254 lines
44 KiB
Plaintext
254 lines
44 KiB
Plaintext
|
§
|
|||
|
|
4$�h £ ã ó — d Z ddlmZ ddlZddlmZ ddlmZ ddlmZm Z m
|
|||
|
|
Z
|
|||
|
|
m Z m Z m
|
|||
|
|
Z
|
|||
|
|
mZmZ ddlmZmZmZmZmZmZmZmZmZmZmZmZm Z m
|
|||
|
|
Z
|
|||
|
|
m Z m Z m Z m Z m!Z!m"Z"m#Z#m$Z$m Z m%Z%m&Z&m'Z'm(Z(m)Z)mZm*Z*m
|
|||
|
|
Z
|
|||
|
|
m Z m Z m#Z#m'Z'm)Z)mZ ddl)Z)ddlT ddl+m,Z,m-Z- dd l.m/Z/ ddlZddl0Z1dd
|
|||
|
|
l2m3Z3 ddlmZ dd l4mZmZ5 d d
|
|||
|
|
d d
|
|||
|
|
d
|
|||
|
|
dœZ6 ej7 d d e6¬¦ « d„ ¦ « Z8e, G d„ de¦ « ¦ « Z9 G d„ de#¦ « Z: G d„ de:¦ « Z;dS )z8
|
|||
|
|
2025.8.4
|
|||
|
|
2025.8.5
|
|||
|
|
4.55.1
|
|||
|
|
0.21.0
|
|||
|
|
__UNSLOTH_VERSIONING__
|
|||
|
|
é )ÚTensorN)Ú
|
|||
|
|
functional)ÚAnyÚListÚOptionalÚTupleÚUnionÚDictÚSetÚCallable)%ÚAutoModelForCausalLMÚ
|
|||
|
|
AutoTokenizerÚBaseImageProcessorr Ú DataCollatorÚ DataCollatorForLanguageModelingÚDataCollatorForSeq2SeqÚ
|
|||
|
|
DataLoaderÚDatasetÚEvalLoopOutputÚFeatureExtractionMixinÚIterativeSFTConfigÚIterativeSFTTrainerr Ú
|
|||
|
|
PPODecoratorsÚPathÚ PeftModelÚPreTrainedModelÚPreTrainedTokenizerBaseÚProcessorMixinÚTrainerÚTrainingArgumentsr Úgenerate_model_cardÚget_comet_experiment_urlÚis_peft_availableÚis_wandb_availableÚosÚtorchÚwarningsr r r r r# r% r& )Ú*)Ú dataclassÚfield)ÚVersion)Ú nullcontext)r r TF)Úepilogue_fusionÚ max_autotuneÚ
|
|||
|
|
shape_paddingz
|
|||
|
|
trace.enabledztriton.cudagraphs)ÚdynamicÚ fullgraphÚoptionsc ó’ — t j | d| j d ¦ « dd¬¦ « }t j | d¦ « dd¬¦ « }g }t ||¦ « D ]‘\ }}| t j ¦ « }t j |d| d¦ « ¬¦ « d¦ « }t j
|
|||
|
|
|d¬¦ « }||z
|
|||
|
|
} | | ¦ « Œ’ t j |¦ « }| | j d | j d f¦ « }|S )Néÿÿÿÿé r )ÚchunksÚdim)r7 Úindex)r7 é )
|
|||
|
|
r& ÚchunkÚreshapeÚshapeÚzipÚtoÚfloat32ÚgatherÚ unsqueezeÚsqueezeÚ logsumexpÚappendÚconcat)
|
|||
|
|
Úlogitsr8 Úchunked_logitsÚ
|
|||
|
|
chunked_indexÚall_per_token_logpsÚ chunk_logitsÚ chunk_indexÚselected_logitsÚlogsumexp_valuesÚper_token_logpss
|
|||
|
|
úf/workspace/Fine-tuning/DS-LLM-TEMPLATE-FINETUNING/unsloth_compiled_cache/UnslothIterativeSFTTrainer.pyÚ chunked_selective_log_softmaxrP " s5 € õ ”[ § ¢ °°F´LÀÔ4DÑ!EÔ!EÐPQÐYZÐ[Ñ[Ô[€NÝ”[ § ¢ ¨rÑ!2Ô!2¸QÀaÐHÑHÔH€MØ Ðå%(¨ ¸ Ñ%GÔ%Gð 4ð 4Ñ!ˆ �kØ#—’¥u¤}Ñ5Ô5ˆ Ý œ, |¸2À{×G\ÒG\Ð]_ÑG`ÔG`ÐaÑaÔa×iÒiÐjlÑmÔmˆÝ œ?¨<¸rÐBÑBÔBÐØ)Ð,<Ñ<ˆØ×"Ò" ?Ñ3Ô3Ð3Ð3ØÝ œ,Ð':Ñ;Ô;ÐØ-×5Ò5°v´|ÀA´ÈÌ ÐUVÌÐ6XÑYÔYÐØ Ð ó c óš ‡ — e Zd ZU dZ edddi¬¦ « Zee ed< edddi¬¦ « Z ee
|
|||
|
|
ed < d+ˆ fd*„ Z ˆ xZ S ),ÚUnslothIterativeSFTConfigaÅ
|
|||
|
|
|
|||
|
|
Configuration class for the [`IterativeSFTTrainer`].
|
|||
|
|
|
|||
|
|
This class includes only the parameters that are specific to Iterative SFT training. For a full list of training
|
|||
|
|
arguments, please refer to the [`~transformers.TrainingArguments`] documentation. Note that default values in this
|
|||
|
|
class may differ from those in [`~transformers.TrainingArguments`].
|
|||
|
|
|
|||
|
|
Using [`~transformers.HfArgumentParser`] we can turn this class into
|
|||
|
|
[argparse](https://docs.python.org/3/library/argparse#module-argparse) arguments that can be specified on the
|
|||
|
|
command line.
