training

superduper.ext.transformers.training

Source code

create_quantization_config

create_quantization_config(config: superduper.ext.transformers.training.LLMTrainer)

Parameter	Description
config	The configuration to use.

Create quantization config for LLM training.

handle_ray_results

handle_ray_results(db,
     llm,
     results)

Parameter	Description
db	datalayer, used for saving the checkpoint
llm	llm model, used for saving the checkpoint
results	the ray training results, contains the checkpoint

Handle the ray results.

Will save the checkpoint to db if db and llm provided.

prepare_lora_training

prepare_lora_training(model,
     config: superduper.ext.transformers.training.LLMTrainer)

Parameter	Description
model	The model to prepare for LoRA training.
config	The configuration to use.

Prepare LoRA training for the model.

Get the LoRA target modules and convert the model to peft model.

train_func

train_func(training_args: superduper.ext.transformers.training.LLMTrainer,
     train_dataset: 'Dataset',
     eval_datasets: Union[ForwardRef('Dataset'),
     Dict[str,
     ForwardRef('Dataset')]],
     model_kwargs: dict,
     tokenizer_kwargs: dict,
     trainer_prepare_func: Optional[Callable] = None,
     callbacks=None,
     **kwargs)

Parameter	Description
training_args	training Arguments, see LLMTrainingArguments
train_dataset	training dataset, can be huggingface datasets.Dataset or ray.data.Dataset
eval_datasets	evaluation dataset, can be a dict of datasets
model_kwargs	model kwargs for AutoModelForCausalLM
tokenizer_kwargs	tokenizer kwargs for AutoTokenizer
trainer_prepare_func	function to prepare trainer This function will be called after the trainer is created, we can add some custom settings to the trainer
callbacks	list of callbacks will be added to the trainer
kwargs	other kwargs for Trainer All the kwargs will be passed to Trainer, make sure the Trainer support these kwargs

Base training function for LLM model.

tokenize

tokenize(tokenizer,
     example,
     X,
     y)

Parameter	Description
tokenizer	The tokenizer to use.
example	The example to tokenize.
X	The input key.
y	The output key.

Function to tokenize the example.

train

train(training_args: superduper.ext.transformers.training.LLMTrainer,
     train_dataset: datasets.arrow_dataset.Dataset,
     eval_datasets: Union[datasets.arrow_dataset.Dataset,
     Dict[str,
     datasets.arrow_dataset.Dataset]],
     model_kwargs: dict,
     tokenizer_kwargs: dict,
     db: Optional[ForwardRef('Datalayer')] = None,
     llm: Optional[ForwardRef('LLM')] = None,
     ray_configs: Optional[dict] = None,
     **kwargs)

Parameter	Description
training_args	training Arguments, see LLMTrainingArguments
train_dataset	training dataset
eval_datasets	evaluation dataset, can be a dict of datasets
model_kwargs	model kwargs for AutoModelForCausalLM
tokenizer_kwargs	tokenizer kwargs for AutoTokenizer
db	datalayer, used for creating LLMCallback
llm	llm model, used for creating LLMCallback
ray_configs	ray configs, must provide if using ray
kwargs	other kwargs for Trainer

Train LLM model on specified dataset.

The training process can be run on these following modes:

• Local node without ray, but only support single GPU
• Local node with ray, support multi-nodes and multi-GPUs
• Remote node with ray, support multi-nodes and multi-GPUs

If run locally, will use train_func to train the model. Can log the training process to db if db and llm provided. Will reuse the db and llm from the current process. If run on ray, will use ray_train to train the model. Can log the training process to db if db and llm provided. Will rebuild the db and llm for the new process that can access to db. The ray cluster must can access to db.

Checkpoint

Checkpoint(self,
     identifier: str,
     db: dataclasses.InitVar[typing.Optional[ForwardRef('Datalayer')]] = None,
     uuid: str = None,
     *,
     artifacts: 'dc.InitVar[t.Optional[t.Dict]]' = None,
     path: Optional[str],
     step: int) -> None

Parameter	Description
identifier	Identifier of the leaf.
db	Datalayer instance.
uuid	UUID of the leaf.
artifacts	A dictionary of artifacts paths and DataType objects
path	The path to the checkpoint.
step	The step of the checkpoint.

Checkpoint component for saving the model checkpoint.

LLMCallback

LLMCallback(self,
     cfg: Optional[ForwardRef('Config')] = None,
     identifier: Optional[str] = None,
     db: Optional[ForwardRef('Datalayer')] = None,
     llm: Optional[ForwardRef('LLM')] = None,
     experiment_id: Optional[str] = None)

Parameter	Description
cfg	The configuration to use.
identifier	The identifier to use.
db	The datalayer to use.
llm	The LLM model to use.
experiment_id	The experiment id to use.

LLM Callback for logging training process to db.

This callback will save the checkpoint to db after each epoch. If the save_total_limit is set, will remove the oldest checkpoint.