Skip to content

Table of Contents

anyGPT.config.settings

SimpleConfig Objects

class SimpleConfig()

SimpleConfig - base class for config dataclass objects

__init__

def __init__(**kwargs)

Initializes dataclass based on keys in the yaml dict.

Arguments:

  • kwargs: a dictionary sourced from a yaml config file.

update

def update(kwargs)

Updates the values in the dataclass.

Arguments:

  • kwargs: dictionary of key/value pairs to update.

TorchConfig Objects

@dataclass
class TorchConfig(SimpleConfig)

Torch configuration.

backend

Specifies which backend to use. Default='nccl'. Currently disabled.

device

Specifies which device to use. Default=cuda. Options are 'cpu', 'cuda'

precision

Specifies the precision used during training. Acceptable values are 32, 16-mixed, bf16-mixed, or 64.

compile

Specifies whether to compile the model for training. Default=true

accelerator

Specifies which accelerator to use. Default=auto.