+ TV-j~^RIt^RIt^RIt^RIt^RIt^RIt^RIt^RIt^RIH t ^RI H t ^RI H t HtHtHtHtHtHtHt^RIHt^RIHt]P4!RR4P74R8Xd ^RIHtM^RIHt]P@!]PBRL4^R I"H#t#H$t$H%t%H&t&^R I'H(t(^R I'H)t*R RRRRRRRRRRRRRRRRR/ t+^t,Rt-RRlt.R R!lt/R"R#lt0R$t1R%t2RMR&R'llt3R(t4R)R*lt5R+R,R]03R-R.llt6R/R0lt7RMR1lt8RNR2R3llt)ROR4R/R5R6lllt9RPR7lt:],3R8R9llt;R:R;ltR+/R?R@llt>RQRARBllt?RCRDlt@RERFltARRRGRHlltBRItCRJRKltDR# ]d ]!R 4hi;i)SN)Path)dedent)AnyCallableDictListOptionalTupleTypeUnionMLXLM_USE_MODELSCOPEFalsetrue)snapshot_downloadz/Run `pip install modelscope` to use ModelScope.) tree_flattentree_map tree_reducetree_unflatten)TokenizerWrapper)loadmistralllamallavamistral3zphi-msftphixtral falcon_mambamambajoyai_llm_flash deepseek_v3kimi_k2 qwen2_5_vlqwen2_vl minimax_m2minimax iquestcoderc  RRRRRRRRR^/p^pVF+pVP4'g VR8XgM V^, pK- \VRV4pWRP4P4p\ WAV,,4#) Mg.AGgeAMBGB.N)isdigitfloatstripupperint)xsizessplitxidigitssizes& ]/Volumes/fast/ai/experiments/ui-tars-smoke/.venv/lib/python3.14/site-packages/mlx_lm/utils.py _parse_sizer9<s #sCsD#r1 =E E c   1Ve9 F fI    $ $ &D vd # $$cXV^8dQhR\PR\P/#)qweightreturnmxarray)formats"r8 __annotate__rCHs"777bhh7r:c^p^ V,pVPwr4WB,p^V,^, p\P!.RO4V,pVR,V, V,pVPW54#))rrEr<).N)shaper@rAreshape) r=bits pack_factor out_features packed_in in_featuresmaskshiftsunpackeds & r8_unpack_awq_weightsrUHsj D*K%mmL)K I?D XX. /$ 6F "f,4H   L 66r:c V^8dQhR\\\P3,R\\\3,R\ \\\P3,\\\3,3,/#)r<weightsquantization_configr>)rstrr@rArr )rBs"r8rCrCSs_Y)Y) #rxx- Y)c3hY) 4RXX S#X ./Y)r:cvaVPR^4pV^8wd\RV: R24hVPR^4p/p\VP44EFooSP R4'd\RS R24hSP R4'EdSRRpW R2,pV R 2pV R 2pW,p ^ V,p VP wrW,p W,p\ V4pVPpW,pVPV VV 4p\P!V 4V,pVP\P4V,PRR 7P\P4p\P!V P4p W9dHW,p\ V4pVPpVP\P4)V ,pMI^V^, ,p\P !V P V)\PR 7V ,pVWE R 2&WV R 2&VPV P"4WE R2&V P"pEK \$;QJdV3RlR4F 'gK RM RM!V3RlR44'dEKcVS,VS&EKr VP'4FQwpp\P(!VP"\P*4'gK=VPX4VV&KS RVRV/pVV3#)rMz Only bits=z& is supported for AutoAWQ/GPTQ models. group_sizez.g_idxzFound z in weights. Models with non-contiguous group indices (g_idx) are not currently supported. Please use a model without g_idx or re-quantize the model using mlx_lm.convert..qweightN.scales.qzeros)axis)dtypez.weightz.biasesc3F<"TFpSPV4xK R#5iN)endswith).0suffixkeys& r8 )_transform_awq_weights..s! /QVCLL /Qs!TFi)r\r^r])get ValueErrorlistkeysrcrKrUTrLr@arangeastypeuint32sum contiguousfloat32fullr`anyitems issubdtypefloating)rWrXrMr[ new_weightsprefixr= scales_key qzeros_keyscalesrNrQ packed_outrOn_groupsunpacked_weightrPrepackedrSweightqzerosunpacked_zerosbiases zero_point model_dtypekwmlx_quantizationrfs&& @r8_transform_awq_weightsrSs  " "61 -D qy;'MNOO$((s;JKGLLN# << ! !AA  << # #"XF12G"87+J"87+J(F *K&-mm #K%3L"0H2':O-//O$2I&..