9i1VSSKrSSKrSSKrSSKrSSKJrJrJr SSKrSSK J r SSK r SSK Jr SSKJr SSKJr SSKJr SSKJr SSKJrJr SS KJr SS KJr SS KJr SS KJrJ r SS K!J"r" SSK#J$r$J%r% SSK&J'r' SSK(J)r) SSK*J+r+ SSK,J-r-J.r.J/r/ SSK0J1r1 SSK2J3r3J4r4J5r5 SSK6J7r7J8r8J9r9 \)"5r:\Rv"\RxS9"SS\55r=g)N)CallableDictOptional) SummaryWriter)nn)optim) DataLoader)Trainers)Model TorchModel)kws_nearfield_dataset) BaseTrainer)TRAINERS)load_checkpointsave_checkpoint)Config)DEFAULT_MODEL_REVISION ModelFile) create_device) get_logger)set_random_seed) executor_cv executor_testexecutor_train) compute_det)query_token_set read_lexicon read_token) average_modelconvert_to_kaldicount_parameters) module_namec ^\rSrSrSS\4S\S\S\\S\\S\\4 U4SjjjrS r S \S \ \\ 44S jr S \ R4S jrSSjrSSjrSrU=r$)KWSNearfieldTrainer%Nmodelwork_dircfg_file arg_parse_fnmodel_revisionc X>[U[5(aRURX5UlUc8[R R UR[R5nO.UcS5e[R RU5Ul[TU]-X45 [R"U5nSUlURR!SS5Ul[%UR!SS55 UR'5 X lUR*S:Xaf[R R-UR(5(d [R."UR(5 [0R3S U35 [R R URS 5n[R R-U5(d US 35e[5U5Ul[R R URS 5n [R R-U 5(d U S 35e[9U 5Ul[R R URS 5n [R R-U 5(d U S 35eXR<S'XplUR*S:Xa[R R-UR(5(d [R."UR(5 [R R UR(S5n [AU S5n [BRD"URG55n U RIU 5 SSS5 gg!,(df  g=f)z Args: model (str): model id in modelscope work_dir (str): main directory for training and evaluating cfg_file (str): config file for training and evaluating kwargs: seed (int): random seed Nz?Config file should not be None if model is not from pretrained!pytorch dist_backendncclseedirzCurrent working dir is ztrain/tokens.txtz is missingztrain/lexicon.txtz&train/feature_transform.txt.80dim-l2r2 cmvn_file config.yamlw)% isinstancestrget_or_download_model_dir model_dirospathjoinr CONFIGURATIONdirnamesuper__init__r from_filelaunchertraingetr.r init_distr(rankexistsmakedirsloggerinfor token_tabler lexicon_tabler'configsopenyamldumpto_dictwrite)selfr'r(r)r*r+kwargsrK token_file lexicon_filefeature_transform_filesaved_config_pathfoutdata __class__s o/var/www/html/land-doc-ocr/venv/lib/python3.13/site-packages/modelscope/trainers/audio/kws_nearfield_trainer.pyr>KWSNearfieldTrainer.__init__)s eS ! !!;;'DN77<<(1(?(?A' j)j j'WW__X6DN 0""8,! #MM--nfE  63/0  99>77>>$--00 DMM* KK1(< =WW\\$..2DE ww~~j))Ej\+EE)%j1ww||DNN4GH ww~~l++I ~[-II+),7!# NND"Fww~~455 3%&k 2 35%; k" 99>77>>$--00 DMM* " T]]M J '-yy!23 4 .- .-s 6N N)c URSS5nUb+[RRU5(aX0lO/[RR UR S5UlURSS5UlUS(dS5eUS(dS 5eUS (dS 5eUSUlUSUl US Ul URS S 5=(a URS:XaSOS Ul URSn[R"U5nS US'S US'S US'[RR UR S5n[RR UR S5n[#URURX@R$UR&URUS5Ul[#URURXPR$UR&URUS5Ul[-UR(SURSS 5URR.R0R2URR.R0RSS5S9Ul[-UR*SURSS 5URR6R0R2URR6R0RSS5S9UlUR;UR5Ul[?UR<5nURS:Xa$[@RCSREU55 URb[RRUR5(a[GURUR<5 [HRJ"SSUR5n 0n [RRU 5(a9[MU S5n [NRP"U [NRRS9n SSS5 OO[@RCS5 0n U RS S!5S"-Ul*URTURS#S$'U RS%URS#S&S%5n XRS#S&S%'URS'S(5n URVS":aS)URX3n [[U 5Ul.URVS":a[^R`Rc5(deUR<ReUR\5Ul[^RfRhRkUR<5UlO*UR<ReUR\5UlURS:Xa[RRUR 