xi}JdZddlZddlZddlmZddlmZddlm Z dde fdZ y) z Gradio UI Training Tab Module Contains the dataset builder and LoRA training interface components. The outer gr.Tab wrapper is now created in __init__.py. N)t)create_help_button)DEBUG_TRAININGreturnc86ttjjdk7}|rdnd}|rdnd}|rdnd}t j dt j 5t jtd5tdt j dtd d t j5t jd 5t j d t j5t jtd dtddgd}t jtddd}dddt jtdd} dddt jd 5t j dt j5t jtddtddgd} t jtddd} dddt jtd d} ddddddt j d!t j5t jd" 5t jgd#gd$td%dd&'} dddt jd 5t j d(td)d*t jtd+d,td-.}t jtd/d&td0dg1}t jd2dd3dgd4}t jd5dd6dgd4}t jtd7d8td9dg:}t j td;dftd?d@fgd@tdAtdBdgC}t j"dDddEdDtdFtdGdgH}ddddddt j dItdJd*t j5t jd 5t j$tdKt jtdLdtdMdg1}t jtdNdtdOdg1}dddt jd 5t jtdPddQR}ddddddt jtdSdd"T}t j dUtdVd*t j5t jd 5t j"dDdDddDtdWtdXdgH}t j&tdYdZd[}t jtd\d}dddt jd" 5t j5t jtd]dd^_}dddt j5t jtd`dda_}t j(gdbdctddtdedgC}dddt j5t jtdfdgdh_}t jtdidgtdjddk} t j*d}!dddt j5t j,tdldDm}"t jtdntdop}#t j(gdqtdrs}$t j,tdtddu}%dddt j5t j(gdvdwtdxy}&t jtdzd&{}'t jtd|d}}(dddt jtd~d})ddddddt j dtdd*t j5t jd 5t jtdddtddg}*dddt jd 5t jtdddQR}+ddddddt jtddd"T},t j dtdd*t j$tdt j5t jd 5t jtddtddg:}-dddt jd 5t jtdddQR}.ddddddt jtdd}/t j$tdt j5t j(dddgdddg}0dddt j5t jd 5t jtdddtddg}1dddt jd 5t jtdddQR}2ddddddt jtdddT}3dddt jtd5tdt j5t jd" 5t j dtdd*t j$tdt jtdddtddg}4t jtdd}}5t jtdddT}6dddt jd 5t j d(tdd*t j"ddddtdtddgH}7t j"ddddtdtddgH}8t j"ddddtd}9ddddddt j dtdd*t j5t j,tddtddg1}:t j"|d||td};t j"ddddtdtddgH}t j"ddddtdītdūdgH}?t j,tdƫddDȫ}@dddt j5t jtdɫddtd˫dg}Adddt j5t jddddg:}Bdddt j d!t j5t jd 5t jtdϫddQR}Cdddt jd 5t jtdЫddQR}Dddddddt jtdҫdd"T}Et j5t jtdӫddEddի}Ft j.td֫d׫}Gdddt j dtd٫d*t j5t jtdګddܬ.}Ht jtdݫd}}Idddt jtdޫd}Jdddt jd߫5tdt j5t jd" 5t j dt j$dt jdddddg}Kt jdd}}Lt jdddT}Mdddt jd 5t j dt j"dddddddgH}Nt j"dddddddgH}Ot j,dddDddg}Pt jddddg1}Qt jddddg1}Rt jddddg1}St jdd&ddg1}Tddddddt j dt j5t j,ddddg1}Ut j"ddddd}Vt j"ddddd}Wt j"ddddd}Xdddt j5t j"ddddd}Yt j"ddddÐdddgH}Zt j,dddDȫ}[dddt j5t jdddddg}\dddt j d!t j5t jd 5t jdddQR}]dddt jd 5t jdddQR}^ddddddt jd dd"T}_t j5t jd ddEddի}`t j.d d׫}adddt j d t j5t jd dd.}bt jdd}}cdddt j5t j(dgddddgC}dt jdd}}edddt jdd}fddddddt j*d}gt j*ddd}hiddd d d d d ddd d!d"d#d$d%d&d'id(d)d*d+d,d-d.d/d0 d1!d2"d3#d4$d5%d6&d7'd8(id9)d:*d;+d<,d=-d>.d?/d@0dA1dB2dC3dD|gdE4dF5dG6dH7dI8idJ9dK:dL;dM<dN=dO>dP?dQ@dRAdSBdTCdUDdVEdWFdXGdYHdZIid[Jd\Kd]Ld^Md_Nd`OdaPdbQdcRddSdeTdfUdgVdhWdiXdjYdkZ[\]^_`abdecf|hdl S#1swYxYw#1swYxYw#1swY9xYw#1swYxYw#1swY#xYw#1swYxYw#1swYTxYw#1swYYxYw#1swYxYw#1swYgxYw#1swYlxYw#1swYxYw#1swYJxYw#1swYxYw#1swYxxYw#1swYxYw#1swYixYw#1swYNxYw#1swYSxYw#1swYxYw#1swYxYw#1swYxYw#1swYxYw#1swYxYw#1swYxYw#1swYbxYw#1swYxYw#1swY xYw#1swY xYw#1swY xYw#1swY xYw#1swY xYw#1swY xYw#1swY 6xYw#1swY xYw#1swY vxYw#1swY LxYw#1swY xYw#1swY xYw#1swY xYw#1swY MxYw#1swYxYw#1swYxYw#1swYxYw#1swYxYw#1swY$xYw#1swYxYw#1swY4xYw#1swYxYw#1swYxYw#1swYxYw#1swYxYw#1swY.xYw#1swYxYw#1swYxYw#1swYzxYw#1swYxYw(maCreate the training tab content (without the outer gr.Tab wrapper). The outer gr.Tab is now created in __init__.py so that Generation and Training tabs are siblings under the same gr.Tabs container. Args: dit_handler: DiT handler instance llm_handler: LLM handler instance init_params: Dictionary containing initialization parameters and state. If None, service will not be pre-initialized. Returns: Dictionary of Gradio components for event handling OFFdiu