|
|||
|
|
|
|||
|
|
Parameters:
|
|||
|
|
> Parameters that control the model
|
|||
|
|
|
|||
|
|
model_init_kwargs (`dict[str, Any]` or `None`, *optional*, defaults to `None`):
|
|||
|
|
Keyword arguments for [`~transformers.AutoModelForCausalLM.from_pretrained`], used when the `model`
|
|||
|
|
argument of the [`IterativeSFTTrainer`] is provided as a string.
|
|||
|
|
|
|||
|
|
> Parameters that control the data preprocessing
|
|||
|
|
|
|||
|
|
max_length (`int` or `None`, *optional*, defaults to `None`):
|
|||
|
|
Maximum length of the tokenized sequence. Sequences longer than `max_length` are truncated.
|
|||
|
|
truncation_mode (`str`, *optional*, defaults to `"keep_end"`):
|
|||
|
|
The truncation mode to use, either `"keep_end"` or `"keep_start"`.
|
|||
|
|
optimize_device_cache (`bool`, *optional*, defaults to `False`):
|
|||
|
|
Whether to optimize accelerator cache for slightly more memory-efficient training.
|
|||
|
|
|
|||
|
|
NÚhelpzvLLM SamplingParams)ÚdefaultÚmetadataÚvllm_sampling_paramsr4 z8Chunk size to reduce memory usage. -1 is most efficient.Úunsloth_num_chunksFÚnor5 é r éú ç-C ëâ6
|
|||
|
|
?ç{®Gáz„?çÍÌÌÌÌÌì?ç+‡ÙÎ÷ï?ç:Œ0âŽyE>ç ð?ç @Úlinearçš™™™™™¹?ÚpassiveÚwarningTÚstepsr9 éô éO
|
|||
|
|
ÚO1ÚautoÚ ç Ú
|
|||
|
|
adamw_8bitÚlengthÚ
|
|||
|
|
every_saveÚlasté Úkeep_endc‡ óö •— |dk rt d|› d�¦ « ‚|dk rt d|› d�¦ « ‚|€|#dk r
|
|||
|
|
|$dk rd}d }# t ¦ « j dŽi d
|
|||
|
|
|“d |“d |“d
|
|||
|
|
|“d|“d|“d|“d|“d| “d|
|
|||
|
|
“d| “d| “d|
|
|||
|
|
“d|“d|“d|“d|“d|“d |“d |“d |“d |“d |“d!|“d"|“d#|“d$|“d%| “d&| “d'| “d(| “d)| “d*|!“d+|"“d,|#“d-|$“d.|%“d/|&“d0|'“d1|(“d2|)“d3|*“d4|+“d5|,“d6|-“d7|.“d8|/“d9|0“d:|1“d;|2“d<|3“d=|4“d>|5“d?|6“d@|7“dA|8“dB|9“dC|:“dD|;“dE|<“dF|=“dG|>“dH|?“dI|@“dJ|A“dK|B“dL|C“dM|D“dN|E“dO|F“dP|G“dQ|H“dR|I“dS|J“dT|K“dU|L“dV|M“dW|N“dX|O“dY|P“dZ|Q“d[|R“d\|S“d]|T“d^|U“d_|V“d`|W“da|X“db|Y“dc|Z“dd|[“de|\“df|]“dg|^“dh|_“di|`“dj|a“dk|b“dl|c“dm|d“dn|e“do|f“dp|g“dq|h“dr|i“ds|j“dt|k“du|l“dv|m“dw|n“dx|o“dy|p“dz|q“d{|r“d||s“d}|t“d~|u“d|v“d€|w“d�|x“d‚|y“dƒ|z“d„|{“d…||“d†|}“d‡|~“dˆ|“d‰|€“dŠ|�“d‹|‚“dŒ|ƒ“d�|„“|‡¤Ž |…| _ |†| _ d S )�NgH¯¼šò×z>z Unsloth: Your learning rate of `zi` is too small and less than 1e-7! Consider increasing it, otherwise gradient updates will be close to 0!r9 za` is way too larger > 1! Consider decreasing it to 1e-1, otherwise gradient updates will explode!rg rh Ú unsloth_training_checkpointsrY Ú
|
|||
|
|
output_dirÚoverwrite_output_dirÚdo_trainÚdo_evalÚ
|
|||
|
|
do_predictÚ
|
|||
|
|
eval_strategyÚprediction_loss_onlyÚper_device_train_batch_sizeÚper_device_eval_batch_sizeÚper_gpu_train_batch_sizeÚper_gpu_eval_batch_sizeÚgradient_accumulation_stepsÚeval_accumulation_stepsÚ
|
|||
|
|
eval_delayÚtorch_empty_cache_stepsÚ
|
|||
|
|
learning_rateÚ weight_decayÚ
