261 lines
27 KiB
Plaintext
261 lines
27 KiB
Plaintext
|
o
|
|||
|
|
9—°h$¤ ã @ sº d Z ddlmZ ddlZddlmZ ddlmZ ddlmZm Z m
|
|||
|
|
Z
|
|||
|
|
m Z m Z m
|
|||
|
|
Z
|
|||
|
|
mZmZ ddlmZmZmZmZmZmZmZmZmZmZmZmZm Z m
|
|||
|
|
Z
|
|||
|
|
m Z m Z m Z m Z m!Z!m"Z"m#Z#m$Z$m Z m%Z%m&Z&m'Z'm(Z(m)Z)mZm*Z*m
|
|||
|
|
Z
|
|||
|
|
m Z m Z m#Z#m'Z'm)Z)mZ ddl)Z)ddlT ddl+m,Z,m-Z- dd l.m/Z/ ddlZddl0Z1dd
|
|||
|
|
l2m3Z3 ddlmZ dd l4mZmZ5 d d
|
|||
|
|
d d
|
|||
|
|
d
|
|||
|
|
dœZ6ej7d d e6d�dd„ ƒZ8e,G dd„ deƒƒZ9 G dd„ de#ƒZ:G dd„ de:ƒZ;dS )z9
|
|||
|
|
2025.8.9
|
|||
|
|
2025.8.10
|
|||
|
|
4.55.4
|
|||
|
|
0.21.0
|
|||
|
|
__UNSLOTH_VERSIONING__
|
|||
|
|
é )ÚTensorN)Ú
|
|||
|
|
functional)ÚAnyÚListÚOptionalÚTupleÚUnionÚDictÚSetÚCallable)%ÚAutoModelForCausalLMÚ
|
|||
|
|
AutoTokenizerÚBaseImageProcessorr Ú DataCollatorÚ DataCollatorForLanguageModelingÚDataCollatorForSeq2SeqÚ
|
|||
|
|
DataLoaderÚDatasetÚEvalLoopOutputÚFeatureExtractionMixinÚIterativeSFTConfigÚIterativeSFTTrainerr Ú
|
|||
|
|
PPODecoratorsÚPathÚ PeftModelÚPreTrainedModelÚPreTrainedTokenizerBaseÚProcessorMixinÚTrainerÚTrainingArgumentsr Úgenerate_model_cardÚget_comet_experiment_urlÚis_peft_availableÚis_wandb_availableÚosÚtorchÚwarningsr r r r r" r$ r% )Ú*)Ú dataclassÚfield)ÚVersion)Ú nullcontext)r r TF)Úepilogue_fusionÚ max_autotuneÚ
|
|||
|
|
shape_paddingz
|
|||
|
|
trace.enabledztriton.cudagraphs)ÚdynamicÚ fullgraphÚoptionsc
|
|||
|
|
C s¾ t j| d| jd ¡ddd�}t j| d¡ddd�}g }t||ƒD ](\}}| t j¡}t j|d| d¡d� d¡}t j
|
|||
|
|
|dd�}|| } | | ¡ q! t |¡}| | jd | jd f¡}|S )Néÿÿÿÿé r )ÚchunksÚdim)r5 Úindex)r5 é )
|
|||
|
|
r% ÚchunkÚreshapeÚshapeÚzipÚtoÚfloat32ÚgatherÚ unsqueezeÚsqueezeÚ logsumexpÚappendÚconcat)
|
|||
|
|
Úlogitsr6 Úchunked_logitsÚ
|
|||
|
|
chunked_indexÚall_per_token_logpsÚ chunk_logitsÚ chunk_indexÚselected_logitsÚlogsumexp_valuesÚper_token_logps© rM úZ/workspace/DS-LLM-TEMPLATE-FINETUNING/unsloth_compiled_cache/UnslothIterativeSFTTrainer.pyÚ chunked_selective_log_softmax" s