|Y THYY{+d2F+v5:::CJJ299U ]]688,F$ ,"5V!<!/!1!1 )// ;;fD4!8_  {"**MPVV.4K('* +.46('* +.4mmFLL.IK('* + ,,K /Q  /Q    's|K G$J!!#1 =="++ . .XXk2KN$ j ( ((r:c$V^8dQhR\/#)r<config)dict)rBs"r8rCrCs&&&r:cVR,p\PW4p\P!RV 24pTP TP3# \dRT R2p\ T4hi;i)z Retrieve the model and model args classes based on the configuration. Args: config (dict): The model configuration. Returns: A tuple containing the Model class and the ModelArgs class. model_typezmlx_lm.models.z Model type z not supported.)MODEL_REMAPPINGrj importlib import_module ImportErrorrkModel ModelArgs)rrarchmsgs& r8 _get_classesrsx %J $$Z&get_total_parameters..s 1bii0Hr:is_leafc4\VR4'd_\VR4'g^MVPPpWPP^ ,VP,,#\ R\ VP4444#)rMbiasc3>"TFwrVPxK R#5irb)r7)rd_vs& r8rg8get_total_parameters..nparams..sC&Bda166&Bs)hasattrrr7rrMrrr parameters)rns& r8nparams%get_total_parameters..nparamsse 1f   F++Axx}}r)QVV33 3Cl1<<>&BCCCr:c38<"TFwrS!V4xK R#5irb)rdrrrs& r8rg'get_total_parameters..s3ldawqzzls)r leaf_modulesrr)modelrrs& @r8get_total_parametersrs5 &HLD 3l3 33r:cT\RV^4p\V4pV^,V, #)cj\V\P4'dWP,#T#rb)rr@rAnbytes)accr2s&&r8r)compute_bits_per_weight..s"Arxx)@)@sXX~IcIr:)rr)r model_bytes model_paramss& r8compute_bits_per_weightrs/I5RSK(.L ?\ ))r:ctV^8dQhR\R\\,R\\,R\/#)r<path_or_hf_reporevisionallow_patternsr>)rYr rr)rBs"r8rCrCs8&&&sm&I& &r:c\V4pVP4'g&T;'g.ROp\\VVVR74pV#)a` Ensures the model is available locally. If the path does not exist locally, it is downloaded from the Hugging Face Hub. Args: path_or_hf_repo (str): The local path or Hugging Face repository ID of the model. revision (str, optional): A revision id which can be a branch name, a tag, or a commit hash. Returns: Path: The local file path. )rr) *.jsonmodel*.safetensors*.pytokenizer.model *.tiktokentiktoken.model*.txt*.jsonl*.jinja)rexistsr)rrr model_paths&&& r8 _downloadrsV o&J     '   ,  !-   r:c.\\VRR74#)T)local_files_only)rr)hf_repos&r8hf_repo_to_pathrs !'DA BBr:c0V^8dQhR\R\/#)r<rr>)rr)rBs"r8rCrCsDTr:c\VR, R4;_uu_4p\P!V4pRRR4VR, pVP4'dV/p\VR4;_uu_4p\P!V4pRRR4TP RR4;p'dTXR&X# +'giL;i +'giLD;i \PdL^i;i) config.jsonrNgeneration_config.json eos_token_idF)openjsonrrJSONDecodeErrorrj)rfrgeneration_config_filegeneration_configrs& r8 load_configrs j=(# . .!1 /(*BB$$&& ,c22a$(IIaL!3 -00G G< G%1F> " M / .32##   s;B1C1CC1 C  C CCC/.C/FTc>V^8dQhR\R\R\R\\\\ 3,,R\ \.\\\P,\3,3,R\\P\3,/#)r<rlazystrict model_configget_model_classesr>) rboolr rrYrrrr r rr)rBs"r8rCrCsJJJ J J4S>* J  d299ot.C(D DE J  299d? Jr:c^aaa\V4oVeSPV4\P!