|
|||
|
|
adam_beta1Ú
|
|||
|
|
adam_beta2Ú adam_epsilonÚ
|
|||
|
|
max_grad_normÚnum_train_epochsÚ max_stepsÚlr_scheduler_typeÚ warmup_ratioÚ warmup_stepsÚ log_levelÚlog_level_replicaÚlog_on_each_nodeÚ logging_dirÚlogging_strategyÚlogging_first_stepÚ
|
|||
|
|
logging_stepsÚlogging_nan_inf_filterÚ
|
|||
|
|
save_strategyÚ
|
|||
|
|
save_stepsÚsave_total_limitÚsave_safetensorsÚsave_on_each_nodeÚsave_only_modelÚ'restore_callback_states_from_checkpointÚno_cudaÚuse_cpuÚuse_mps_deviceÚseedÚ data_seedÚ
|
|||
|
|
jit_mode_evalÚuse_ipexÚbf16Úfp16Úfp16_opt_levelÚhalf_precision_backendÚbf16_full_evalÚfp16_full_evalÚtf32Ú
|
|||
|
|
local_rankÚ ddp_backendÚ
|
|||
|
|
tpu_num_coresÚtpu_metrics_debugÚdebugÚdataloader_drop_lastÚ
|
|||
|
|
eval_stepsÚdataloader_num_workersÚdataloader_prefetch_factorÚ
|
|||
|
|
past_indexÚrun_nameÚ disable_tqdmÚremove_unused_columnsÚ label_namesÚload_best_model_at_endÚmetric_for_best_modelÚgreater_is_betterÚignore_data_skipÚfsdpÚfsdp_min_num_paramsÚ fsdp_configÚ"fsdp_transformer_layer_cls_to_wrapÚaccelerator_configÚ deepspeedÚlabel_smoothing_factorÚoptimÚ
|
|||
|
|
optim_argsÚ adafactorÚgroup_by_lengthÚlength_column_nameÚ report_toÚddp_find_unused_parametersÚddp_bucket_cap_mbÚddp_broadcast_buffersÚdataloader_pin_memoryÚ dataloader_persistent_workersÚskip_memory_metricsÚuse_legacy_prediction_loopÚ push_to_hubÚresume_from_checkpointÚ hub_model_idÚ hub_strategyÚ hub_tokenÚhub_private_repoÚhub_always_pushÚ hub_revisionÚgradient_checkpointingÚ gradient_checkpointing_kwargsÚinclude_inputs_for_metricsÚeval_do_concat_batchesÚ fp16_backendÚpush_to_hub_model_idÚpush_to_hub_organizationÚpush_to_hub_tokenÚ
|
|||
|
|
mp_parametersÚauto_find_batch_sizeÚfull_determinismÚ torchdynamoÚ ray_scopeÚ ddp_timeoutÚ
|
|||
|
|
torch_compileÚtorch_compile_backendÚtorch_compile_modeÚinclude_tokens_per_secondÚ include_num_input_tokens_seenÚneftune_noise_alphaÚoptim_target_modulesÚbatch_eval_metricsÚ
|
|||
|
|
eval_on_startÚuse_liger_kernelÚliger_kernel_configÚeval_use_gather_objectÚ average_tokens_across_devicesÚmodel_init_kwargsÚ
|
|||
|
|
max_lengthÚtruncation_modeÚoptimize_device_cache© )ÚFloatingPointErrorÚ
|
|||
|
|
OverflowErrorÚsuperÚ__init__rW rX )‰Úselfrv rw rx ry rz r{ r| r} r~ r r€ r� r‚ rƒ r„ r… r† r‡ rˆ r‰ rŠ r‹ rŒ r� rŽ r� r� r‘ r’ r“ r” r• r– r— r˜ r™ rš r› rœ r� rž rŸ r r¡ r¢ r£ r¤ r¥ r¦ r§ r¨ r© rª r« r¬ r r® r¯ r° r± r² r³ r´ rµ r¶ r· r¸ r¹ rº r» r¼ r½ r¾ r¿ rÀ rÁ r rà rÄ rÅ rÆ rÇ rÈ rÉ rÊ rË rÌ rÍ rÎ rÏ rÐ rÑ rÒ rÓ rÔ rÕ rÖ r× rØ rÙ rÚ rÛ rÜ rÝ rÞ rß rà rá râ rã rä rå ræ rç rè ré rê rë rì rí rî rï rð rñ rò ró rô rõ rö r÷ rø rù rW rX ÚkwargsÚ __class__s‰ €rO rþ z"UnslothIterativeSFTConfig.__init__Z s9 ø€ ðT ˜4Ò Ð Õ'9ð ;VÐ]jð ;Vð ;Vð ;Vñ (Wô (Wð "WØ ˜1Ò Ð ¥Mð 3FÐUbð 3Fð 3Fð 3Fñ %Gô %Gð GØ Ð -°7Ò":Ð":¸zÈSÒ?PÐ?PØ7ˆJØ ˆMà�‰ŒÔð D Dð D Dð D DØ#˜ðD Dà#7Ð#7ðD Dð �xðD Dð �gð D Dð
