U
9%e_ @ s d Z ddlZddlZddlmZ ddlmZmZmZmZ ddl Z
ddl m Z ddl
mZmZ ddlmZ dd lmZ eeZeG d
d d ZeG d d
d
ZG dd dZG dd deZG dd deZG dd deZ G dd deZ G dd deZ dS )zJ
Callbacks to use with the Trainer class and customize the training loop.
N) dataclass)DictListOptionalUnion)tqdm )IntervalStrategy
has_length)TrainingArguments)loggingc @ s& e Zd ZU dZdZee ed< dZe ed< dZ
e ed< dZ e ed< dZ e ed < dZ
e ed
< dZe ed < dZeed < dZeeeef ed
< dZee ed< dZee ed< dZeed< dZeed< dZeed< dZeed< dZeee eee ef f ed< dd Z edddZ e edd d Z dS ) TrainerStateaN
A class containing the [`Trainer`] inner state that will be saved along the model and optimizer when checkpointing
and passed to the [`TrainerCallback`].
In all this class, one step is to be understood as one update step. When using gradient accumulation, one update
step may require several forward and backward passes: if you use `gradient_accumulation_steps=n`, then one update
step requires going through *n* batches.
Args:
epoch (`float`, *optional*):
Only set during training, will represent the epoch the training is at (the decimal part being the
percentage of the current epoch completed).
global_step (`int`, *optional*, defaults to 0):
During training, represents the number of update steps completed.
max_steps (`int`, *optional*, defaults to 0):
The number of update steps to do during the current training.
logging_steps (`int`, *optional*, defaults to 500):
Log every X updates steps
eval_steps (`int`, *optional*):
Run an evaluation every X steps.
save_steps (`int`, *optional*, defaults to 500):
Save checkpoint every X updates steps.
total_flos (`float`, *optional*, defaults to 0):
The total number of floating operations done by the model since the beginning of training (stored as floats
to avoid overflow).
log_history (`List[Dict[str, float]]`, *optional*):
The list of logs done since the beginning of training.
best_metric (`float`, *optional*):
When tracking the best model, the value of the best metric encountered so far.
best_model_checkpoint (`str`, *optional*):
When tracking the best model, the value of the name of the checkpoint for the best model encountered so
far.
is_local_process_zero (`bool`, *optional*, defaults to `True`):
Whether or not this process is the local (e.g., on one machine if training in a distributed fashion on
several machines) main process.
is_world_process_zero (`bool`, *optional*, defaults to `True`):
Whether or not this process is the global main process (when training in a distributed fashion on several
machines, this is only going to be `True` for one process).
is_hyper_param_search (`bool`, *optional*, defaults to `False`):
Whether we are in the process of a hyper parameter search using Trainer.hyperparameter_search. This will
impact the way data will be logged in TensorBoard.
Nepochr global_step max_stepsi
logging_steps
eval_steps
save_stepsnum_train_epochs
total_flos log_history best_metricbest_model_checkpointTis_local_process_zerois_world_process_zeroFis_hyper_param_search
trial_name trial_paramsc C s | j d krg | _ d S N)r self r! \/var/www/html/Darija-Ai-API/env/lib/python3.8/site-packages/transformers/trainer_callback.py
__post_init__d s
zTrainerState.__post_init__) json_pathc C sB t jt| dddd }t|ddd}|| W 5 Q R X dS ) zDSave the content of this instance in JSON format inside `json_path`. T)indent sort_keys
wutf-8encodingN)jsondumps dataclassesasdictopenwrite)r r$ Z json_stringfr! r! r" save_to_jsonh s zTrainerState.save_to_jsonc C s2 t |ddd}| }W 5 Q R X | f t|S )z3Create an instance from the content of `json_path`.rr* r+ )r1 readr- loads)clsr$ r3 textr! r! r" load_from_jsonn s zTrainerState.load_from_json)!__name__
__module__ __qualname____doc__r r float__annotations__r intr r r r r r r r r strr r r boolr r r r r r# r4 classmethodr: r! r! r! r" r
" s*
/ r
c @ sf e Zd ZU dZdZeed< dZeed< dZeed< dZ eed< dZ
eed< dd Z d
d Z d d
Z
dS )TrainerControlaA
A class that handles the [`Trainer`] control flow. This class is used by the [`TrainerCallback`] to activate some
switches in the training loop.
Args:
should_training_stop (`bool`, *optional*, defaults to `False`):
Whether or not the training should be interrupted.