5(d [Rl"UR 5 [RR UR S*5n[MUS+5n[NRn"URRq55nURsU5 SSS5 [@RuS,5 SnURS:Xa\[Rl"UR SS-9 [w[RR UR UR55nURS#RS.S/5nURS#S&n[xRz"UR<R}5US%US0S19n[xR~RUS2S3S4S5S6S79nSnURTS:XaUURS:XaE[RR UR S85n[UR<USSSS S5 [@RuS,5 0nUS9US9'URS:S"5US:'UUS.'URVUS;'URUS<'URXUS='URS#S>n[RR5n[URTU5GHnUR(RU5 UUS 'URSS%n[@RuS?REUU55 [UR<UUR4UR\UU5 [UR<UR8UR\U5unn[@RuS@REUUU55 URS:Xa[RR UR SAREU55n[UR<USSSS S5 [HRJ"SSU5n [UUUSB9n[MU S+5n[NRn"U5nURsU5 SSS5 URSCUU5 URSDUU5 UnURU5 GM Ub URS:XaUR5 [RR5U- n[@RuSEREUR5SF- 55 g!,(df  GN=f!,(df  GN=f!,(df  N=f)Ga% Args: kwargs: train_data (int): wave list with kaldi style for training cv_data (int): wave list with kaldi style for cross validation trans_data (str): transcription list with kaldi style, merge train and cv checkpoint (str): basemodel checkpoint, if None, default to use base.pt in model path tensorboard_dir (str): path to save tensorboard results, create 'tensorboard_dir' in work_dir by default need_dump (bool): wether to dump data with mapping tokens or not checkpointNz train/base.pttensorboard_dir tensorboard train_dataz'please config train data in dict kwargscv_dataz$please config cv data in dict kwargs trans_dataz/please config transcription data in dict kwargs need_dumpFrT preprocessor speed_perturbspec_augshufflezdump_train.txtz dump_cv.txt pin_memoryprefetch) batch_sizerh num_workersprefetch_factorzthe number of model params: {}z.pt$z.yamlr)Loaderz'Training with random initialized paramsepochrrA start_epochlr optimizerdevicegpucuda:r2r3zStart training...)exist_ok log_interval weight_decay)rsr{ming?gư>g{Gz?)modefactorpatiencemin_lr thresholdzinit.pt grad_clip grad_accum world_sizerD local_rank max_epochszEpoch {} TRAIN info lr {}z0Epoch {} EVAL info cv_loss {:.6f}, cv_acc {:.2f}z{}.pt)rprscv_lossz epoch/cv_losszepoch/lrTotal time spent: {:.2f} hours @)NrBr8r9rEr]r:r7r^r`rarbrDrcrKcopydeepcopyr(r rIrJ train_dataset cv_datasetr rA dataloaderworkers_per_gputrain_dataloader evaluation cv_dataloader build_modelr'r"rGwarningformatrresubrLrMload FullLoaderrrrrr train_devicetorchcuda is_availabletorparallelDistributedDataParallelrFrNrOrPrHrrAdam parameters lr_schedulerReduceLROnPlateaurdatetimenowrange set_epoch param_groupsrrdict add_scalarstepclose total_seconds) rQargsrRtrain_checkpoint train_confcv_confdump_train_file dump_cv_file num_params info_pathinfosfin lr_last_epoch device_namerVrWrXwriterry optim_confrtr final_epochsave_model_pathtraining_config max_epoch totaltimerprsrcv_acc info_dicts rZrAKWSNearfieldTrainer.trainks& "::lD9  'BGGNN;K,L,L.O ggll4>>?KDO%zz*;]Kl#N%NN#i H"HH   ML M .i(  .  H-1YY!^T \\.1 -- +#( # " '',,t}}6FGww||DMM=A 2 OOT__j:J:J   G0 doo079I9I040B0B04 04 6 !+   zz,6 **55EE LL..99==j!L !N ( OOzz,6 //::JJ LL33>>BBA  %%dll3 %djj1 99> NN;BB:N O ?? &277>>$//+J+J DOOTZZ 8vw@IEww~~i(()S)S IIc$//BE*)) NND EE 99Wb1A5/3/?