|
|||
|
|
$˜ð D Dð *˜Mð
|
|||
|
|
D Dð $8Ð#7ðD Dð +FÐ*EðD Dð *DÐ)CðD Dð (@Ð'?ðD Dð '>Ð&=ðD Dð +FÐ*EðD Dð '>Ð&=ðD Dð $˜ð D Dð '>Ð&=ð D Dð *˜Mð!D Dð" (˜<ð#D Dð$ $˜ð%D Dð& $˜ð'D Dð( (˜<ð)D Dð* *˜Mð+D Dð, 0Ð /ð-D Dð. "˜ ð/D Dð0 !2Ð 1ð1D Dð2 (˜<ð3D Dð4 (˜<ð5D Dð6 "˜ ð7D Dð8 !2Ð 1ð9D Dð: 0Ð /ð;D Dð< &˜+ð=D Dð> 0Ð /ð?D Dð@ "4Ð!3ðAD DðB *˜MðCD DðD &<Ð%;ðED DðF *˜MðGD DðH $˜ðID DðJ 0Ð /ðKD DðL 0Ð /ðMD DðN !2Ð 1ðOD DðP .˜oðQD DðR 7^Ð6]ðSD DðT �gðUD DðV �gðWD DðX ,˜^ðYD DðZ �4ð[D Dð\ "˜ ð]D Dð^ *˜Mð_D Dð` �xðaD Dðb �4ðcD Dðd �4ðeD Dðf ,˜^ðgD Dðh &<Ð%;ðiD Dðj ,˜^ðkD Dðl ,˜^ðmD Dðn �4ðoD Dðp $˜ðqD Dðr &˜+ðsD Dðt *˜MðuD Dðv !2Ð 1ðwD Dðx �EðyD Dðz $8Ð#7ð{D Dð| $˜ð}D Dð~ &<Ð%;ðD Dð@ *DÐ)CðAD DðB $˜ðCD DðD �xðED DðF (˜<ðGD DðH %:Ð$9ðID DðJ &˜+ðKD DðL &<Ð%;ðMD DðN %:Ð$9ðOD DðP !2Ð 1ðQD DðR 0Ð /ðSD DðT �4ðUD DðV #6Ð"5ðWD DðX &˜+ðYD DðZ 2TÐ1Sð[D Dð\ "4Ð!3ð]D Dð^ "˜ ð_D Dð` &<Ð%;ðaD Dðb �EðcD Dðd $˜ðeD Dðf "˜ ðgD Dðh .˜oðiD Dðj "4Ð!3ðkD Dðl "˜ ðmD Dðn *DÐ)CðoD Dðp !2Ð 1ðqD Dðr %:Ð$9ðsD Dðt %:Ð$9ðuD Dðv -JÐ,IðwD Dðx #6Ð"5ðyD Dðz *DÐ)Cð{D Dð| &˜+ð}D Dð~ &<Ð%;ðD Dð@ (˜<ðAD DðB (˜<ðCD DðD "˜ ðED DðF 0Ð /ðGD DðH .˜oðID DðJ (˜<ðKD DðL &<Ð%;ðMD DðN -JÐ,IðOD DðP *DÐ)CðQD DðR &<Ð%;ðSD DðT (˜<ðUD DðV $8Ð#7ðWD DðX (@Ð'?ðYD DðZ !2Ð 1ð[D Dð\ *˜Mð]D Dð^ $8Ð#7ð_D Dð` 0Ð /ðaD Dðb &˜+ðcD Dðd "˜ ðeD Dðf &˜+ðgD Dðh *˜MðiD Dðj %:Ð$9ðkD Dðl "4Ð!3ðmD Dðn )BÐ(AðoD Dðp -JÐ,IðqD Dðr #6Ð"5ðsD Dðt $8Ð#7ðuD Dðv "4Ð!3ðwD Dðx *˜MðyD Dðz 0Ð /ð{D Dð| #6Ð"5ð}D Dð~ &<Ð%;ðD Dð@ -JÐ,IðAD DðB !2Ð 1ðCD DðD $˜ðED DðF .˜oðGD DðH %:Ð$9¸FðID Dð D Dð D DðJ %9ˆÔ!Ø"4ˆÔ Ð Ð rQ )†NNFFFrY Fr5 r5 NNrZ rZ r r[ r\ r] r^ r_ r` ra rb r4 rc rd r re rf TNrg Fr9 Frg rh NTFFFFFFri ri FFFFrj rk FFNr4 NNFrl FNr Nr4 NNTNFNNFrl r NNNNrm rn NFFro NNNNTFTFFNNrp NNFNFNFTrk NNNrl TFNrq rr FNNFFNNFFFNFTNNrs FNr4 )
|
|||
|
|
Ú__name__Ú
|
|||
|
|
__module__Ú __qualname__Ú__doc__r* rW r r Ú__annotations__rX Úintrþ Ú
|
|||
|
|
__classcell__©r s @rO rS rS 3 s\ ø€ € € € € € ð ð ð: +0¨%ØØÐ 1Ð2ð+ñ +ô +И( 3œ-ð ð ñ ð */¨ØØÐ VÐWð*ñ *ô *И #œ ð ð ñ ð Ø #ØØØØ Ø $Ø&'Ø%&Ø#'Ø"&Ø&'Ø"#ØØ"%Ø ØØØØ ØØ ØØ $ØØØ Ø %Ø ØØ"Ø "ØØ!&Ø ØØ Ø Ø !Ø Ø27ØØØ ØØØ ØØØØ Ø!'Ø Ø ØØØØ Ø !ØØ $ØØ!"Ø%)ØØØØ $ØØ!&Ø $Ø Ø ØØ ØØ-1Ø !ØØ!$Ø ØØØ Ø %ØØ%)Ø Ø $Ø $Ø(-Ø "Ø%*ØØ!%ØØ#ØØ Ø ØØ!&Ø(,Ø%*Ø!%Ø Ø #Ø#'Ø ØØ #Ø ØØØØ Ø $Ø !Ø$)Ø(-Ø "Ø #Ø "Ø Ø Ø "Ø!&Ø(,Ø ØØ$Ø %Ø #Ø ðOV5ð V5ð V5ð V5ð V5ð V5ð V5ð V5ð V5ð V5rQ rS c ó” ‡ — e Zd ZdZddgZ d deeef deee e
|
|||
|
|
f dee d eee e
|
|||
|
|
ee f f d
|
|||
|
|
eeeeeef d eej j ej j j f d eeej ej gej f d
|
|||
|
|
eeege
|
|||
|
|
f fˆ fd„
|
|||
|
|
Zdede defd„Z dej dej dej fd„Z e de ej de ej de ej de e de e f
|
|||
|
|
d„¦ « Z! e"j# ¦ « d!dee ej dee ej dee ej dee e dee e f
|
|||
|
|