🎵 LoRA Training for ACE-Step

Build datasets from your audio files and train custom LoRA adapters

ztraining.tab_dataset_buildertraining_datasetz

ztraining.quick_start_titlez

Choose one: Load existing dataset OR Scan new directory

)scaleu#

📂 Load Existing Dataset

ztraining.load_dataset_labelz./datasets/my_lora_dataset.jsonztraining.load_dataset_infozhas-info-container)label placeholderinfo elem_classesr ztraining.load_btnprimary)variantr Nztraining.load_statusF)r interactiveu

🔍 Scan New Directory

ztraining.scan_labelz/path/to/your/audio/folderztraining.scan_infoztraining.scan_btn secondaryztraining.scan_statusz
)#FilenameDurationLyricsLabeledBPMKeyCaption)numberstrr r r r r r ztraining.found_audio_filesT)headersdatatyperrwrapu

⚙️ z training.dataset_settings_headerz

ztraining.dataset_namemy_lora_datasetz!training.dataset_name_placeholder)rvaluerztraining.all_instrumentalztraining.all_instrumental_info)rr%rrzFormat Lyrics (LM)zMUse LM to format/structure user-provided lyrics from .txt files (coming soon))rr%rrrzTranscribe Lyrics (LM)z4Use LM to transcribe lyrics from audio (coming soon)ztraining.custom_tagze.g., 8bit_retro, my_styleztraining.custom_tag_info)rrrrztraining.tag_prependprependztraining.tag_appendappendztraining.tag_replacereplaceztraining.tag_positionztraining.tag_position_info)choicesr%rrrr ztraining.genre_ratioztraining.genre_ratio_info)minimummaximumstepr%rrru

🤖 ztraining.step2_titleztraining.step2_instructionztraining.skip_metasztraining.skip_metas_infoztraining.only_unlabeledztraining.only_unlabeled_infoztraining.auto_label_btnlg)rsizeztraining.label_progress)rrlinesu

👀 ztraining.step3_titleztraining.select_sampleztraining.select_sample_infoztraining.audio_previewfilepath)rtyperztraining.filenameztraining.captionzMusic description...)rr0rztraining.genrezpop, electronic, dance...)Use Global RatiorGenrer3ztraining.prompt_override_labelztraining.prompt_override_infoztraining.lyrics_editable_labelz&[Verse 1] Lyrics here... [Chorus] ...ztraining.raw_lyrics_labelztraining.no_lyrics_placeholder)rr0rrvisiblez training.bpm)r precisionztraining.key_labelztraining.key_placeholder)rr)2346zN/Aztraining.time_sig)r)rztraining.duration_s)rr7r) instrumentalenzhjakoesfrdeptruunknownr=ztraining.language)r)r%rztraining.instrumental)rr%ztraining.save_changes_btn)rztraining.edit_statusu

