espnet2.train.deepspeed_trainer.DeepSpeedTrainerOptions

source

class espnet2.train.deepspeed_trainer.DeepSpeedTrainerOptions(resume: bool, seed: int, train_dtype: str | dtype, log_interval: int | None, output_dir: Path | str, max_epoch: int, deepspeed_config: Path | str)

Bases: object

DeepSpeed Trainer Options for configuring training with DeepSpeed.

This class holds the options required to set up the DeepSpeed trainer, including parameters for resuming training, setting seeds, data types, logging intervals, output directories, and maximum epochs.

resume

Whether to resume training from a checkpoint.

• Type: bool

seed

Seed for random number generation.

• Type: int

train_dtype

Data type for training (e.g., ‘fp16’).

• Type: Union[str, torch.dtype]

log_interval

Interval for logging training progress.

• Type: Optional[int]

output_dir

Directory for saving outputs and checkpoints.

• Type: Union[Path, str]

max_epoch

Maximum number of epochs for training.

• Type: int

deepspeed_config

Path to the DeepSpeed configuration file.

• Type: Union[Path, str]

Examples

>>> options = DeepSpeedTrainerOptions(
...     resume=True,
...     seed=42,
...     train_dtype='fp16',
...     log_interval=100,
...     output_dir='./output',
...     max_epoch=10,
...     deepspeed_config='./deepspeed_config.json'
... )
>>> print(options)
DeepSpeedTrainerOptions(resume=True, seed=42, train_dtype='fp16',
... log_interval=100, output_dir=PosixPath('output'),
... max_epoch=10, deepspeed_config=PosixPath('deepspeed_config.json'))

deepspeed_config

*: Path | str*

log_interval

*: int | [None](../asr/AbsDecoder.md#espnet2.asr.decoder.abs_decoder.AbsDecoder.None)*

max_epoch

*: int*

output_dir

*: Path | str*

resume

*: bool*

seed

*: int*

train_dtype

*: str | dtype*