d„¦ « Z$d„ Z%ˆ fd„Z& d"d ee d ee d eee e df fd „Z'ˆ xZ(S )#Ú_UnslothIterativeSFTTrainerrl Útrlz
|
|||
|
|
iterative-sftN©NNÚmodelÚargsÚ
|
|||
|
|
data_collatorÚ eval_datasetÚprocessing_classÚ
|
|||
|
|
optimizersÚ preprocess_logits_for_metricsÚcompute_metricsc
|
|||
|
|
óê •— t |t ¦ « r|n |j j } |€.| d¦ « d }
|
|||
|
|
t |
|
|||
|
|
› d�¦ « }nit |t ¦ « rTt |t
|
|||
|
|
¦ « s?| ¦ « } |j | d< | d¦ « t di | ¤Ž}|€t j | ¦ « }|j �)t |t ¦ « st j d¦ « t |t ¦ « r| ||¦ « }t! ¦ « r t |t" ¦ « rd| _ nd| _ || _ t) |j d d¦ « | _ |€;| j rt- |d
|
|||
|
|
d ¬ ¦ « | _ n#t1 | j d¬
|
|||
|
|
¦ « | _ n|| _ |j | _ |j | _ |j | _ t9 ¦ « ||| j |||||¬¦ « t= | j d¦ « r | j | j! ¦ « | " | j# j$ ¦ « | j% & | j | j' | j( ¦ « \ | _ | _' | _( | j dk rdnd| j _) t= | d¦ « stU d¦ « ‚| j tV _ d S )Nú/r4 z
|
|||
|
|
-IterativeSFTr× râ zŠYou passed model_init_kwargs to the `IterativeSFTConfig`, but your model is already instantiated. The `model_init_kwargs` will be ignored.TFÚis_encoder_decoderéœÿÿÿé )Úlabel_pad_token_idÚpad_to_multiple_of)Úmlm)r r r r r r r r Úadd_model_tagsrs ÚleftÚrightÚ acceleratorzXYour `Trainer` does not have an `accelerator` object. Consider upgrading `transformers`.rú ),Ú
|
|||
|
|
isinstanceÚstrÚconfigÚ
|
|||
|
|
_name_or_pathÚsplitr r Úto_dictr× Úpopr Úfrom_pretrainedrö r' ÚwarnÚ_create_model_from_pathr# r Ú
|
|||
|
|
is_peft_modelr Úgetattrr r r r r÷ rø rù rý rþ Úhasattrr r Ú
|
|||
|
|
_tag_namesÚ create_optimizer_and_schedulerr rŒ r! ÚprepareÚ optimizerÚ lr_schedulerÚtruncation_sideÚAttributeErrorr )
|
|||
|
|
rÿ r r r r r r r r Úmodel_idÚ
|
|||
|
|
model_nameÚ dict_argsr s
|
|||
|
|
€rO rþ z$_UnslothIterativeSFTTrainer.__init__x s ø€ õ" ' ucÑ 2Ô 2ÐR�5�5¸¼ Ô8RˆØ ˆ<Ø!Ÿ š ¨Ñ,Ô,¨RÔ0ˆJÝ%¨Ð&BÐ&BÐ&BÑCÔCˆDˆDÝ
|
|||
|
|
˜Õ /Ñ
|
|||
|
|
0Ô
|
|||
|
|
0ð 3½ÀDÕJ\Ñ9]Ô9]ð 3Ø Ÿ š ™œˆIØ%)¤^ˆI�kÑ "Ø �MŠMÐ-Ñ .Ô .Ð .Ý%Ð2Ð2¨ Ð2Ð2ˆDð Ð #Ý ,Ô <¸XÑ FÔ FÐ ð Ô !Ð -µjÀÍÑ6LÔ6LÐ -Ý ŒMð;ñ
|
|||
|
|
ô
|
|||
|
|
ð
|
|||
|
|
õ �e�SÑ !Ô !ð >Ø×0Ò0°¸Ñ=Ô=ˆEõ Ñ Ô ð '¥:¨eµYÑ#?Ô#?ð 'Ø!%ˆDÔ Ð à!&ˆDÔ à 0ˆÔ Ý")¨%¬,Ð8LÈeÑ"TÔ"TˆÔ à Ð ØÔ&ð
|
|||
|
|
gÝ%;Ø$¸ÐRSð&ñ &ô &�Ô"Ð"õ &EÀTÔEZÐ`eÐ%fÑ%fÔ%f�Ô"Ð"à!.ˆDÔ à œ/ˆŒØ #Ô 3ˆÔ Ø%)Ô%?ˆÔ"å
|
|||
|
|
‰Œ×ÒØØØ Ô,Ø%Ø -Ø +Ø!Ø*Gð ñ
|
|||
|
|
ô
|
|||
|
|
ð
|
|||
|
|
õ �4”:Ð /Ñ 0Ô 0ð 7Ø ŒJ× %Ò % d¤oÑ 6Ô 6Ð 6à ×+Ò+¨D¬IÔ,?Ñ@Ô@Ð@ð 9=Ô8H×8PÒ8PØ ŒJ˜œ¨Ô(9ñ9
|
|||
|
|
ô 9
|
|||
|
|
Ñ5ˆŒ
|
|||
|
|
�D”N DÔ$5ð ;?Ô:NÐR\Ò:\Ð:\°°ÐbiˆÔ Ô-å�t˜]Ñ+Ô+ð Ý Øjñô ð
|
|||
|
|
ð /3Ô.H�
|
|||
|
|
Ô+Ð+Ð+rQ Ú
|
|||
|
|
model_pathÚreturnc ó8 — |j pi }t j |fi |¤ŽS )z0Creates a model from a path or model identifier.)rö r
|
|||
|
|
r) )rÿ r9 r rö s rO r+ z3_UnslothIterativeSFTTrainer._create_model_from_pathÖ s* € à Ô 2Ð 8°bÐÝ#Ô3°JÐTÐTÐBSÐTÐTÐTrQ Ú input_idsÚattention_maskÚlabelsc ó´ ‡ — |€ d„ |D ¦ « }‰ j r�‰ d„ t |||¦ « D ¦ « ¦ « ‰ j j ¦ « }| dd ¦ « d|d |d ‰ j j k < nJ‰ d„ t ||¦ « D ¦ « ¦ « ‰ j j ¦ « }‰ j �o‰ j