💾 ztraining.step4_titleztraining.save_pathz./datasets/dataset_name.jsonztraining.save_path_info)rr%rrrztraining.save_dataset_btnztraining.save_statusu

⚡ ztraining.step5_titleztraining.step5_introztraining.load_existing_labelztraining.load_existing_infoztraining.load_dataset_btnztraining.step5_detailszPreprocess ForLoRALoKrzILoRA keeps compatibility mode; LoKr uses per-sample source-style context.)rr)r%rrztraining.tensor_output_dirz./datasets/preprocessed_tensorsztraining.tensor_output_infoztraining.preprocess_btnztraining.preprocess_progressztraining.tab_train_loratraining_trainu

📊 ztraining.train_section_tensorsz$training.train_tensor_selection_descz!training.preprocessed_tensors_dirz"training.preprocessed_tensors_info)rrr%rrztraining.dataset_infoztraining.train_section_lora@ztraining.lora_rankztraining.lora_rank_infoiztraining.lora_alphaztraining.lora_alpha_infogg?g?g?ztraining.lora_dropout)r+r,r-r%ru

🎛️ ztraining.train_section_paramsztraining.learning_ratega2U0*3?ztraining.learning_rate_infoiztraining.max_epochsztraining.batch_sizeztraining.batch_size_infoztraining.gradient_accumulationz#training.gradient_accumulation_info2ztraining.save_every_n_epochsg?g@g@ztraining.shiftztraining.shift_infoz training.seed*)rr%r7ztraining.output_dirz ./lora_outputztraining.output_dir_infozResume Checkpointz#./lora_output/checkpoints/epoch_200z3Directory of a saved LoRA checkpoint to resume fromztraining.start_training_btnztraining.stop_training_btnstopztraining.training_progressztraining.training_log)rrr0 max_linesr ztraining.training_loss_title)rr u

📦 ztraining.export_headerztraining.export_pathz./lora_output/final_loraz./lora_output/my_loraztraining.export_lora_btnztraining.export_statusu🚀 Train LoKr training_lokru"

📊 Preprocessed Tensors

zvSelect the directory containing preprocessed tensor files (`.pt` files). These are created in the Dataset Builder tab.zPreprocessed Tensors Directoryz@Path to directory containing manifest.json and tensor .pt files.z Load Datasetz Dataset Infou

⚙️ LoKr Settings

zLoKr Linear Dimz/Adapter rank-like width for LoKr linear layers.zLoKr Linear Alphaz!Scaling factor for LoKr adapters.z LoKr Factorz--1 uses automatic Kronecker factor selection.)rr%r7rrzDecompose Bothz&Enable decomposition on both matrices.z Use Tuckerz!Enable Tucker decomposition mode.z Use Scalarz%Enable scalar calibration in LyCORIS.zWeight Decompose (DoRA)z6Enable DoRA-style weight decomposition when supported.u(

🎛️ Training Parameters

z Learning RategMbP?z;LoKr commonly uses a higher LR than LoRA. Tune per dataset.iz Max Epochsz Batch SizezGradient AccumulationzSave Every N EpochsShiftz$Turbo model training timestep shift.SeedzOutput Directoryz ./lokr_outputz9Where LoKr checkpoints and final weights will be written.zStart LoKr Trainingz Stop TrainingzTraining Progressz Training LogzLoKr Training Lossu