\VR, 44pV'gV'd\ RV 24h/oVF(pSP\ P !V44K* SPR4;pey\PPRW, 4p\PPV4p VPPV 4V PV PrM V!SR7wrRS9d&SPR/4p RV 9d V R,SR&V P!S4p V !V 4o\#SR 4'dSP%S4oVVV3R lpSPR R4;pe V!V4MSPRR 4;p'dVR ,pVR8Xd^RIHpV!SV4oMlVR8XdR^ R^RR/pVSR &VSR&V!V4MJVR8XdR^ R^RR/pVSR &VSR&V!V4M(VR9d"\+SV4wopVSR &VSR&V!V4SPRR 4'dIRp\-VSP/4\0P2P4R7pSP7V4SP94SP;\=SP?44VR7V'g%\ P8!SPA44SS3#)a Load and initialize the model from a given path. Args: model_path (Path): The path to load the model from. lazy (bool): If False eval the model parameters to make sure they are loaded in memory before returning, otherwise they will be loaded when needed. Default: ``False`` strict (bool): Whether or not to raise an exception if weights don't match. Default: ``True`` model_config (dict, optional): Optional configuration parameters for the model. Defaults to an empty dictionary. get_model_classes (Callable[[dict], Tuple[Type[nn.Module], Type]], optional): A function that returns the model class and model args class given a config. Defaults to the ``_get_classes`` function. Returns: Tuple[nn.Module, dict[str, Any]]: The loaded and initialized model and config. Raises: FileNotFoundError: If the weight files (.safetensors) are not found. ValueError: If the model class or args class are not found or cannot be instantiated. NrzNo safetensors found in model_file custom_model)rrX text_configsanitizec <VV3Rlp\P!SVR,VR,VPRR4VR7R#)cx<VSR,9dSR,V,#\VR4'gR#V R2S9#) quantization to_quantizedFr])r)prrrWs&&r8class_predicate6load_model.._quantize..class_predicate]sAF>**n-a001n--S=G+ +r:r[rMmodeaffine)r[rMrrN)rquantizerj)rrrrrWs& r8 _quantizeload_model.._quantize\s< , #L1f%!!&(3+  r:rF quant_methodbitnet)bitnet_quantizemxfp4r[rMrzcompressed-tensorsrquantize_activationsc\V\P4'dVPR9d \ R4hVP RR4'd \ R4hVP PwrV^ VP,,p\P!W!VPVPVP4#V#)nvfp4z8Mode ({m.mode}) does not support activation quantizationrFz?Linear layer with bias does not support activation quantization)rmxfp8) rrQuantizedLinearrrkrjrrKrMQQLinearr[)rout_dimsin_dimss& r8 _maybe_qqload_model.._maybe_qqs!R//0066!33$R55''$Y%&HHNN!2<'{{7allAFFAFFSSr:r)r)awqgptq)!rupdateglobrYFileNotFoundErrorr@rrjrutilspec_from_file_locationmodule_from_specloader exec_modulerr from_dictrrmodels.bitlinear_layersrrrrrr is_moduleupdate_moduleseval load_weightsrlrwr)rrrrr weight_fileswfrspecr model_classmodel_args_classr model_argsrrrXrrrleavesrrrWs&&&&& @@@r8 load_modelrs< $F l#99S.B!BCDL F"::, GHHGrwwr{#jj.. ;~~55   # ~~..t4 %(, DNN%(9(H% F*jj3 K /,78M,NF( )!