|
|||
|
|
dk r!ˆ fd„| ¦ « D ¦ « }nC‰ j
|
|||
|
|
d k r!ˆ fd
|
|||
|
|
„| ¦ « D ¦ « }nt d ‰ j
|
|||
|
|
› �¦ « ‚|S ) Nc ó6 — g | ]}t j |¦ « ‘ŒS rú )r& Ú ones_like)Ú.0Úidss rO ú
|
|||
|
|
<listcomp>zD_UnslothIterativeSFTTrainer.prepare_model_inputs.<locals>.<listcomp>Ý s" € Ð HÐ HÐ H°s�eœo¨cÑ 2Ô 2Ð HÐ HÐ HrQ c ó" — g | ] \ }}}|||d œ‘Œ
|
|||
|
|
S )©r< r= r> rú )rB rC ÚattÚlabs rO rD zD_UnslothIterativeSFTTrainer.prepare_model_inputs.<locals>.<listcomp>á s8 € ð ð ð á%˜˜S #ð #&¸ÈÐLÐLðð ð rQ Údecoder_input_idsr r> c ó — g | ]
|
|||
|
|
\ }}||d œ‘Œ S ))r< r= rú )rB rC rG s rO rD zD_UnslothIterativeSFTTrainer.prepare_model_inputs.<locals>.<listcomp>í s$ € ÐjÐjÐj¹x¸sÀC˜s°cÐ:Ð:ÐjÐjÐjrQ Ú
|
|||
|
|
keep_startc ó6 •— i | ]\ }}||d ‰j … “ŒS ©N©r÷ ©rB ÚkÚvrÿ s €rO ú
|
|||
|
|
<dictcomp>zD_UnslothIterativeSFTTrainer.prepare_model_inputs.<locals>.<dictcomp>ó s, ø€ Ð UÐ UÐ U¹$¸!¸Q˜a Ð#4 T¤_Ð#4Ô!5Ð UÐ UÐ UrQ rs c ó8 •— i | ]\ }}||‰j d … “ŒS rM rN rO s €rO rR zD_UnslothIterativeSFTTrainer.prepare_model_inputs.<locals>.<dictcomp>õ s/ ø€ Ð VÐ VÐ V¹4¸1¸a˜a D¤OÐ#3Ð#5Ð#5Ô!6Ð VÐ VÐ VrQ zUnknown truncation mode: )
|
|||
|
|
r r r= r> r Údevicer( r Ú pad_token_idr÷ rø ÚitemsÚ
|
|||
|
|
ValueError)rÿ r< r= r> Ú
|
|||
|
|
input_datas ` rO Úprepare_model_inputsz0_UnslothIterativeSFTTrainer.prepare_model_inputsÛ s� ø€ Ø Ð !Ø HÐ H¸iÐ HÑ HÔ HˆNà Ô "ð $Ø ×+Ò+ðð å),¨Y¸ÈÑ)OÔ)Oðñ ô ñô ÷
|
|||
|
|
Šb�”Ô"Ñ#Ô#ð
|
|||
|
|
ð
|
|||
|
|
�NŠNÐ.°Ñ 5Ô 5Ð 5à_cˆJ�xÔ ¨HÔ!5¸Ô9NÔ9[Ò![Ñ \Ð \ð ×+Ò+ØjÐjÍ3ÈyÐZhÑKiÔKiÐjÑjÔjñô çŠb�”Ô"Ñ#Ô#ð
|
|||
|
|
ð
|
|||
|
|
Œ?Ð &ØÔ# |Ò3Ð3Ø UÐ UÐ UÐ UÀ*×BRÒBRÑBTÔBTÐ UÑ UÔ U�
|
|||
|
|
�
|
|||
|
|
ØÔ%¨Ò3Ð3Ø VÐ VÐ VÐ VÀ:×CSÒCSÑCUÔCUÐ VÑ VÔ V�
|
|||
|
|
�
|
|||
|
|
å Ð!S¸TÔ=QÐ!SÐ!SÑTÔTÐTàÐrQ ÚtextsÚ texts_labelsc
|
|||
|
|
óV — |�€6|€št ddg| |g¦ « D ]ƒ\ }}t |t ¦ « s!t |› dt |¦ « › �¦ « ‚t |d t
|
|||
|
|
j ¦ « s(t d|› dt |d ¦ « › �¦ « ‚Œ„�n„t g d¢| ||g¦ « D ]ƒ\ }}t |t ¦ « s!t |› dt |¦ « › �¦ « ‚t |d t
|
|||
|
|
j ¦ « s(t d|› dt |d ¦ « › �¦ « ‚Œ„nêt |t ¦ « s t d t |¦ « › �¦ « ‚t |d t ¦ « s%t d
|
|||
|
|
t |d ¦ « › �¦ « ‚|�tt |t ¦ « s t d t |¦ « › �¦ « ‚t |d t ¦ « s%t d t |d ¦ « › �¦ « ‚| ||||fS )
|
|||
|
|
aŠ
|
|||
|
|
Check if the input data is valid for training.
|
|||
|
|
|
|||
|
|
Args:
|
|||
|
|
input_ids (list[`torch.LongTensor`]):
|
|||
|
|
List of tensors containing the input_ids
|
|||
|
|
attention_mask (list[`torch.LongTensor`]):
|
|||
|
|
List of tensors containing the attention_mask
|
|||
|
|
labels (list[`torch.FloatTensor`]):
|
|||
|
|
List of tensors containing the labels
|
|||
|
|
texts (list[`str`]):
|
|||
|
|
List of string containing the text input.
|
|||
|
|
texts_labels (list[`str`]):
|
|||
|
|
List of string containing the text labels.
