd6ZQd7ZRy)8N) AutoConfig)logging) AqlmConfigAutoRoundConfig AwqConfigBitNetQuantConfigBitsAndBytesConfigCompressedTensorsConfig EetqConfigFbgemmFp8ConfigFineGrainedFP8Config FPQuantConfig GPTQConfig HiggsConfig HqqConfig Mxfp4ConfigQuantizationConfigMixinQuantizationMethod QuantoConfig QuarkConfig SinqConfig SpQRConfig TorchAoConfig VptqConfig) HfQuantizer)AqlmHfQuantizer)AutoRoundQuantizer) AwqQuantizer)BitNetHfQuantizer)Bnb4BitHfQuantizer)Bnb8BitHfQuantizer)CompressedTensorsHfQuantizer)EetqHfQuantizer)FbgemmFp8HfQuantizer)FineGrainedFP8HfQuantizer)FPQuantHfQuantizer)GptqHfQuantizer)HiggsHfQuantizer)HqqHfQuantizer)Mxfp4HfQuantizer)QuantoHfQuantizer)QuarkHfQuantizer)SinqHfQuantizer)SpQRHfQuantizer)TorchAoHfQuantizer)VptqHfQuantizerawqbitsandbytes_4bitbitsandbytes_8bitgptqaqlmquantoquarkfp_quanteetqhiggshqqzcompressed-tensors fbgemm_fp8torchaobitnetvptqspqr)fp8z auto-roundmxfp4sinqc6eZdZdZedefdZedZy)AutoQuantizationConfigz The Auto-HF quantization config class that takes care of automatically dispatching to the correct quantization config given a quantization config stored in a dictionary. quantization_config_dictc v|jd}|jdds|jddr*|jddrdnd}tj|z}n | td|tvr,td|d t t jt|}|j|S) N quant_method load_in_8bitF load_in_4bit_4bit_8bitThe model's quantization config from the arguments has no `quant_method` attribute. Make sure that the model has been correctly quantizedUnknown quantization type, got  - supported types are: ) getrBITS_AND_BYTES ValueError AUTO_QUANTIZATION_CONFIG_MAPPINGlistAUTO_QUANTIZER_MAPPINGkeys from_dict)clsrHrJsuffix target_clss X/mnt/e/genesis-system/.venv/lib/python3.12/site-packages/transformers/quantizers/auto.pyrYz AutoQuantizationConfig.from_dict|s/33NC # ' ' >BZB^B^_motBu 8 < <^U SWY`F-<>>( # #r^rGcreZdZdZedeezfdZedZedeezdedzfdZ e dZ y) AutoHfQuantizerz The Auto-HF quantizer class that takes care of automatically instantiating to the correct `HfQuantizer` given the `QuantizationConfig`. r`c Dt|trtj|}|j}|t j k(r|jr|dz }n|dz }|tvr,td|dttjt|}||fi|S)NrNrMrPrQ) isinstancermrGrYrJrrSrKrWrTrVrX)rZr`rfrJr\s r] from_configzAutoHfQuantizer.from_configs )4 0"8"B"BCV"W *77  -<< <"//' ' 5 51,@/44678:  ,L9 -888r^c Ptj|fi|}|j|S)N)rGrbrr)rZrerfr`s r]rbzAutoHfQuantizer.from_pretraineds*4DDEbmflm233r^quantization_config_from_argsNc |d}nd}t|tr;t|trtj|}ntj|}|g|j j |j j k7r:td|j j d|j j dt|tttttttfrW|U|j}|jD]\}}t!||||dt#|j%dz }|dk7r-t|ttfst'j(||St*j-||S)z handles situations where both quantization_config from args and quantization_config from model config are present. zYou passed `quantization_config` or equivalent parameters to `from_pretrained` but the model you're loading already has a `quantization_config` attribute. The `quantization_config` from the model will be used.zThe model is quantized with z but you are passing a z| config. Please make sure to pass the same quantization config class to `from_pretrained` with different loading attributes.z"However, loading attributes (e.g. z]) will be overwritten with the one you passed to `from_pretrained`. The rest will be ignored.)rqrmrrYrG __class__rhrTrrr r rrget_loading_attributesitemssetattrrVrXwarningswarnloggerinfo)rZr`rt warning_msgloading_attr_dictattrvals r]merge_quantization_configsz*AutoHfQuantizer.merge_quantization_configss ) 4y  K )4 07I&5&?&?@S&T#&<&F&FGZ&[# * 5#--66:W:a:a:j:jj./B/L/L/U/U.VVmoLoVoVo_o_n``FF  ###+( .9 = T T V .446 8 c+T37 8 ?EVE[E[E]@^?_`}~ ~K " Z0CkSgEh%i MM+ &#" KK $""r^c ^|jdd}|jdds|jddr*|jddrdnd}tj|z}n | td|tvr8t j d|d ttjd yy ) NrJrKFrLrMrNrOrPrQz~. Hence, we will skip the quantization. To remove the warning, you can delete the quantization_config attribute in config.jsonT) rRrrSrTrUr}warningrVrWrX)rHrJr[s r]supports_quant_methodz%AutoHfQuantizer.supports_quant_methods/33NDI # ' ' >BZB^B^_motBu 8 < <^U SWY`F-<.register_config_fnsH 5 5xx/CDE E#67HI I36(0 r^ra)rrs` r]register_quantization_configrs r^namecfd}|S)zRegister a custom quantizer.c~tvrtddt|ts t d|t<|S)Nz Quantizer 'rz!Quantizer must extend HfQuantizer)rWrTrrr)rZrs r]register_quantizer_fnz1register_quantizer..register_quantizer_fn(sG ) ){4&0DEF F#{+?@ @'*t$ r^ra)rrs` r]register_quantizerr%s ! r^c"t|d}|r!tj|jsd}|s|Q|r&tj |j||_n||_tj |j|}nd}||j |||j|}|j|}|j|}t|jdds&|jj}t|d||d<|||fS)Nr`F) pre_quantized) device_map weights_only dequantizevaluequant) hasattrrorr`rrrvalidate_environmentupdate_device_mapupdate_tp_planupdate_ep_planrcrJ)configr`rr user_agentr hf_quantizerrJs r]get_hf_quantizerr5sF$9:M_BB6C]C]^ +7 )8)S)S**,?*F &*=F &&22  & &'3  ))!% * "33J? ,,V4,,V4|77uM';;HHL"),"NJw   ++r^)Sr{models.auto.configuration_autorutilsrutils.quantization_configrrrr r r r r rrrrrrrrrrrrrrbaserquantizer_aqlmrquantizer_auto_roundr quantizer_awqr quantizer_bitnetr!quantizer_bnb_4bitr"quantizer_bnb_8bitr#quantizer_compressed_tensorsr$quantizer_eetqr%quantizer_fbgemm_fp8r&quantizer_finegrained_fp8r'quantizer_fp_quantr(quantizer_gptqr)quantizer_higgsr* quantizer_hqqr+quantizer_mxfp4r,quantizer_quantor-quantizer_quarkr.quantizer_sinqr/quantizer_spqrr0quantizer_torchaor1quantizer_vptqr2rWrU get_loggerrhr}rGrostrrrrrar^r]rs70+4'/22F+6@2+-)-/-++1+ <++ O   O     " O  >6&!   O!" O#$ %$  +0$ 9$+$+$ J $  J $  J $ l$ [$ $ 9$1$/$ [$}$ $  J!$" J#$$ !  +$ 0   H %&#&#Rssl   !S ! !,r^