++F3J  #Euj!!..) " >488 E, & +@% H H  H*>: 8 # @#E+>?E W $("faIL%1F> ",8F( ) l # 1 1("faJL%1F> ",8F( ) l # _ ,$:7DW$X !G\%1F> ",8F( ) l # zz(%00  )U%7%7%9299CVCVW V$ JJL tGMMO,V<    "# &=r:cdV^8dQhR\PR\R\P/#)r<r adapter_pathr>)rrrY)rBs"r8rCrCs)///#/"))/r:c^RIHpV!W4#)rF) load_adapters) tuner.utilsr )rr_load_adapterss&& r8r r s< % ..r:c<\V.ROR7p\VVVR7#)zXLoad a huggingface tokenizer and try to infer the type of streaming detokenizer to use. r eos_token_idsrrrrrrrr)r_load_tokenizer)rtokenizer_config_extrar&s&&&r8load_tokenizerr*s.   J # r:cV^8dQhR\R\\\\3,,R\\\\3,,R\\,R\R\R\\,R\ \ \P\3,\ \P\\\\3,3,3,/#) r<rtokenizer_configrrr return_configrr>) rYr rrrr r rrr)rBs"r8rCrCs1 1 1 tCH~.1 4S>*1 3- 1  1  1 sm1   "))% %& "))%tCH~ 5681 r:c\WR7p\WtVR7wrVe\W4pVP4\ WqV P RR4R7p V'dWV 3#W3#)a Load the model and tokenizer from a given path or a huggingface repository. Args: path_or_hf_repo (Path): The path or the huggingface repository to load the model from. tokenizer_config (dict, optional): Configuration parameters specifically for the tokenizer. Defaults to an empty dictionary. model_config(dict, optional): Configuration parameters specifically for the model. Defaults to an empty dictionary. adapter_path (str, optional): Path to the LoRA adapters. If provided, applies LoRA layers to the model. Default: ``None``. lazy (bool): If ``False`` eval the model parameters to make sure they are loaded in memory before returning, otherwise they will be loaded when needed. Default: ``False`` return_config (bool: If ``True`` return the model config as the last item.. revision (str, optional): A revision id which can be a branch name, a tag, or a commit hash. Returns: Union[Tuple[nn.Module, TokenizerWrapper], Tuple[nn.Module, TokenizerWrapper, Dict[str, Any]]]: A tuple containing the loaded model, tokenizer and, if requested, the model config. Raises: FileNotFoundError: If config file or safetensors are not found. ValueError: If model class or args class are not found. )r)rNrr%)rrr rr*rj) rr,rrrr-rrrr tokenizers &&&&&&& r8rrsiH?>JzlKMEe2 FJJ~t4TI''r:r,c V^8dQhR\\PP,R\\PP,R\R\\ \ \3,,/#)r<pipeline_group tensor_groupr-r,)r r@ distributedGrouprrrYr)rBs"r8rCrCscV V R^^112V 2>>//0V  V tCH~. V r:c\V.ROR7p\VRRR7wrg\VR4;'d\VPR4p\VR4p VeV'g \ R 4hVeV 'g \ R 4hV'gV 'g \ R 4hYu;JdfQMMMV 'd \ P P4pM&V'd\ P P4pVeVPPV4\VR , R 4;_uu_4p \P!V 4R,p RRR4\4p \VP44FCwrX PV R4RJ;p'd \ R4hV P!W,4KE \W R7M \V4\#TT;'gRR/VPRR4R7p\VRRR7wrnVeVP%V4VeVPPV4\ P&!VP44\ P&!