|
|||
|
|
|
|||
|
|
Returns:
|
|||
|
|
`tuple`: The input data.
|
|||
|
|
Nr< r> z! must be a list of tensors - got r z Elements in z must be tensors - got rF z''text' must be a list of strings - got z)Elements in 'text' must be strings - got z.'text_labels' must be a list of strings - got z0Elements in 'text_labels' must be strings - got )r= r" ÚlistrW Útyper& r r# )r< r= r> rZ r[ ÚnameÚ tensor_lists rO Ú_step_safety_checkerz0_UnslothIterativeSFTTrainer._step_safety_checkerû s‚ € ð4 ‰=Ø Ð%Ý),¨k¸8Ð-DÀyÐRXÐFYÑ)ZÔ)Zð mð mÑ%�D˜+Ý% kµ4Ñ8Ô8ð hÝ (¨DÐ)fÐ)fÕSWÐXcÑSdÔSdÐ)fÐ)fÑ gÔ gÐgÝ% k°!¤nµe´lÑCÔCð mÝ (Ð)k¸Ð)kÐ)kÕUYÐZeÐfgÔZhÑUiÔUiÐ)kÐ)kÑ lÔ lÐlðmñmõ *-Ø=Ð=Ð=À È>Ð[aÐ?bñ*ô *ð mð mÑ%�D˜+õ & kµ4Ñ8Ô8ð hÝ (¨DÐ)fÐ)fÕSWÐXcÑSdÔSdÐ)fÐ)fÑ gÔ gÐgÝ% k°!¤nµe´lÑCÔCð mÝ (Ð)k¸Ð)kÐ)kÕUYÐZeÐfgÔZhÑUiÔUiÐ)kÐ)kÑ lÔ lÐlðmð mõ ˜e¥TÑ*Ô*ð
|
|||
|
|
ZÝ Ð!XÍ4ÐPUÉ;Ì;Ð!XÐ!XÑYÔYÐYÝ ˜e AœhÑ,Ô,ð
|
|||
|
|
_Ý Ð!]ÍTÐRWÐXYÔRZÉ^Ì^Ð!]Ð!]Ñ^Ô^Ð^ØÐ'Ý! ,µÑ5Ô5ð lÝ$Ð%jÕVZÐ[gÑVhÔVhÐ%jÐ%jÑkÔkÐkÝ! ,¨q¤/µ3Ñ7Ô7ð qÝ$Ð%oÕX\Ð]iÐjkÔ]lÑXmÔXmÐ%oÐ%oÑpÔpÐpà˜.¨&°%¸ ÐEÐErQ c ó† ‡ ‡— ‰ j ¦ « ‰ j j dk rGt j d¦ « ‰ j j ¦ « ‰ _ ‰ j j ‰ _
|
|||
|
|
|€|€t d¦ « ‚|� |�t j
|
|||
|
|
dt ¦ « |€|€‰ j rt d¦ « ‚|�
|
|||
|
|
|dd… nd}|�
|
|||
|
|
|dd… nd}|�
|
|||
|
|
|dd… nd}|�
|
|||
|
|
|dd… nd}|�
|
|||
|
|
|dd… nd}‰ |||||¦ « \ }}}}}|�/‰ |‰ j ddd¬ ¦ « }|d
|
|||
|
|
|d }}|�%‰ |‰ j ddd¬ ¦ « d
|
|||
|
|
}|€|}‰ |||¦ « }t) | ¦ « ¦ « }i }| |¦ « ˆ fd „} t/ j |¦ « }
|
|||
|
|
|
|
|||
|
|
d
|
|||
|
|
¦ « t5 |
|
|||
|
|
‰ j j d| ¬¦ « } t9 | ¦ « D �]—\ } Š‰ j ‰ j ¦ « 5 ˆfd„|D ¦ « }‰ ‰ j |¦ « }
|
|||
|
|
‰ j j dk r|
|
|||
|
|
! ¦ « }
|
|||
|
|
|
|
|||
|
|
" ¦ « }‰ j # |
|
|||
|
|
¦ « ‰ j j$ rH‰ j j% �<‰ j & ‰ j ' ¦ « ‰ j j% ¦ « ‰ j( ) ¦ « ‰ j( * ¦ « ‰ j+ �‰ j+ ) ¦ « ‰ j xj dz
|
|||
|
|
c_ ‰ xj |z
|
|||
|
|
c_ ‰ , ¦ « ddd¦ « n # 1 swxY w Y �Œ™dS )aÐ
|
|||
|
|
Run an optimisation step given a list of input_ids, attention_mask, and labels or a list of text and
|
|||
|
|
text_labels.