📦 Export LoKr

z Export Pathz./lokr_output/final_lokrz./lokr_output/my_lokru📦 Export LoKrz Latest (auto)zCheckpoint EpochzDSelect a specific epoch checkpoint to export, or keep Latest (auto).u↻ Refresh Epochsz Export Status) is_training should_stopload_json_path load_json_btnload_json_statusaudio_directoryscan_btn scan_statusaudio_files_table dataset_nameall_instrumental format_lyricstranscribe_lyrics custom_tag tag_position skip_metasonly_unlabeledauto_label_btnlabel_progresssample_selector preview_audiopreview_filename edit_caption edit_genreprompt_override genre_ratio edit_lyricsraw_lyrics_displayhas_raw_lyrics_stateedit_bpm edit_keyscale edit_timesig edit_duration edit_languageedit_instrumental save_edit_btn edit_status save_pathsave_dataset_btn save_statusload_existing_dataset_pathload_existing_dataset_btnload_existing_statuspreprocess_modepreprocess_output_dirpreprocess_btnpreprocess_progressdataset_builder_statetraining_tensor_dirload_dataset_btntraining_dataset_info lora_rank lora_alpha lora_dropout learning_rate train_epochstrain_batch_sizegradient_accumulationsave_every_n_epochstraining_shift training_seedlora_output_dirresume_checkpoint_dirstart_training_btnstop_training_btntraining_progress training_logtraining_loss_plot export_pathexport_lora_btn export_statuslokr_training_tensor_dirlokr_load_dataset_btnlokr_training_dataset_infolokr_linear_dimlokr_linear_alpha lokr_factorlokr_decompose_bothlokr_use_tuckerlokr_use_scalarlokr_weight_decomposelokr_learning_ratelokr_train_epochslokr_train_batch_sizelokr_gradient_accumulationlokr_save_every_n_epochslokr_training_shift) lokr_training_seedlokr_output_dirstart_lokr_training_btnstop_lokr_training_btnlokr_training_progresslokr_training_loglokr_training_loss_plotlokr_export_pathlokr_export_epochrefresh_lokr_export_epochs_btnexport_lokr_btnlokr_export_statustraining_state)r rstripuppergrHTMLTabsTabrrRowColumnTextboxButton DataframeCheckboxRadioSliderMarkdownAudioDropdownStateNumberPlot)i dit_handler llm_handler init_paramsdebug_training_enabled epoch_min epoch_step epoch_defaultr]r^r_r`rarbrcrdrerfrgrhrirtrjrkrlrmrnrorprqrrrsrurvrwrxryrzr{r|r}r~rrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrsi D/mnt/workspace/ACE-Step-1.5/acestep/ui/gradio/interfaces/training.pycreate_training_sectionrsL!0668>>@EI+I,#J/ATMGG    VVA45 6 1 2 GG0012N0O/PQ  YYQ'GGAB)+"#$A"B(I!"#?!@PdOe"# * )+ !4G2HR[cd(e "(*zz 67$)($(YYQ'GG>?*,**"#$9":(D!"#7!8H\G]"# + $&99Q/B-C[`a#b"#%** 67$)#K(!> GGFOYYQ')+ k!\ <=$)! )%(YYQ'GGk!,N*O)PPUVW#%:: 78/$%&I$J$L (*{{ ;<"?@PdOe($ %'KK2#l}Q|R$) %M)+ 6#Scwbx$) )%"$ 56$@9:J^I_"J $&8856 B45x@56 B! ( 78;eD EYYQ'KK"> ?@!# 56#9:J^I_"J &([[ 9:#=>NbMc&N(YYQ'%'YY34 )!&N(( ZZ12!N GGmA&<$=#>eD EYYQ'&(ii ! ! 89<=MaLb 'O%'HH 89'$)%M (*zz 34$)($!