aitune.torch.config | NVIDIA AITune Documentation

Inplace configuration.

Module Contents

Classes

Name	Description
`AITuneConfig`	AITune configuration.
`AITuneMode`	AITune execution mode used for tuning.

Functions

Name	Description
`aitune_cache_dir`	Configure cache dir location based on environment variable.

Data

DEFAULT_DEVICE

DEFAULT_DEVICE_AFTER_TUNING

DEFAULT_MAX_NUM_SAMPLES_STORED

DEFAULT_MIN_NUM_SAMPLES

DEFAULT_TUNING_DATA_OUTPUT_PATH

config

API

class aitune.torch.config.AITuneConfig()

AITune configuration.

_cache_dir

Path = aitune_cache_dir()

_min_num_samples

int = DEFAULT_MIN_NUM_SAMPLES

_tuning_data_output_path

Path

cache_dir

Path

Get the cache directory.

device_after_tuning

str = DEFAULT_DEVICE_AFTER_TUNING

enable_diffusers_integration

bool = AITUNE_DIFFUSERS_INTEGRATION

enable_transformers_integration

bool = AITUNE_TRANSFORMERS_INTEGRATION

max_num_samples_stored

int | float = DEFAULT_MAX_NUM_SAMPLES_STORED

min_num_samples

int

Get the minimum number of samples to collect before optimizing.

strict_mode

bool = True

tuning_data_output_path

Path

Get the output path for tuning data.

aitune.torch.config.AITuneConfig.to_dict() -> dict[str, typing.Any]

Return all configuration attributes keyed by their public name.

class aitune.torch.config.AITuneMode

Bases: enum.Enum

AITune execution mode used for tuning.

DECLARATIVE

= 'DECLARATIVE'

JIT

= 'JIT'

aitune.torch.config.aitune_cache_dir() -> pathlib.Path

Configure cache dir location based on environment variable.

Returns: Path

Cache dir from AITUNE_CACHE_DIR environment variable, or default.

aitune.torch.config.DEFAULT_DEVICE = 'cuda:0'

aitune.torch.config.DEFAULT_DEVICE_AFTER_TUNING = 'meta'

aitune.torch.config.DEFAULT_MAX_NUM_SAMPLES_STORED = 1

aitune.torch.config.DEFAULT_MIN_NUM_SAMPLES = 100

aitune.torch.config.DEFAULT_TUNING_DATA_OUTPUT_PATH = _AITUNE_CACHE_DIR / 'tuning_data' / 'report.json'

aitune.torch.config.config = AITuneConfig()