|
|||
|
|
rO c s† e Zd ZU dZedddid�Zee ed< edddid�Z ee
|
|||
|
|
ed < eddd
|
|||
|
|
id�Z ee
|
|||
|
|
ed <
|
|||
|
|
! ! " # $ $ % & ' ( # $ ) * + d.‡ fd,d-„ Z ‡ Z
|
|||
|
|
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_paramsr2 z8Chunk size to reduce memory usage. -1 is most efficient.Úunsloth_num_chunksz'Maximum sequence length to truncate to.Úmax_seq_lengthFÚnor3 é r éú ç-C ëâ6
|
|||
|
|
?ç{®Gáz„?çÍÌÌÌÌÌì?ç+‡ÙÎ÷ï?ç:Œ0âŽyE>ç ð?ç @Úlinearçš™™™™™¹?ÚpassiveÚwarningTÚstepsr7 éô éO
|
|||
|
|
ÚO1ÚautoÚ ç Ú
|
|||
|
|
adamw_8bitÚlengthÚ
|
|||
|
|
every_saveÚlasté Úkeep_endcˆ ‰ s� |dk r t d|› d�ƒ‚|dkrtd|› d�ƒ‚|d u r(|#dkr(|$dkr(d}d }#tƒ jdŽi d
|
|||
|
|
|“d |“d |“d
|
|||
|
|
|“d|“d|“d|“d|“d| “d|
|
|||
|
|
“d| “d| “d|
|
|||
|
|
“d|“d|“d|“d|“d|“d |“d |“d |“d |“d |“d!|“d"|“d#|“d$|“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�|„“|ˆ¤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!r7 za` is way too larger > 1! Consider decreasing it to 1e-1, otherwise gradient updates will explode!re rf Ú unsloth_training_checkpointsrW 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_cacherM )ÚFloatingPointErrorÚ
|
|||
|
|
OverflowErrorÚsuperÚ__init__rT rU rV )‰Úselfrs rt ru rv 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 rT rU rV Úkwargs©Ú __class__rM rN rú ^ s: ÿþýüûúùø ÷
|
|||
|
|
ö õ ô
|
|||
|
|
óòñðïîíìëêéèçæå ä ã â á à!ß"Þ#Ý$Ü%Û&Ú'Ù(Ø)×*Ö+Õ,Ô-Ó.Ò/Ñ0Ð1Ï2Î3Í4Ì5Ë6Ê7É8È9Ç:Æ;Å<Ä=Ã>Â?Á@ÀA¿B¾C½D¼E»FºG¹H¸I·J¶KµL´M³N²O±P°Q¯R®ST¬U«VªW©X¨Y§Z¦[¥\¤]£^¢_¡` aŸbžc�dœe›fšg™h˜i—j–k•l”m“n’o‘p�q�rŽs�tŒu‹vŠw‰xˆy‡z†{…|„}ƒ~‚� �ÿ �þ �ý �ü �û
|
|||
|
|
z"UnslothIterativeSFTConfig.__init__)‡NNFFFrW Fr3 r3 NNrX rX r rY rZ r[ r\ r] r^ r_ r` r2 ra rb r rc rd TNre Fr7 Fre rf NTFFFFFFrg rg FFFFrh ri FFNr2 NNFrj FNr Nr2 NNTNFNNFrj r NNNNrk rl NFFrm NNNNTFTFFNNrn NNFNFNFTri NNNrj TFNro rp FNNFFNNFFFNFTNNrq FNr2 N)Ú__name__Ú