\ P P)\ P*!R4\ P,R74V'dVVV3#VV3# +'giEL;i)rr$TF)rrrpipelineshardNzGThe model does not support pipelining but a pipeline_group was providedzMThe model does not support tensor parallelism but a tensor_group was providedz'The model does not support any shardingmodel.safetensors.index.jsonr weight_mapz>..0L ^^002N! ^,*==s C Cs99S>,7LDe  !1!1!34DA(,,Q5==y= R OOLO , 5 $3$770$7jj6I *4>HE L!! ^,GGE   GGBNN " "288C= " @Ai''iED C Cs 3K K& cV\V\PP4RV4#rb)rHr@r3r<)rAr-s&&r8 pipeline_loadrJRs bnn113T= IIr:c<V^8dQhR\R\R\/#)r<rWmax_file_size_gbr>)rr1rl)rBs"r8rCrCVs!Dr:cV^,p.p/^rTVP4FFwrgWWP,V8dVPV4/^rTWtV&WWP, pKH VPV4V#)z Splits the weights into smaller shards. Args: weights (dict): Model weights. max_file_size_gb (int): Maximum size of each shard in gigabytes. Returns: list: List of weight shards. )rwrappend)rWrLmax_file_size_bytesshardsr7 shard_sizerrs&& r8 make_shardsrRVsv+b0 FA:   #6 6 MM% "A:ahh   MM% Mr:cvV^8dQhR\\\3,R\\\R3,/#)r<pathhf_pathN)r rYr)rBs"r8rCrCns.//E#t),/uS$_7M/r:c2^RIHpHpVfVPV!RR74pMVP V4pRVP nRVP nVP PfR.VP nM?RVP P9d%VP ;PR., unVe\V4VP n RVn VP\PPVR44R#) z Uploads the model to Hugging Face hub. Args: path (Union[str, Path]): Local path to the model. hf_path (Union[str, Path, None]): Path to the original Hugging Face model. ) ModelCard ModelCardDataNen)languagemlxztext-generationr+ README.md)huggingface_hubrWrX from_templaterdata library_name pipeline_tagtagsrY base_modeltextsaveosrTjoin)rTrUrWrXcards&& r8create_model_cardrins9&&}d'CD~~g&"DII.DII yy~~  diinn $ 5'!"7| DIIIbggll4-.r:c0V^8dQhR\R\/#)r<rT upload_repo)rY)rBs"r8rCrCs?Y?Y?Y#?Yr:c ^RIHpHpHp^RIHpVP 4\V4R, pVPV4pVPPpVeRV RV RV RV RV R 2 p MR p \R V R V R V R24Vn VPV4V!4p V PVRR7V PVVRR7\!RV R24R#)z Uploads the model to Hugging Face hub. Args: path (str): Local path to the model. upload_repo (str): Name of the HF repo to upload to. )HfApirWlogging) __version__r\Nz This model [z](https://huggingface.co/z,) was converted to MLX format from [z!) using mlx-lm version **z **. r+z # z z ## Use with mlx ```bash pip install mlx-lm ``` ```python from mlx_lm import load, generate model, tokenizer = load("av") prompt = "hello" if tokenizer.chat_template is not None: messages = [{"role": "user", "content": prompt}] prompt = tokenizer.apply_chat_template( messages, add_generation_prompt=True, return_dict=False, ) response = generate(model, tokenizer, prompt=prompt, verbose=True) ``` T)repo_idexist_okr) folder_pathrp repo_typez0Upload successful, go to https://huggingface.co/z for details.)r]rmrWrnr+roset_verbosity_inforrr_rcrrdre create_repoupload_large_folderprint) rTrkrmrWrnro card_pathrhrU provenanceapis && r8 upload_to_hubr{s:9  T [(I >>) $Dii""G M!:;-H''.i/H R +}-    -  "#. /  DI6 IIi 'COOK$O7  <[M WXr: donate_modelcvV^8dQhR\\\3,R\PR\ RR/#)r< save_pathrr|r>N)r rYrrrr)rBs"r8rCrCs=9 9 S$Y9 999  9  9 r:c\V\4'd \V4pVPRRR7\ \ VP 444p\V4p\V4pV^8dRMRp\RVP444pRRVR\V4/R //pV'd+VP\R VP 444VP4?