If `True`, this variable will not be set back to `False`. The training will just stop.
should_epoch_stop (`bool`, *optional*, defaults to `False`):
Whether or not the current epoch should be interrupted.
If `True`, this variable will be set back to `False` at the beginning of the next epoch.
should_save (`bool`, *optional*, defaults to `False`):
Whether or not the model should be saved at this step.
If `True`, this variable will be set back to `False` at the beginning of the next step.
should_evaluate (`bool`, *optional*, defaults to `False`):
Whether or not the model should be evaluated at this step.
If `True`, this variable will be set back to `False` at the beginning of the next step.
should_log (`bool`, *optional*, defaults to `False`):
Whether or not the logs should be reported at this step.
If `True`, this variable will be set back to `False` at the beginning of the next step.
Fshould_training_stopshould_epoch_stop should_saveshould_evaluate
should_logc C s
d| _ dS )z rF rC r@ rG rH rI rJ rK rL rM r! r! r! r" rE v s
rE c @ s e Zd ZdZeeedddZeeedddZeeedddZ eeedd d
Z
eeedd d Z eeedd
dZ eeedddZ
eeedddZeeedddZeeedddZeeedddZeeedddZeeeddd Zd S ) TrainerCallbacka
A class for objects that will inspect the state of the training loop at some events and take some decisions. At
each of those events the following arguments are available:
Args:
args ([`TrainingArguments`]):
The training arguments used to instantiate the [`Trainer`].
state ([`TrainerState`]):
The current state of the [`Trainer`].
control ([`TrainerControl`]):
The object that is returned to the [`Trainer`] and can be used to make some decisions.
model ([`PreTrainedModel`] or `torch.nn.Module`):
The model being trained.
tokenizer ([`PreTrainedTokenizer`]):
The tokenizer used for encoding the data.
optimizer (`torch.optim.Optimizer`):
The optimizer used for the training steps.
lr_scheduler (`torch.optim.lr_scheduler.LambdaLR`):
The scheduler used for setting the learning rate.
train_dataloader (`torch.utils.data.DataLoader`, *optional*):
The current dataloader used for training.
eval_dataloader (`torch.utils.data.DataLoader`, *optional*):
The current dataloader used for training.
metrics (`Dict[str, float]`):
The metrics computed by the last evaluation phase.
Those are only accessible in the event `on_evaluate`.
logs (`Dict[str, float]`):
The values to log.
Those are only accessible in the event `on_log`.
The `control` object is the only one that can be changed by the callback, in which case the event that changes it
should return the modified version.
The argument `args`, `state` and `control` are positionals for all events, all the others are grouped in `kwargs`.
You can unpack the ones you need in the signature of the event using them. As an example, see the code of the
simple [`~transformer.PrinterCallback`].
Example:
```python
class PrinterCallback(TrainerCallback):
def on_log(self, args, state, control, logs=None, **kwargs):
_ = logs.pop("total_flos", None)
if state.is_local_process_zero:
print(logs)
```argsstatecontrolc K s dS )zS
Event called at the end of the initialization of the [`Trainer`].
Nr! r rP rQ rR kwargsr! r! r" on_init_end s zTrainerCallback.on_init_endc K s dS )z<
Event called at the beginning of training.
Nr! rS r! r! r" on_train_begin s z TrainerCallback.on_train_beginc K s dS )z6
Event called at the end of training.
Nr! rS r! r! r" on_train_end s z TrainerCallback.on_train_endc K s dS )z<
Event called at the beginning of an epoch.
Nr! rS r! r! r" on_epoch_begin s z TrainerCallback.on_epoch_beginc K s dS )z6
Event called at the end of an epoch.
Nr! rS r! r! r" on_epoch_end s z TrainerCallback.on_epoch_endc K s dS )z
Event called at the beginning of a training step. If using gradient accumulation, one training step might take
several inputs.
Nr! rS r! r! r"
on_step_begin s z TrainerCallback.on_step_beginc K s dS )zU
Event called at the end of an substep during gradient accumulation.
Nr! rS r! r! r" on_substep_end s z TrainerCallback.on_substep_endc K s dS )z
Event called at the end of a training step. If using gradient accumulation, one training step might take
several inputs.
Nr! rS r! r! r" on_step_end s zTrainerCallback.on_step_endc K s dS )z9
Event called after an evaluation phase.
Nr! rS r! r! r" on_evaluate s zTrainerCallback.on_evaluatec K s dS )z=
Event called after a successful prediction.