|
|||
|
|
__module__Ú __qualname__Ú__doc__r) rT r r Ú__annotations__rU ÚintrV rú Ú
|
|||
|
|
__classcell__rM rM rý rN rP 3 s.
|
|||
|
|
þþþ �÷rP c sâ e Zd ZdZddgZ d(deeef deee e
|
|||
|
|
f dee d eee e
|
|||
|
|
ee f f d
|
|||
|
|
eeeeeef d eejjejjjf d eeejejgejf d
|
|||
|
|
eeege
|
|||
|
|
f f‡ fdd„
|
|||
|
|
Zdede defdd„Z dejdejdejfdd„Z e de ej de ej de ej de e de e f
|
|||
|
|
dd „ƒZ!e" #¡ d)dee ej dee ej dee ej dee e dee e f
|
|||
|
|
d d „ƒZ$d d „ Z%‡ fd!d"„Z& d*d#ee d$ee d%eee e df fd&d'„Z'‡ Z(S )+Ú_UnslothIterativeSFTTrainerrj Útrlz
|
|||
|
|
iterative-sftN©NNÚmodelÚargsÚ
|
|||
|
|
data_collatorÚ eval_datasetÚprocessing_classÚ
|
|||
|
|
optimizersÚ preprocess_logits_for_metricsÚcompute_metricsc
|
|||
|
|
sà t |tƒr|n|jj} |d u r | d¡d }
|
|||
|
|
t|
|
|||
|
|
› d�ƒ}n t |tƒr=t |tƒs=| ¡ } |j| d< | d¡ tdi | ¤Ž}|d u rFt
|
|||
|
|
| ¡}|j d urUt |tƒsUt
|
|||
|
|
d¡ t |tƒr`| ||¡}tƒ rlt |tƒrld| _nd| _|| _t|jd dƒ| _|d u r“| jrŠt|d
|
|||
|
|
d d �| _n t| jdd
|
|||
|
|
�| _n|| _|j| _|j| _|j| _t ƒ j ||| j|||||d� t | j dƒr¾| j | j!¡ | "| j#j$¡ | j% &| j | j'| j(¡\| _ | _'| _(| jdkrÝdnd| j_)t | dƒsêt*dƒ‚| jt+_d S )Nú/r2 z
|
|||
|
|
-IterativeSFTrÔ rß zŠYou passed model_init_kwargs to the `IterativeSFTConfig`, but your model is already instantiated. The `model_init_kwargs` will be ignored.TFÚis_encoder_decoderéœÿÿÿé )Úlabel_pad_token_idÚpad_to_multiple_of)Úmlm)r r
|
|||
|
|
r r r
|
|||
|
|
r r r Úadd_model_tagsrq ÚleftÚrightÚ acceleratorzXYour `Trainer` does not have an `accelerator` object. Consider upgrading `transformers`.rM ),Ú
|
|||
|
|
isinstanceÚstrÚconfigÚ
|
|||
|
|
_name_or_pathÚsplitr r Úto_dictrÔ Úpopr
|
|||
|
|
Úfrom_pretrainedró r& ÚwarnÚ_create_model_from_pathr" r Ú
|
|||
|
|
is_peft_modelr
|
|||
|
|
Úgetattrr r r r rô rõ rö rù rú Úhasattrr r Ú
|
|||
|
|
_tag_namesÚ create_optimizer_and_schedulerr
|
|||
|
|
r‰ r ÚprepareÚ optimizerÚ lr_schedulerÚtruncation_sideÚAttributeErrorr ) rû r r
|
|||
|
|
r r r
|
|||
|
|
r r r Úmodel_idÚ
|
|||
|
|
model_nameÚ dict_argsrý rM rN rú ~ sl
|
|||
|
|
|
|||
|
|
|
|||
|
|
ÿ
|
|||
|
|
|
|||
|
|
ÿø ÿ
|
|||
|
|
ÿ z$_UnslothIterativeSFTTrainer.__init__Ú
|
|||
|
|
model_pathÚreturnc C s |j pi }tj|fi |¤ŽS )z0Creates a model from a path or model identifier.)ró r r# )rû r3 r
|
|||
|
|
ró rM rM rN r% Ü s
|
|||
|
|
z3_UnslothIterativeSFTTrainer._create_model_from_pathÚ input_idsÚattention_maskÚlabelsc sò |d u r dd„ |D ƒ}ˆ j r4ˆ dd„ t|||ƒD ƒ¡ ˆ jj¡}| dd ¡ d|d |d ˆ jjk< nˆ dd„ t||ƒD ƒ¡ ˆ jj¡}ˆ j d urwˆ j
|
|||
|
|
dkr]‡ fd d
|
|||
|
|
„| ¡ D ƒ}|S ˆ j
|
|||
|
|
d kro‡ fd d
