biM6ddlZddlmZddlmZddlmZmZmZm 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.ZIejJeKZLGd/d0ZMGd1d2ZNd3eOfd4ZPd5eOfd6ZQd7ZRdS)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-roundmxfp4sinqcDeZdZdZedefdZedZdS)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 |d}|dds|ddr*|ddrdnd}tj|z}n|td|tvr9td|d t t t|}||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 W/root/projects/butler/venv/lib/python3.11/site-packages/transformers/quantizers/auto.pyrYz AutoQuantizationConfig.from_dict|s/33NCC # ' ' > > BZB^B^_motBuBu  8 < <^U S S`WWY`F->>>[>( # #[ # # #r^rGceZdZdZedeezfdZedZedeezdedzfdZ e dZ dS) AutoHfQuantizerz The Auto-HF quantizer class that takes care of automatically instantiating to the correct `HfQuantizer` given the `QuantizationConfig`. r`c bt|trt|}|j}|t jkr|jr|dz }n|dz }|tvr9td|dtt t|}||fi|S)NrNrMrPrQ) isinstancermrGrYrJrrSrKrWrTrVrX)rZr`rfrJr\s r] from_configzAutoHfQuantizer.from_configs )4 0 0 X"8"B"BCV"W"W *7  -< < <"/ (' ' 5 5 5:,::/446677::  ,L9 z-88888r^c Ptj|fi|}||S)N)rGrbrr)rZrerfr`s r]rbzAutoHfQuantizer.from_pretraineds04DEbmmflmm2333r^quantization_config_from_argsNc :|d}nd}t|trDt|trtj|}nt|}|D|jj|jjkr*td|jjd|jjdt|tttttttfri|g|}|D]\}}t!||||dt#|dz }|dkr1t|ttfst'j|nt*||S) z handles situations where both quantization_config from args and quantization_config from model config are present. NzYou 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 K )4 0 0 \7II \&5&?@S&T&T##&<&F&FGZ&[&[# * 5#-6:W:a:jjjF/B/L/UFFoLoVo_FFF  ###+(   ~.9 = T T V V .4466 8 8 c+T37777 ~EVE[E[E]E]@^@^~~~ ~K "  Z0CkSgEh%i%i  M+ & & & & KK $ $ $""r^c |dd}|dds|ddr*|ddrdnd}tj|z}n|td|tvrGt d|d tt d dSd S) NrJrKFrLrMrNrOrPrQz~. Hence, we will skip the quantization. To remove the warning, you can delete the quantization_config attribute in config.jsonT) rRrrSrTrUr}warningrVrWrX)rHrJr[s r]supports_quant_methodz%AutoHfQuantizer.supports_quant_methods/33NDII # ' ' > > BZB^B^_motBuBu  8 < <^U S S`WWY`F-.register_config_fns\ 5 5 5DDDDEE E#677 JHII I36(0 r^ra)rrs` r]register_quantization_configrs$ r^namecfd}|S)zRegister a custom quantizer.ctvrtddt|tst d|t<|S)Nz Quantizer 'rz!Quantizer must extend HfQuantizer)rWrTrrr)rZrs r]register_quantizer_fnz1register_quantizer..register_quantizer_fn(s[ ) ) )E4EEEFF F#{++ A?@@ @'*t$ r^ra)rrs` r]register_quantizerr%s$ ! r^c<t|d}|r!t|jsd}|s|Q|r&t|j||_n||_t|j|}nd}||||||}||}| |}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_quantizerr5sSF$9::M_BB6C]^^  +7  =)8)S)S*,?**F & &*=F &&22  &'3   ))!% *   "33J?? ,,V44,,V44|7uMM O';HL"),"N"NJw   ++r^)Sr{models.auto.configuration_autorutilsrutils.quantization_configrrrr r r r r rrrrrrrrrrrrrrbaserquantizer_aqlmrquantizer_auto_roundr quantizer_awqr quantizer_bitnetr!quantizer_bnb_4bitr"quantizer_bnb_8bitr#quantizer_compressed_tensorsr$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]rsW7777770++++++444444''''''//////222222222222FFFFFF++++++666666@@@@@@222222++++++------))))))------//////------++++++++++++111111++++++ <++ O   O     " O  >6&!   O!" O#$ %$  +0$ 9$+$+$ J $  J $  J $ l$ [$ $ 9$1$/$ [$}$ $  J!$" J#$$ !  +$$$ 0  H % %&#&#&#&#&#&#&#&#Rssssssssl      !S ! ! ! ! !,!,!,!,!,r^