Nr! )r rP rQ rR metricsrT r! r! r"
on_predict s zTrainerCallback.on_predictc K s dS )z7
Event called after a checkpoint save.
Nr! rS r! r! r" on_save s zTrainerCallback.on_savec K s dS )z;
Event called after logging the last logs.
Nr! rS r! r! r" on_log s zTrainerCallback.on_logc K s dS )z7
Event called after a prediction step.
Nr! rS r! r! r" on_prediction_step$ s z"TrainerCallback.on_prediction_stepN)r; r< r= r> r r
rE rU rV rW rX rY rZ r[ r\ r] r_ r` ra rb r! r! r! r" rN s 1rN c @ s. e Zd ZdZdd Zdd Zdd Zdd Zed
d Z e
e e d d
dZ
e
e e d ddZe
e e d ddZe
e e d ddZe
e e d ddZe
e e d ddZe
e e d ddZe
e e d dd Ze
e e d d d Ze
e e d d d Ze
e e d d!d"Ze
e e d d#d$Ze
e e d d%d&Zd'd( Zd)S )*CallbackHandlerz>Internal class that just calls the list of callbacks in order.c C sf g | _ |D ]}| | q
|| _|| _|| _|| _d | _d | _tdd | j D sbt
d| j d S )Nc s s | ]}t |tV qd S r )
isinstanceDefaultFlowCallback.0cbr! r! r" 9 s z+CallbackHandler.__init__..zThe Trainer will not work properly if you don't have a `DefaultFlowCallback` in its callbacks. You
should add one before training with `trainer.add_callback(DefaultFlowCallback). The current list ofcallbacks is
:) callbacks add_callbackmodel tokenizer optimizer lr_schedulertrain_dataloadereval_dataloaderanyloggerwarning
callback_list)r rj rl rm rn ro rh r! r! r" __init__. s zCallbackHandler.__init__c C sh t |tr| n|}t |tr"|n|j}|dd | jD krXtd| dd | j | j| d S )Nc S s g | ]
}|j qS r! ) __class__)rg cr! r! r"
D s z0CallbackHandler.add_callback..zYou are adding a zH to the callbacks of this Trainer, but there is already one. The currentzlist of callbacks is
:)rd typerw rj rs rt ru append)r callbackrh Zcb_classr! r! r" rk A s
z CallbackHandler.add_callbackc C sb t |tr6| jD ]"}t ||r| j| | S qn(| jD ] }||kr<| j| | S q.)joinrj r r! r! r" ru a s z CallbackHandler.callback_listrO c C s | d|||S )NrU
call_eventr rP rQ rR r! r! r" rU e s zCallbackHandler.on_init_endc C s d|_ | d|||S )NFrV )rF r r r! r! r" rV h s z CallbackHandler.on_train_beginc C s | d|||S )NrW r r r! r! r" rW l s z CallbackHandler.on_train_endc C s d|_ | d|||S )NFrX )rG r r r! r! r" rX o s z CallbackHandler.on_epoch_beginc C s | d|||S )NrY r r r! r! r" rY s s z CallbackHandler.on_epoch_endc C s" d|_ d|_d|_| d|||S )NFrZ )rJ rI rH r r r! r! r" rZ v s z CallbackHandler.on_step_beginc C s | d|||S )Nr[ r r r! r! r" r[ | s z CallbackHandler.on_substep_endc C s | d|||S )Nr\ r r r! r! r" r\ s zCallbackHandler.on_step_endc C s d|_ | jd||||dS )NFr] r^ )rI r r rP rQ rR r^ r! r! r" r] s zCallbackHandler.on_evaluatec C s | j d||||dS )Nr_ r r r r! r! r" r_ s zCallbackHandler.on_predictc C s d|_ | d|||S )NFr` )rH r r r! r! r" r` s zCallbackHandler.on_savec C s d|_ | jd||||dS )NFra )logs)rJ r )r rP rQ rR r r! r! r" ra s zCallbackHandler.on_logc C s | d|||S )Nrb r r r! r! r" rb s z"CallbackHandler.on_prediction_stepc
K sP | j D ]D}t|||||f| j| j| j| j| j| jd|}|d k r|}q|S )N)rl rm rn ro rp rq )rj getattrrl rm rn ro rp rq )r eventrP rQ rR rT r| resultr! r! r" r s$
zCallbackHandler.call_eventN)r; r< r= r> rv rk r r propertyru r r
rE rU rV rW rX rY rZ r[ r\ r] r_ r` ra rb r r! r! r! r" rc + s*
rc c @ s4 e Zd ZdZeeedddZeeedddZdS )re zx
A [`TrainerCallback`] that handles the default flow of the training loop for logs, evaluation and checkpoints.
