mindformers.trainer.QuestionAnsweringTrainer¶

class mindformers.trainer.QuestionAnsweringTrainer(model_name: str = None)[源代码]¶

QuestionAnswering Task For Trainer. :param model_name: The model name of Task-Trainer. Default: None :type model_name: str

实际案例

>>> import numpy as np
>>> from mindspore.nn import AdamWeightDecay, TrainOneStepCell
>>> from mindformers.core.lr import build_lr
>>> from mindformers.trainer import GeneralTaskTrainer
>>> from mindformers.tools.register import MindFormerConfig
>>> from mindformers.models import BertForQuestionAnswering, BertConfig
>>> config = MindFormerConfig("configs/qa/run_qa_bert_base_uncased.yaml")
>>> #1) use config to train
>>> cls_task = QuestionAnsweringTrainer(model_name='qa_bert_base_uncased')
>>> cls_task.train(config=config)
>>> #2) use instance function to train

引发: NotImplementedError – If train method or evaluate method or predict method not implemented.

evaluate(config: Union[dict, mindformers.tools.register.config.MindFormerConfig, mindformers.trainer.config_args.ConfigArguments, mindformers.trainer.training_args.TrainingArguments, None] = None, network: Union[mindspore.nn.cell.Cell, mindformers.models.base_model.BaseModel, None] = None, dataset: Union[mindformers.dataset.base_dataset.BaseDataset, mindspore.dataset.engine.datasets_user_defined.GeneratorDataset, None] = None, callbacks: Union[mindspore.train.callback._callback.Callback, List[mindspore.train.callback._callback.Callback], None] = None, compute_metrics: Union[dict, set, None] = None, **kwargs)[源代码]¶

Evaluate task for TokenClassification Trainer. This function is used to evaluate the network.

The trainer interface is used to quickly start training for general task. It also allows users to customize the network, dataset, callbacks, compute_metrics.

参数

config (Optional[Union[dict, MindFormerConfig, ConfigArguments, TrainingArguments]]) – The task config which is used to configure the dataset, the hyper-parameter, optimizer, etc. It supports config dict or MindFormerConfig or TrainingArguments or ConfigArguments class. Default: None.
network (Optional[Union[Cell, BaseModel]]) – The network for trainer. It supports model name or BaseModel or MindSpore Cell class. Default: None.
dataset (Optional[Union[BaseDataset]]) – The evaluate dataset. It support real dataset path or BaseDateset class or MindSpore Dataset class. Default: None.
callbacks (Optional[Union[Callback, List[Callback]]]) – The eval callback function. It support CallBack or CallBack List of MindSpore. Default: None.
compute_metrics (Optional[Union[dict, set]]) – The metric of evaluating. It support dict or set in MindSpore’s Metric class. Default: None.

predict(config: Union[dict, mindformers.tools.register.config.MindFormerConfig, mindformers.trainer.config_args.ConfigArguments, mindformers.trainer.training_args.TrainingArguments, None] = None, input_data: Union[str, list, None] = None, network: Union[mindspore.nn.cell.Cell, mindformers.models.base_model.BaseModel, None] = None, tokenizer: Optional[mindformers.models.base_tokenizer.BaseTokenizer] = None, **kwargs)[源代码]¶

Executes the predict of the trainer.

参数

config (Optional[Union[dict, MindFormerConfig, ConfigArguments, TrainingArguments]]) – The task config which is used to configure the dataset, the hyper-parameter, optimizer, etc. It supports config dict or MindFormerConfig or TrainingArguments or ConfigArguments class. Default: None.
input_data (Optional[Union[Tensor, str, list]]) – The predict data. Default: None.
network (Optional[Union[Cell, BaseModel]]) – The network for trainer. It supports model name or BaseModel or MindSpore Cell class. Default: None.
tokenizer (Optional[BaseTokenizer]) – The tokenizer for tokenizing the input text. Default: None.

返回

A list of prediction.

train(config: Union[dict, mindformers.tools.register.config.MindFormerConfig, mindformers.trainer.config_args.ConfigArguments, mindformers.trainer.training_args.TrainingArguments, None] = None, network: Union[mindspore.nn.cell.Cell, mindformers.models.base_model.BaseModel, None] = None, dataset: Union[mindformers.dataset.base_dataset.BaseDataset, mindspore.dataset.engine.datasets_user_defined.GeneratorDataset, None] = None, wrapper: Optional[mindspore.nn.wrap.cell_wrapper.TrainOneStepCell] = None, optimizer: Optional[mindspore.nn.optim.optimizer.Optimizer] = None, callbacks: Union[mindspore.train.callback._callback.Callback, List[mindspore.train.callback._callback.Callback], None] = None, **kwargs)[源代码]¶

Train task for TokenClassification Trainer. This function is used to train or fine-tune the network.

The trainer interface is used to quickly start training for general task. It also allows users to customize the network, optimizer, dataset, wrapper, callback.

参数

config (Optional[Union[dict, MindFormerConfig, ConfigArguments, TrainingArguments]]) – The task config which is used to configure the dataset, the hyper-parameter, optimizer, etc. It supports config dict or MindFormerConfig or TrainingArguments or ConfigArguments class. Default: None.
network (Optional[Union[Cell, BaseModel]]) – The network for trainer. It supports model name or BaseModel or MindSpore Cell class. Default: None.
dataset (Optional[Union[BaseDataset, GeneratorDataset]]) – The training dataset. It support real dataset path or BaseDateset class or MindSpore Dataset class. Default: None.
optimizer (Optional[Optimizer]) – The training network’s optimizer. It support Optimizer class of MindSpore. Default: None.
wrapper (Optional[TrainOneStepCell]) – Wraps the network with the optimizer. It support TrainOneStepCell class of MindSpore. Default: None.
callbacks (Optional[Union[Callback, List[Callback]]]) – The training callback function. It support CallBack or CallBack List of MindSpore. Default: None.

引发

NotImplementedError – If wrapper not implemented.