Trainer

Users could freely customize the training process, and can also use our Runner API to conduct classification or generation, while in this case you need to construct a config.yml file, see Play with Configuration.

Classification Runner

class ClassificationRunner(model: openprompt.pipeline_base.PromptForClassification, config: Optional[yacs.config.CfgNode] = None, train_dataloader: Optional[openprompt.pipeline_base.PromptDataLoader] = None, valid_dataloader: Optional[openprompt.pipeline_base.PromptDataLoader] = None, test_dataloader: Optional[openprompt.pipeline_base.PromptDataLoader] = None, loss_function: Optional[Callable] = None, id2label: Optional[Dict] = None)[source]

A runner for simple training without training tricks. Applying training tricks such as ensemble of template or verbalizer, or self-training can use other runner class. This class is specially implemented for classification. For generation task, though it can be integrated in this class via task option, we keep it as another class for simplicity.

Parameters

  • model (PromptForClassification) – One PromptForClassification object.

  • train_dataloader (PromptDataloader, optional) – The dataloader to bachify and process the training data.

  • valid_dataloader (PromptDataloader, optionla) – The dataloader to bachify and process the val data.

  • test_dataloader (PromptDataloader, optional) – The dataloader to bachify and process the test data.

  • config (CfgNode) – A configuration object.

  • loss_function (Callable, optional) – The loss function in the training process.

configure_loss_function()[source]

config the loss function if it’s not passed.

on_fit_start()[source]

Some initialization works

Generation Runner

class GenerationRunner(model: openprompt.pipeline_base.PromptForGeneration, config: Optional[yacs.config.CfgNode] = None, train_dataloader: Optional[openprompt.pipeline_base.PromptDataLoader] = None, valid_dataloader: Optional[openprompt.pipeline_base.PromptDataLoader] = None, test_dataloader: Optional[openprompt.pipeline_base.PromptDataLoader] = None)[source]

A runner for simple training without training tricks. Applying training tricks such as ensemble of template or verbalizer, or self-training can use other runner class. This class is specially implemented for generation.

Parameters

  • model (PromptForGeneration) – One PromptForGeneration object.

  • train_dataloader (PromptDataloader, optional) – The dataloader to bachify and process the training data.

  • valid_dataloader (PromptDataloader, optionla) – The dataloader to bachify and process the val data.

  • test_dataloader (PromptDataloader, optional) – The dataloader to bachify and process the test data.

  • config (CfgNode) – A configuration object.

LM-BFF Classification Runner

class LMBFFClassificationRunner(train_dataset: List[transformers.data.processors.utils.InputExample], valid_dataset: List[transformers.data.processors.utils.InputExample], test_dataset: List[transformers.data.processors.utils.InputExample], verbalizer: Optional[openprompt.prompt_base.Verbalizer] = None, template: Optional[str] = None, config: Optional[yacs.config.CfgNode] = None)[source]

This runner implements the LM-BFF training process in paper Making Pre-trained Language Models Better Few-shot Learners(Gao et al. 2020).

Parameters

  • train_dataset (List[InputExample]) – The dataset for training

  • valid_dataset (List[InputExample]) – The dataset for validation

  • test_dataset (List[InputExample]) – The dataset for test

  • verbalizer (Optional[Verbalizer]) – The manually designed verbalizer for template generation. Defaults to None.

  • template (Optional[Verbalizer]) – The manually designed template for verbalizer generation. Defaults to None.

  • config (CfgNode) – A configuration object

run()[source]

Run LM-BFF. if both auto_v and auto_v are set to True in config, automatic template generation will be performed first.