rO c K s |j dkr|jrd|_|jtjkr8|j |j dkr8d|_|jtjkrf|j |j dkrf|j |j krfd|_
|j tjkr|j dkr|j |j dkrd|_
|j |jkrd|_|S )Nr Tr )r Zlogging_first_steprJ logging_strategyr ZSTEPSr evaluation_strategyr
eval_delayrI
save_strategyr rH r rF rS r! r! r" r\ s*
z DefaultFlowCallback.on_step_endc K sF |j tjkrd|_|jtjkr0|j|jkr0d|_|jtjkrBd|_ |S )NT)
r r EPOCHrJ r r r rI r rH rS r! r! r" rY s z DefaultFlowCallback.on_epoch_endN) r; r< r= r> r r
rE r\ rY r! r! r! r" re s re c @ sT e Zd ZdZdd Zdd Zdd Zdd d
Zd d Zd
d Z dddZ
dd Z dS )ProgressCallbackzU
A [`TrainerCallback`] that displays the progress of training or evaluation.
c C s d | _ d | _d S r ) training_barprediction_barr r! r! r" rv s zProgressCallback.__init__c K s |j rt|jdd| _d| _d S )NT)total
dynamic_ncolsr )r r r r current_steprS r! r! r" rV s z ProgressCallback.on_train_beginc K s& |j r"| j|j| j |j| _d S r )r r updater r rS r! r! r" r\ s z ProgressCallback.on_step_endNc K sB |j r>t|r>| jd kr2tt|| jd kdd| _| jd d S )NT)r Zleaver r )r r
r r lenr r )r rP rQ rR rq rT r! r! r" rb s
z#ProgressCallback.on_prediction_stepc K s$ |j r | jd k r| j d | _d S r r r closerS r! r! r" r] s
z ProgressCallback.on_evaluatec K s$ |j r | jd k r| j d | _d S r r rS r! r! r" r_ s
zProgressCallback.on_predictc K s0 |j r,| jd k r,|dd }| jt| d S Nr )r r popr2 rB r rP rQ rR r rT _r! r! r" ra s zProgressCallback.on_logc K s |j r| j d | _d S r )r r r rS r! r! r" rW s
z ProgressCallback.on_train_end)N)N) r; r< r= r> rv rV r\ rb r] r_ ra rW r! r! r! r" r s
r c @ s e Zd ZdZdddZdS )PrinterCallbackz?
A bare [`TrainerCallback`] that just prints the logs.
Nc K s | dd }|jrt| d S r )r r printr r! r! r" ra s zPrinterCallback.on_log)N)r; r< r= r> ra r! r! r! r" r s r c @ s> e Zd ZdZdeee dddZdd Zd d
Z d d Z
d
S )EarlyStoppingCallbacka/
A [`TrainerCallback`] that handles early stopping.
Args:
early_stopping_patience (`int`):
Use with `metric_for_best_model` to stop training when the specified metric worsens for
`early_stopping_patience` evaluation calls.
early_stopping_threshold(`float`, *optional*):
Use with TrainingArguments `metric_for_best_model` and `early_stopping_patience` to denote how much the
specified metric must improve to satisfy early stopping conditions. `
This callback depends on [`TrainingArguments`] argument *load_best_model_at_end* functionality to set best_metric
in [`TrainerState`]. Note that if the [`TrainingArguments`] argument *save_steps* differs from *eval_steps*, the
early stopping will not occur until the next save step.
r )early_stopping_patienceearly_stopping_thresholdc C s || _ || _d| _d S )Nr )r r early_stopping_patience_counter)r r r r! r! r" rv ( s z EarlyStoppingCallback.__init__c C sV |j r tjntj}|jd ks<|||jrDt||j | jkrDd| _n| jd7 _d S )Nr r )Zgreater_is_betternpZgreaterlessr absr r )r rP rQ rR metric_valueoperatorr! r! r" check_metric_value. s
z(EarlyStoppingCallback.check_metric_valuec K s8 |j std|jd k s td|jtjks4tdd S )Nz rA r r? rv r rV r] r! r! r! r" r s
r ) r> r/ r- r typingr r r r numpyr Z tqdm.autor Z
trainer_utilsr r
Z
training_argsr utilsr Z
get_loggerr; rs r
rE rN rc re r r r r! r! r! r" s*
S1 }22