/? Wm, $"&,,w"7 "DT"JL 3@ Wk*40jj51 ??Q !$//!23K)+6 ??Q JJ++-- .-t'8'89DJ**BB4::NDJt'8'89DJ 99>77>>$--00 DMM* " T]]M J '-yy!5!5!78 4 .  '( 99> KK  5" T]]D,@,@ACF||G,00D \\'*;7 JJ JJ ! ! #$#N35 ));;  <    q TYY!^ ggll4==)DO DJJtT!4 )  '('1+'> $(2|Q(G %*6'(, %"&))(, %LL),7 %%))+ 4++Y7E    ( ( /',OG $''*40B KK3::5"E F 4::y$2G2G,,fo G)$**d6H6H*.*;*;_NOGV KKBII7F, -yyA~"$'',,t}}/6~~e/D#F OT4 %t-FF67OD  # )S)T99Y/DJJt$*!!/7EB!!*b%8K   g &E8H  "tyyA~ LLN%%))+i7  4;;  # # % .0 1k*)>.-P*)s%>$m Am(m+ m m(+ m9 checkpoint_pathreturnc  SUlUb@[RRU5(a[R SU35 UnGO&URcUR SS5n[RRURSUS35Ul[R SUR35 [URURSUS 9n[S:0UD6UlURUR5n[UURUR5n[R S U35 URn[R S UR35 UR S S5b3UR S S5b [R S5 US n US n O-[R S5 URn URn [R SU 35 [R SU 35 [ R""URS5n SU SS'SU SS'SU S'SU S'SU S'UR SS5b USU SS'[%XU UR&UR(SSS5n [+U SUR SS5SURR,R.R0URR,R.R S S!5S"9n UR S#S5cS$5eUS#nUR35R5S%S5R7S&5n0nS1nS'1nS'S0nUHn[9UUR&UR(5unn0UU'UUUS('SRS)U55UUS*'UVs/sHnUR;U5PM nUVs/sHnUR;U5PM n[=UU5H unnUR US5bMUUU'M" M SnUR?5HunnUUS+US,3- nM [R S-U35 UR S.S/5S:=(a [@RBRE5nUR S0S15nURFS2:aU(aS3URH3n[KU5Ul&URUR5Ul'[QX@RN5 URNRSURL5Ul'0nUR S4S5b US4US4'OO[RRUU5n[RRURS5U-5US4'US4Ul+[RRURV5(d [RX"URV5 [R[S65 [\R\R_5n[aURNU URLUUU5n [\R\R_5U- n[R[S7RcURe5S8- 55 [UU U U S99n![gS:0U!D6n"[iU"5 gs snfs snf);a  Args: checkpoint_path (str): evaluating with ckpt or default average ckpt kwargs: test_dir (str): local path for saving test results test_data (str): wave list with kaldi style trans_data (str): transcription list with kaldi style average_num (int): the NO. to do model averaging(checkpoint_path==None) batch_size (int): batch size during evaluating keywords (str): keyword string, split with ',' gpu (int): evaluating with cpu/gpu: -1 for cpu; >=0 for gpu Nz evaluating with specific model: average_numrzavg_z.ptz!default average model not exist: T) dst_modelsrc_pathval_bestavg_numz(average model convert to kaldi network: zevaluating with average model: test_datarbz/evaluating with specific data and transcriptionz5evaluating with cross validation data during trainingz test data: z trans data: rdi filter_conf max_lengthr min_lengthFrerfrgrk batch_confrhrirj)rkrhpersistent_workersrlrmkeywordszat least one keyword is needed ,ztoken_idc3># UHnS[U5-v M g7f)z%s N)r5).0is rZ /KWSNearfieldTrainer.evaluate..s!;MDKqDKs token_str(z) zToken set is: rvrqrucpurrwtest_dirtest_zStart evaluating...rr)rrrb score_file)5avg_checkpointr8r9rErGrrBr:r(rr rrKr!rarbrrr rIrJr rrrstripreplacesplitraddzipitemsrrrrrr test_device test_modelrrbasenamerrFrHrrrrrrprint)#rQrrrReval_checkpointr avg_kwargs model_cvt kaldi_cvtrrb test_conf test_datasettest_dataloader keywords_str keywords_listkeywords_tokenkeywords_idxsetkeywords_strsetkeywords_tokenmapkeywordstrsindexesrtxtidx token_printuse_cudartesting_config base_namerr det_kwargs det_resultss# rZevaluateKWSNearfieldTrainer.evaluate-sn #  &277>>/+J+J NN2?2CD F-O""* **]B7&(ggll4==59'#3F'H#78K8K7LMO!"11!]]