\\V44Fwp WI,p R WI&VP!V ^,V4p W , p \"P$!\V 4V R R /R7V P'4Fp WR ,V &K ? Ky \)VR ,4Uu/uFqVR ,V,bK upVR &\+VR, R4;_uu_4p\,P.!VV^R7R R R 4R #uupi +'giR #;i)z?Save model weights and metadata index into specified directory.T)parentsrqz"model-{:05d}-of-{:05d}.safetensorszmodel.safetensorsc38"TFqPxK R#5irb)r)rdrs& r8rgsave_model..s8'7!XX'7smetadata total_sizetotal_parametersr9c.\P!.4#rbr?)rs&r8rsave_model..s  r:NrBr[)rr8rindent)rrYrmkdirrrrrRlenrrvaluesrrrclearrangerBr@save_safetensorsrmsortedrrdump)r~rr|rWrP shards_countshard_file_formatr index_datair7 shard_name shard_path weight_namerrs&&$ r8 save_modelrs)S!!O  OOD4O0< 0 0 234G  !Fv;L !  -  8w~~'788J *  4U ;  b J X4e6F6F6HIJ MMO 3v;   &--a!e\B +  C OUh=NO ::| $[ 1(  17z,7O0P 0P1:l #A &&0P J| i88# > >!    ? >   ? > >sG/ G44 H c\V^8dQhR\PR\R\\,R\\,R\ R\\ \ \P.\\\3,3,,R\\P\3,/#)r<rrr[rMrquant_predicater>) rrrr r1rYrr rr )rBs"r8rCrCsL#L# 99L# L# L# 3- L#  L# hRYY'7tTz9J'JKL L# 299d?L#r:cXaaa a a Rp\P!V4o S;'g\VRR4oV!VSV4wopRSRVRV/o RS 9dRo MR o S S R&V VV VV 3R lp\P!VSVVVR 7S R,S R &\ V4p\ R VR R24VS 3#)a Applies quantization to the model weights. Args: model (nn.Module): The model to be quantized. config (dict): Model configuration. group_size (Optional[int]): Group size for quantization. bits (Optional[int]): Bits per weight for quantization. mode (str): The quantization mode. quant_predicate (Callable): A callable that decides how to quantize each layer based on the path. Accepts the layer `path` and the `module`. Returns either a bool to signify quantize/no quantize or a dict of quantization parameters to pass to `to_quantized`. Returns: Tuple: Tuple containing quantized model and config. cZRRRRRRRR/pW0,wrET;'gTT;'gT3#)rrrr)@rE) rE)rE)rr)rr[rM mode_defaultsdefault_group_size default_bitss&&& r8defaults_for_mode)quantize_model..defaults_for_mode sG g W W W  ,9+>(///1E1EEEr:rNr[rMrrTFc"<\VR4'gR#VPPR,S,^8wdR#RpSe S!W4p\V\4'dVSR,V&V#S'dV'd SSR,V&V#)rFTrri)rrrKrr)rTmodulebool_or_paramsfine_grained_configr[ quant_paramsrquantized_configs&& r8wrapped_predicate)quantize_model..wrapped_predicate7sv~.. ==  r "Z /1 4  &,T:N nd + +5C ^ ,T 2!^5A ^ ,T 2r:)rrrXz[INFO] Quantized model with z.3fz bits per weight.)copydeepcopygetattrrrrrw) rrr[rMrrrrbpwrrrs &&f&&f @@@r8quantize_modelrs4F}}V,%PP8I4)PO(z4@J *fdFDIL))##+7(  KK  ) /?~.N*+ !