|
|||
|
|
|
|||
|
|
Args:
|
|||
|
|
input_ids (list[`torch.LongTensor`]):
|
|||
|
|
List of tensors containing the input_ids (if not provided, text will be used)
|
|||
|
|
attention_mask (list[`torch.LongTensor`], , *optional*):
|
|||
|
|
List of tensors containing the attention_mask
|
|||
|
|
labels (list[`torch.FloatTensor`], *optional*):
|
|||
|
|
List of tensors containing the labels (if set to None, will default to input_ids)
|
|||
|
|
texts (list[`str`], *optional*):
|
|||
|
|
List of strings containing the text input (if not provided, input_ids will directly be used)
|
|||
|
|
texts_labels (list[`str`], *optional*):
|
|||
|
|
List of strings containing the text labels (if set to None, will default to text)
|
|||
|
|
|
|||
|
|
Returns:
|
|||
|
|
`dict[str, Any]`: A summary of the training statistics
|
|||
|
|
r rm Nz@Step should include `input_ids` or `texts` as keyword arguments.ztBoth `input_ids` and `texts` argument are provided. `input_ids` will be ignored. Please provide only one of the two.z€No 'labels' or 'text_labels' are provided. When using an encoder-decoder architecture, 'labels' or 'text_labels' must be passed.TÚpt)r÷ Ú
|
|||
|
|
truncationÚpaddingÚreturn_tensorsr< r= c óÄ •‡— t ¦ « }| d D ]FŠ‰dv r@t j ˆfd„| D ¦ « ¦ « ‰j j ¦ « |‰< ŒG|S )Nr rF c ó •— g | ]
|
|||
|
|
}|‰ ‘Œ S rú rú )rB ÚdÚkeys €rO rD zF_UnslothIterativeSFTTrainer.step.<locals>.collator.<locals>.<listcomp>† s ø€ Ð3IÐ3IÐ3I¸q°A°c´FÐ3IÐ3IÐ3IrQ )Údictr& Ústackr> r rT )ÚdataÚ return_dictrj rÿ s @€rO Úcollatorz2_UnslothIterativeSFTTrainer.step.<locals>.collator‚ sq øø€ Ý ™&œ&ˆKؘA”wð
|
|||
|
|
að
|
|||
|
|
a�ØÐCÐCÐCÝ',¤{Ð3IÐ3IÐ3IÐ3IÀDÐ3IÑ3IÔ3IÑ'JÔ'J×'MÒ'MÈdÌjÔN_Ñ'`Ô'`�K Ñ$øØ Ð rQ r& )Ú
|
|||
|
|
batch_sizeÚshuffleÚ
|
|||
|
|
collate_fnc ó" •— i | ] }|‰| “Œ S rú rú )rB rP Úbatchs €rO rR z4_UnslothIterativeSFTTrainer.step.<locals>.<dictcomp>• s ø€ Ð HÐ HÐ H° 5¨¤8Ð HÐ HÐ HrQ r9 )-r ÚtrainÚstateÚ global_stepr& Útensorr> r rT Útr_lossÚ_globalstep_last_loggedrW r' r* Ú UserWarningr ra r r÷ rY r] ÚkeysÚupdater Ú from_dictÚ
|
|||
|
|
set_formatr r} Ú enumerater! Ú
|
|||
|
|
accumulateÚ compute_lossÚn_gpuÚmeanÚdetachÚbackwardÚsync_gradientsrŠ Úclip_grad_norm_Ú
|
|||
|
|
parametersr2 ÚstepÚ zero_gradr3 Ú_maybe_log_save_evaluate)rÿ r< r= r> rZ r[ Ú model_inputsÚmodel_inputs_namesÚ
|
|||
|
|
batch_dictro Ú
|
|||
|
|
batch_dataÚstep_dataloaderÚ_ÚlossÚ tr_loss_steprt s ` @rO rŠ z _UnslothIterativeSFTTrainer.step1 sq øø€ ð8
|
|||
|
|
Œ
|
|||
|
|
×ÒÑÔÐà Œ:Ô ! QÒ &Ð &Ý œ<¨Ñ,Ô,×/Ò/°´ Ô0@ÑAÔAˆDŒLØ+/¬:Ô+AˆDÔ (à Ð Ý Ð _Ñ`Ô`Ð `Ø
|
|||
|
|
Ð
|
|||
|
|
" uÐ'8Ý ŒMð6åñ
|
|||
|
|
ô
|
|||
|
|
ð
|
|||
|
|
ð ˆ>˜lÐ 2°tÔ7NÐ 2Ý ð Sñô ð
|
|||
|
|
ð
|
|||
|
|
%.Ð$9�I˜a˜a˜a”L�L¸tˆ Ø.<Ð.H˜ ¨¨¨Ô*Ð*ÈdˆØ $Ð 0�˜˜˜”�°dˆØ !Ð -��a�a�a”�°4ˆØ*6Ð*B�| A A A”�Ȉ àAE×AZÒAZØ �~ v¨u°lñB
|
|||
|
|
ô B
|
|||
|
|
Ñ>ˆ �> 6¨5°,ð Ð Ø ×0Ò0Ø $¤/¸dÈDÐaeð 1ñ ô ˆLð )5°[Ô(AÀ<ÐP`ÔCa�~ˆIà Ð #Ø×*Ò*Ø $¤/¸dÈDÐaeð +ñ ô àôˆFð ˆ>Ø ˆFà×0Ò0°¸NÈFÑSÔSˆ å ! ,×"3Ò"3Ñ"5Ô"5Ñ 6Ô 6Ðàˆ
|
|||
|
|
Ø×Ò˜,Ñ'Ô'Ð'ð ð ð ð ð õ Ô& zÑ2Ô2ˆ
|
|||
|
|
Ø× Ò ˜gÑ&Ô&Ð&å$Ø Ø”yÔ<ØØ ð
|
|||
|
|
ñ
|
|||
|
|
ô
|
|||
|
|
ˆõ " /Ñ2Ô2ð 0ñ 0‰HˆAˆuØÔ!×,Ò,¨T¬ZÑ8Ô8ð
|
|||
|
|
0ð
|
|||
|
|
0Ø HÐ HÐ HÐ HÐ5GÐ HÑ HÔ H� Ø×(Ò(¨¬°\ÑBÔB�à”9”? QÒ&Ð&Ø Ÿ9š9™;œ;�Dà #Ÿ{š{™}œ}� àÔ ×)Ò)¨$Ñ/Ô/Ð/àÔ#Ô2ð °t´yÔ7NÐ7ZØÔ$×4Ò4Ø œ
|
|||
|
|
×-Ò-Ñ/Ô/Ø œ Ô/ñô ð ð
|
|||
|
|
”×#Ò#Ñ%Ô%Ð%Ø”×(Ò(Ñ*Ô*Ð*ØÔ$Ð0ØÔ%×*Ò*Ñ,Ô,Ð,à”
|