aitune.torch.module.exact_sample_metadata | NVIDIA AITune Documentation

Exact sample metadata for matching tensor shapes.

Module Contents

Classes

Name	Description
`ExactSampleMetadata`	Contrary to SampleMetadata, ExactSampleMetadata requires exact tensor shapes to match.

API

class aitune.torch.module.exact_sample_metadata.ExactSampleMetadata()

Bases: SampleMetadata

Contrary to SampleMetadata, ExactSampleMetadata requires exact tensor shapes to match.

This is useful for recording inputs and bucketing data by tensor shapes. This class has only one factory function: from_inputs and defines equality and hash based on locator and tensor shapes.

aitune.torch.module.exact_sample_metadata.ExactSampleMetadata.__eq__(
    __value: object
) -> bool

Equality operator.

aitune.torch.module.exact_sample_metadata.ExactSampleMetadata.__hash__() -> int

Compute hash of sample metadata.

aitune.torch.module.exact_sample_metadata.ExactSampleMetadata.from_inputs(
    inputs: dict[str, typing.Any]
) -> aitune.torch.module.exact_sample_metadata.ExactSampleMetadata

staticmethod

Create ExactSampleMetadata from inputs keyed by forward parameter name.