mindformers.trainer.TrainingArguments

class mindformers.trainer.TrainingArguments(output_dir: Optional[str] = None, use_parallel: Optional[bool] = None, profile: Optional[bool] = None, only_save_strategy: Optional[bool] = None, sink_mode: Optional[bool] = None, sink_size: Optional[bool] = None, batch_size: Optional[int] = None, per_device_train_batch_size: Optional[int] = None, per_device_eval_batch_size: Optional[int] = None, learning_rate: Optional[float] = None, weight_decay: Optional[float] = None, adam_beta1: Optional[float] = None, adam_beta2: Optional[float] = None, adam_epsilon: Optional[float] = None, max_grad_norm: Optional[float] = None, num_train_epochs: Optional[float] = None, lr_scheduler_type: Optional[Union[LRType, str]] = None, optim: Optional[Union[OptimizerType, str]] = None, warmup_steps: int = 0, save_strategy: Optional[Union[SaveIntervalStrategy, str]] = None, integrated_save: Optional[bool] = None, save_steps: Optional[int] = None, save_seconds: Optional[int] = None, save_total_limit: Optional[int] = None, seed: Optional[int] = None, data_seed: Optional[int] = None, resume_from_checkpoint: Optional[str] = None, auto_trans_ckpt: Optional[bool] = None, resume_training: Optional[bool] = None, do_eval: Optional[bool] = None, eval_step_interval: Optional[int] = None, eval_epoch_interval: Optional[int] = None)

TrainingArguments is the subset of the arguments we use in our default config which is relate to the training in MindSpore.

convert_args_to_mindformers_config(task_config: Optional[MindFormerConfig] = None)

convert training arguments to mindformer config type for adapting hugging-face.

property get_device_id

get device id for training.

property get_device_num

get device num for training.

property get_rank_id

get rank id for training.