(*YYQ'')zz"#$6"7"#(>( "%'ZZ"#$4"5"#(C& +-++$L"4"#$D"E!"#B!CSgRh + "&(jj"#$D"E"#(T' .0ZZ"#$?"@"#()*J(K(-$) .*02xx,"#%99"#N"3&'$)+ "#$8"9()*D(E) (*{{$C"#$7"8( )+ "#$9":&'(-) "&(* $u"0"#$7"8) -/KK"#$;"<"&-))+ !4O2PZe(f "#%** 67$)#KG(-| GGmA&<$=#>eD EYYQ' "  45?$B89I]H^ !I(YYQ'')yy56 )!($(**./!K GGl1%;#<"=UC D KK01 2YYQ'13 >?$E<=MaLb2.( YYQ'02 56 +!1-($&::./!$ KK23 4"$++*#V, duItJ #YYQ',.JJ <=?$E<=MaLb -)(YYQ'%'YY34 )!&N(#%**67!# c 7p VVA/0 1 / 0YYQ'GGi*J(K'LERSKK"H IJ*,** CD$E?CDThSi +'(*yy3N1OYd'e$,.JJ 78$)-)((YYQ'GGk!,I*J)K5QR " ! #  4589I]H^ !I"$ ! #! 569:J^I_ "J$&99 # #!! 78 $L+(+d GG&q)H'I&J%P Q " 4589I]H^! "yy% #'12   $&991256FZE[ $ )+ <=@AQePf )%1B&(ii :; '#"$,-01AU@V "!# O,! %0"$**12) /56FZE[ #(* - EN^r]s)% GGFOYYQ')+78 )!*&( YYQ'(* 67 &!)%(!# 45!!  !zz34 %    &(WW:;&" GGmA&>$?#@F G jj234 7 #%))A.H,IS^"_ JJ01!MW2bVV% &  /YYQ'GG@AKKH 02zz>$E?_pDoE 0,-/IInk,Z)13,$)2.!(,YYQ'GG;<&(ii ! # /N^r]s 'O)+ ! #!1@PdOe )%#%))+ "#L\p[q #K +-++.#EUiTj+' ')kk*#@PdOe'O ')kk*#DThSi'O -/KK7"Ueydz-)Q(/J GG> ?%'YY)Vfze{&" %'II & %!)+ & )%.0YY1 .*/>+-99 / ,(')ii!?OcNd '#&(YY &"%0"$**,) /Tdxcy # GGFOYYQ'.0ii- )!/+( YYQ'-/YY' &!.*(&(ZZ)!& " $&JJ( %  %!+-''.+' GG3 4#%::'4 7$ #%)),> "T $&KK,-),_pDoE %! 24;OYd1e."$%!" }'W ^HHTNXXeEJKNg .g  g  , g  ? g  H g  {g  .g   g  ,g  g  .g  jg   g  jg  .!g " .#g $ .%g & ?'g ( )g * ,+g ,  -g . j/g 0 ?1g 2 {3g 4 {5g 6 07g 8  49g : H;g < =g >  ?g @ Ag B Cg D .Eg F Gg H {Ig J YKg L ,Mg N {Og R %&@Sg T $%>Ug V  4Wg X ?Yg Z !6[g \ .]g ^ 2_g ` !6ag d 2eg f ,gg h !6ig j Ykg l jmg n  og p qg r  sg t ,ug v !6wg x 2yg z .{g | }g ~ ?g @ !6Ag B 0Cg D .Eg F .Gg H  Ig J 0Kg L {Mg N ?Og P Qg T #$0(' ('(k' &W s*Al AAf0 Aa+"*Aa AA`7 (Aa Aa+!*Aa AAa (Aa Aa+ 1Af0;Ab(Aa8 :AbEAb 3Ab;>Af09Ab9A7Ab Ab9%"Ab, Ab9AAf0.Ad!A1Ac 6Ad!Ad )"Ac Ad 'AAc 8Ad A#Ac- 7Ad B Ac: Ad Af0Ae%.Ad. Ae1"Ad; AeA=Af0Ae//-Ae Ae/:"Ae" Ae/$AAf0>Ae<Af08Af#.Af =Af#"Af =Af#)Af0.&Al Ai4Ag B/Af= :AgB$Ag <Ag>AiB.Ag$0Ai A4Ag1Ai.Ag> Ai&Ah 1Ai2Ah2@ "Ah @+Ah2A "Ah% A+Ah2A3=AiB0AAh?C4>AiD2AAi E4(AiFAlF9 AlGAjG0A5Ai& I%AjJCAi3 MAjM1AlNA)Aj O8AlPAAjQ(AlR Aj'R$1AlSAkS,Aj4 TAkT$Ak T>AkU5AlU;4AkV/2AlW!4Ak(XAlX19Ak5Y* AlZ Al`7Aa`<Aa aAa a Aa+aAaaAa aAa( a#Aa+a+Aa5 a0Af0a8Ab a=AbbAb b AbbAb bAf0bAb) b$Ab9b,Ab6 b1Ab9b9Ac b>Af0cAc c Ad!cAccAd c Ac*c%Ad c-Ac7c2Ad c:Adc?Ad dAdd Ad dAd dAd!d!Ad+ d&Af0d.Ad8 d3Aed;Ae eAeeAe e Af0eAe eAe/e"Ae, e'Ae/e/Ae9 e4Af0e<Af fAf0f Af fAf#fAf fAf#f#Af- f(Af0f0Af: f5Alf=Ag gAgg Ag gAggAg! gAig$Ag. g)Aig1Ag; g6Aig>Ah hAih Ah hAihAh" hAh2h%Ah/ h*Ah2h2Ah< h7Aih?Ai iAii Ai iAiiAi# iAli&Ai0 i+Aji3Ai= i8AjjAj jAlj Aj jAljAj$ jAlj'Aj1 j,Alj4Aj> j9AkkAk kAkkAk kAlkAk% k Alk(Ak2 k-Alk5Ak? k:AllAl lAllAl)N) __doc__osgradioracestep.ui.gradio.i18nracestep.ui.gradio.help_contentracestep.constantsrdictrrrs) $=,t 4t r