|
|||
|
|
„| ¡ D ƒ}|S t d
|
|||
|
|
ˆ j
|
|||
|
|
› �ƒ‚|S )Nc S s g | ]}t |¡‘qS rM )r% Ú ones_like)Ú.0ÚidsrM rM rN Ú
|
|||
|
|
<listcomp>ã ó zD_UnslothIterativeSFTTrainer.prepare_model_inputs.<locals>.<listcomp>c S s g | ] \}}}|||d œ‘qS )©r5 r6 r7 rM )r9 r: ÚattÚlabrM rM rN r; ç s
|
|||
|
|
ÿÿÚdecoder_input_idsr r7 c S s g | ] \}}||d œ‘qS ))r5 r6 rM )r9 r: r> rM rM rN r; ó s Ú
|
|||
|
|
keep_startc s i | ] \}}||d ˆ j … “qS ©N©rô ©r9 ÚkÚv©rû rM rN Ú
|
|||
|
|
<dictcomp>ù s zD_UnslothIterativeSFTTrainer.prepare_model_inputs.<locals>.<dictcomp>rq c s" i | ]
|
|||
|
|
\}}||ˆ j d … “qS rB rC rD rG rM rN rH û s " zUnknown truncation mode: )
|
|||
|
|
r r r; r< r Údevicer" r
|
|||
|
|
Ú pad_token_idrô rõ ÚitemsÚ
|
|||
|
|
ValueError)rû r5 r6 r7 Ú
|
|||
|
|
input_datarM rG rN Úprepare_model_inputsá s4
|
|||
|
|
þÿ
|
|||
|
|
û ÿ
|
|||
|
|
þ
|
|||
|
|
|
|||
|
|
|
|||
|
|
ûþz0_UnslothIterativeSFTTrainer.prepare_model_inputsÚtextsÚ texts_labelsc C s| |du rs|du r=t ddg| |gƒD ]*\}}t|tƒs%t|› dt|ƒ› �ƒ‚t|d tjƒs;td|› dt|d ƒ› �ƒ‚qnzt g d¢| ||gƒD ]*\}}t|tƒs[t|› dt|ƒ› �ƒ‚t|d tjƒsqtd|› dt|d ƒ› �ƒ‚qGnDt|tƒs�td t|ƒ› �ƒ‚t|d tƒs“td
|
|||
|
|
t|d ƒ› �ƒ‚|dur·t|tƒs¥td t|ƒ› �ƒ‚t|d tƒs·td t|d ƒ› �ƒ‚| ||||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.
|
|||
|
|
Nr5 r7 z! must be a list of tensors - got r z Elements in z must be tensors - got r= 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 ÚlistrL Útyper% r r )r5 r6 r7 rO rP ÚnameÚ tensor_listrM rM rN Ú_step_safety_checker s8
|
|||
|
|
ÿý ÿ
|
|||
|
|
ÿû
|
|||
|
|
|
|||
|
|
z0_UnslothIterativeSFTTrainer._step_safety_checkerc sæ ˆj ¡ ˆjjdkrt d¡ ˆjj¡ˆ_ ˆjjˆ_
|
|||
|
|
|du r'|du r't dƒ‚|dur5|dur5t
|
|||
|
|
dt¡ |du rD|du rDˆjrDt dƒ‚|durN|dd… nd}|durZ|dd… nd}|durf|dd… nd}|durr|dd… nd}|dur~|dd… nd}ˆ |||||¡\}}}}}|dur¦ˆj|ˆjdddd �}|d
|
|||
|
|
|d }}|dur·ˆj|ˆjdddd �d
|
|||
|
|
}|du r½|}ˆ |||¡}t| ¡ ƒ}i }| |¡ ‡fd d
|
|||
|
|
„} t |¡}
|
|||
|
|
|
|
|||
|
|
d¡ t|
|
|||
|
|
ˆjjd| d�} t | ƒD ]�\} ‰ ˆj ˆj ¡�m ‡ fdd„|D ƒ}ˆ ˆj |¡}
|
|||
|
|
ˆjj dk�r|
|
|||
|
|
!¡ }
|
|||
|
|
|
|
|||
|
|
"¡ }ˆj #|
|
|||
|
|
¡ ˆj j$�r8ˆjj%du�r8ˆj &ˆj '¡ ˆjj%¡ ˆj( )¡ ˆj( *¡ ˆj+du�rMˆj+ )¡ ˆj jd7 _ˆ j |7 _ ˆ ,¡ W d ƒ n 1 �skw Y qïdS )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 rk 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)rô Ú
|
|||
|
|
truncationÚpaddingÚreturn_tensorsr5 r6 c sF t ƒ }| d D ]‰ ˆ dv r t ‡ fdd„| D ƒ¡ ˆjj¡|ˆ < q|S )Nr r= c s g | ]}|ˆ ‘qS rM rM )r9 Úd©ÚkeyrM rN r; Œ s zF_UnslothIterativeSFTTrainer.step.<locals>.collator.<locals>.<listcomp>)Údictr% Ústackr< r rI )ÚdataÚ return_dictrG r[ rN Úcollatorˆ s &€z2_UnslothIterativeSFTTrainer.step.<locals>.collatorr% )Ú
