quaterion.train.cache.cache_config module

class CacheConfig(cache_type: ~quaterion.train.cache.cache_config.CacheType | None = CacheType.AUTO, mapping: ~typing.Dict[str, ~quaterion.train.cache.cache_config.CacheType] = <factory>, key_extractors: KeyExtractorType | ~typing.Dict[str, KeyExtractorType] = <factory>, batch_size: int | None = 32, num_workers: int | None = None, save_dir: str | None = None)

Bases: object

Determine cache settings.

This class should be passed to configure_caches()

batch_size: int | None = 32

Batch size to be used in CacheDataLoader during caching process. It does not affect others training stages.

cache_type: CacheType | None = 'auto'

Cache type used for cacheable encoders not set in mapping

key_extractors: KeyExtractorType | Dict[str, KeyExtractorType]

Mapping of encoders to key extractor functions required to cache non-hashable objects.

mapping: Dict[str, CacheType]

Mapping of encoder_name to CacheType

num_workers: int | None = None

Num of workers to be used in CacheDataLoader during caching process. It does not affect others training stages.

save_dir: str | None = None

If provided, cache fill be saved to the given directory and re-used between launches

class CacheType(value)

Bases: str, Enum

Available tensor devices to be used for caching.

AUTO = 'auto'

Use CUDA if it is available, else use CPU.

CPU = 'cpu'

Tensors device is CPU.

GPU = 'gpu'

Tensors device is GPU.

NONE = 'none'

Disable cache

KeyExtractorType

Type of function to extract hash value from the input object. Required if there is no other way to distinguish values for caching

alias of Callable[[Any], Hashable]