% (C (S 1B CD " ""r:cXV^8dQhR\PR\P/#r<rr>)rr)rBs"r8rCrCUs"++BII+"))+r:c ^RIHpHp.pVP4EF3wrERV9p\ V\ P 4'd\ PpRV/pMM\ V\ P4'd/p\ PpM\ WQ4'dRV/pTpMK\P!VPVPVPVPVP VP"4p V P$RRR1,p V!V /VBp V'dVP&V nWn VP)WK34EK6 \+V4^8dVP-\/V44V#)z Dequantize the quantized layers in the model. Args: model (nn.Module): The model with quantized layers. Returns: nn.Module: The model with dequantized layers. )QuantizedSwitchLinear SwitchLinearrNri)models.switch_layersrr named_modulesrrrLinearQuantizedEmbedding Embeddingr@ dequantizerr~rr[rMrrKrrNrrr) rrrdequantize_layersnamerrclskwargsrargsrs & r8dequantize_modelrUs8J++-  fb00 1 1))Cd^F  5 5 6 6F,,C  6 6d^FC  MM MM MM    KK KK  ||DbD!    [[AF  $+5.8 ! ^,=>? Lr:cVV^8dQhR\R\\\3,RR/#)r<r config_pathr>N)rr rYr)rBs"r8rCrCs/)) )sDy!) )r:cJVPRR4VPRR4RV9d VR,VR&\\VP444p\ VR4;_uu_4p\ P !W^R7RRR4R# +'giR#;i)zSave the model configuration to the ``config_path``. The final configuration will be sorted before saving for better readability. Args: config (dict): The model configuration. config_path (Union[str, Path]): Model configuration file path. _name_or_pathN vision_configrrXrr)poprrrwrrr)rrrDs&& r8 save_configrs} JJ% JJ%(.~(>$%&( )F k3  3 &a(   s .B B" c V^8dQhR\\\3,R\\\3,R\PR\ R\ \\3,R\/#)r<dst_pathsrc_path_or_reporr/rr|) r rYrrrrrrr)rBs"r8rCrCsb))CI)CI&) 99) ) cN )  )r:c\V4pVP4'gTp\V4pMRp\V4p\WRR7\ W@R, R7VP V4RFEp\ P !\Wh, 44Fp \P!W4K KG \W4R#)NT)r|r)r)rr) rrrrrsave_pretrainedrrYshutilrri) rrrr/rr|src_pathrrfiles &&&&&& r8reres$%H ??  ""7+H~HxT2}$<= h' /IIc(,/0D KK '10h(r:c\\V4\V44p\V4FpW,W,8wgKVu# V#)a Calculates the length of the common prefix of two lists. Args: list1: The first list of strings. list2: The second list of strings. Returns: The length of the common prefix. Returns 0 if lists are empty or do not match at the first element. )minrr)list1list2min_lenrs&& r8common_prefix_lenrs@#e*c%j)G7^ 8ux H Nr:cDV^8dQhR\PR\/#r)rrr)rBs"r8rCrCs  ryy T r:c\P!VP4pRVP9# \\ 3dR#i;i)z Check if the model supports input_embeddings in its call signature. Args: model (nn.Module): The model to check. Returns: bool: True if the model supports input_embeddings, False otherwise. input_embeddingsF)inspect signature__call__rrk TypeError)rrs& r8#does_model_support_input_embeddingsrsC%%enn5 !Y%9%999  "s.1AA)ii)NN)NNNFFN)NNF)F)rN)T)Errrrrrfresourcerpathlibrtextwraprtypingrrrrr r r r mlx.corecorer@mlx.nnrgetenvlower modelscoperrr] setrlimit RLIMIT_NOFILE mlx.utilsrrrrtokenizer_utilsrrr(rMAX_FILE_SIZE_GBr9rUrrrrrrrrr r*rHrJrRrir{rrrrrerrrr:r8rs      99 #W-335?M02 8))<8II.4w Z G} }*)7  %7Y)x&* 4*&RC*-1HT JZ/ 01 hV 26 V V rJ8H0/4?YD9  9 xL#^+\)4)84 k MKLLMs 4E..E>