|
|||
|
|
batch_sizeÚshuffleÚ
|
|||
|
|
collate_fnc s i | ]}|ˆ | “qS rM rM )r9 rE )ÚbatchrM rN rH › r< z4_UnslothIterativeSFTTrainer.step.<locals>.<dictcomp>r7 )-r ÚtrainÚstateÚ global_stepr% Útensorr< r
|
|||
|
|
rI Útr_lossÚ_globalstep_last_loggedrL r& r$ Ú UserWarningr rU r
|
|||
|
|
rô rN rQ ÚkeysÚupdater Ú from_dictÚ
|
|||
|
|
set_formatr rz Ú enumerater Ú
|
|||
|
|
accumulateÚ compute_lossÚn_gpuÚmeanÚdetachÚbackwardÚsync_gradientsr‡ Úclip_grad_norm_Ú
|
|||
|
|
parametersr, ÚstepÚ zero_gradr- Ú_maybe_log_save_evaluate)rû r5 r6 r7 rO rP Ú model_inputsÚmodel_inputs_namesÚ
|
|||
|
|
batch_dictra Ú
|
|||
|
|
batch_dataÚstep_dataloaderÚ_ÚlossÚ tr_loss_steprM )re rû rN r{ 7 s�
|
|||
|
|
|
|||
|
|
ýÿ
|
|||
|
|
ÿ ÿ ÿþ
|
|||
|
|
|
|||
|
|
|
|||
|
|
ü þ
|
|||
|
|
|
|||
|
|
|
|||
|
|
|
|||
|
|
å€ÿz _UnslothIterativeSFTTrainer.stepc C sØ | j jd ur | jj| j j dkr | jjdkr | | j¡ | j jd urf| jj| j j dkrh| jjdkrji }| | j¡ ¡
|
|||
|
|
¡ }| j| j8 _t || jj| j dƒ|d< |
|
|||
|
|
¡ |d< | jj| _ | |¡ d S d S d S d S )Nr r3 r„ r‚ )r
|
|||
|
|
r° rg rh Úevaluater r“ Ú_nested_gatherrj ru ÚitemÚroundrk Ú_get_learning_rateÚlog)rû ÚlogsÚtr_loss_scalarrM rM rN r} · s
|
|||
|
|
òz4_UnslothIterativeSFTTrainer._maybe_log_save_evaluatec sL | j jd u rt| j jƒj}n | j j d¡d }| j|d� tƒ ||¡ d S )Nr r2 )r1 ) r
|
|||
|
|
rÒ r rs rS r Úcreate_model_cardrù Ú_save_checkpoint)rû r Útrialr1 rý rM rN r� Ï s
|
|||
|
|
z,_UnslothIterativeSFTTrainer._save_checkpointr1 Ú dataset_nameÚtagsc
|
|||
|
|
C sÚ | ¡ sdS t| jjdƒr tj | jjj¡s | jjj}nd}|du r&tƒ }n
|
|||
|
|
t |t
|
|||
|
|
ƒr/|h}nt|ƒ}t| jjdƒr?| d¡ | | j
|
|||
|
|
¡ t||| j||tƒ rXtjdurXtjjndtƒ dd�}| tj | jjd¡¡ dS )aî
|
|||
|
|
Creates a draft of a model card using the information available to the `Trainer`.
|
|||
|
|
|
|||
|
|
Args:
|
|||
|
|
model_name (`str` or `None`, *optional*, defaults to `None`):
|
|||
|
|
Name of the model.
|
|||
|
|
dataset_name (`str` or `None`, *optional*, defaults to `None`):
|
|||
|
|
Name of the dataset used for training.
|
|||
|
|
tags (`str`, `list[str]` or `None`, *optional*, defaults to `None`):
|
|||
|
|
Tags to be associated with the model card.
|
|||
|
|
Nr Úunsloth_versionÚunslothz
|
|||
|
|
Iterative SFT)Ú
|
|||
|
|
base_modelr1 rÒ r‘ r’ Ú wandb_urlÚ comet_urlÚ trainer_namez README.md)Úis_world_process_zeror( r r r$ ÚpathÚisdirr Úsetr r Úaddrn r) r rÒ r# ÚwandbÚrunÚurlr! ÚsaveÚjoinr
|
|||
|
|
rs )rû r1 r‘ r’ r• Ú
|
|||
|
|
model_cardrM rM rN rŽ × s0
|
|||
|
|
|