!#  '4&Aj&A# ,,T\\: ,''MM >ykJL#11O NN1$2E2E1FG I ::k4 ( 4 ::lD ) 5 NNL M{+I -J NNG I IJYK01j\23MM$,,~">? 17 - .12 - .%* /" % *$ ) ::lD ) 54:<4HIl #L 1,YI-1-=-=-1-?-?-24 % zz,6# //::JJ LL33>>BBA  zz* '+, N-M N,j) $**,44S"=CCCH #")$aL$G+GT5E5E,0,>,>@MD'&(N7 #29N7 #J /3577;MDK;M4MN7 #K 0-1 2T_  #T 2-4 5W_  #W 5g.S$((d3;-0%c*/% )//1HC cU!C5+ +K2 }56 ::eR(A-K%**2I2I2Kjj51 ??Q 8!$//!23K(5**4<<89//,,T-=-=> ::j$ ' 3)/ );N: &((9I)+dmm6= 6I*KN: &&z2 ww~~dmm,, KK &  )*%%))+ "4??O#'#3#3^#2ND %%))+i7  4;;  # # % .0 1!!!  "/J/  kk 3 5s ['![,c[R"URUSS9n[U[5(a[ US5(a UR $[U[R5(aU$g)zInstantiate a pytorch model and return. By default, we will create a model using config from configuration file. You can override this method in a subclass. T)cfg_dicttrainingr'N) r from_pretrainedr7r4r hasattrr'rModule)rQrKr's rZrKWSNearfieldTrainer.build_models]%% NNWt= eZ ( (WUG-D-D;;  ryy ) )L*c [R"SS5cS[RS'[R"SS5cS[RS'[R"SS5cS[RS'[R"SS5cS[RS'[R"SS5cS [RS'[[RS5Ul[[RS5Ul[[RS5Ul[RSUl[RSUlUS :Xa^UR S :aM[RS RUR 55 [R"URS S 9 O:US :a4[R"URS S 9 [R"5 [RSRUR URUR URUR55 g)NRANK0 LOCAL_RANK WORLD_SIZE1 MASTER_ADDR localhost MASTER_PORT29500rz'init dist on multiple gpus, this gpu {}zenv://)backend init_methodz-RANK {}/{}/{}, Master addr:{}, Master port:{})r8getenvenvironintrDrr master_addr master_portrGrHrdistinit_process_groupr.barrierrQ train_nodess rZrCKWSNearfieldTrainer.init_dists 99VT " *!$BJJv  99\4 ( 0'*BJJ| $ 99\4 ( 0'*BJJ| $ 99]D ) 1(3BJJ} % 99]D ) 1(/BJJ} % 6*+ bjj67bjj67::m4::m4 ! " ELLOO%&'' --8E 1_  # #))x A LLN CJJ OOTYY9I9I    rcUS:XaUURS:aD[RSRUR55 [ R "5 ggUS:a+[ R"5 [ R "5 gg)Nrz*destory dist on multiple gpus, this gpu {})rrGrHrrr$destroy_process_groupr&r's rZ uninit_distKWSNearfieldTrainer.uninit_distsm ! " @GG)***, # 1_ LLN  & & (r)rr]rKrarrr.r@rJrr"r#r'r7rcrDrrr^rrrrIr`rrrrbr(r)r)__name__ __module__ __qualname____firstlineno__rr5rrr>rArfloatr rrrrCr,__static_attributes__ __classcell__)rYs@rZr%r%%s,0481G @!@!@!$C=@! (1 @! "*# @!@!D@1D[["3:.[z bii B ) )rr%)>rrr8rtypingrrrrtorch.distributed distributedr$rM tensorboardXrrrtorch.utils.datar modelscope.metainfor modelscope.modelsr r Mmodelscope.msdatasets.dataset_cls.custom_datasets.audio.kws_nearfield_datasetr modelscope.trainers.basermodelscope.trainers.builderrmodelscope.utils.checkpointrrmodelscope.utils.configrmodelscope.utils.constantrrmodelscope.utils.devicermodelscope.utils.loggerrmodelscope.utils.torch_utilsrkws_utils.batch_utilsrrrkws_utils.det_utilsrkws_utils.file_utilsrrrkws_utils.model_utilsr r!r"rGregister_module"speech_kws_fsmn_char_ctc_nearfieldr%rrrZrKs  ++ & '(/00H*G1.8MM,KK66  ;;=[)+[)=[)r