aitune.torch.backend.tensorrt.tensorrt_profile

TensorRT Profile for specifying optimization profiles.

Module Contents

Classes

Name	Description
`TensorRTProfile`	Class for representing a TensorRT optimization profile.

Data

logger

API

class aitune.torch.backend.tensorrt.tensorrt_profile.TensorRTProfile()

Class for representing a TensorRT optimization profile.

This class provides an interface for defining optimization profiles for TensorRT engines with dynamic shapes.

_profile

= Profile()

profile

Profile

Get the underlying Polygraphy Profile.

aitune.torch.backend.tensorrt.tensorrt_profile.TensorRTProfile.__eq__(
    other: aitune.torch.backend.tensorrt.tensorrt_profile.TensorRTProfile
) -> bool

Check if two TensorRTProfiles are equal.

aitune.torch.backend.tensorrt.tensorrt_profile.TensorRTProfile.__hash__() -> int

Hash the TensorRTProfile.

aitune.torch.backend.tensorrt.tensorrt_profile.TensorRTProfile.__repr__() -> str

Return the official string representation of the profile.

Returns: str

Official string representation

aitune.torch.backend.tensorrt.tensorrt_profile.TensorRTProfile.__str__() -> str

Return string representation of the profile.

Returns: str

String representation

aitune.torch.backend.tensorrt.tensorrt_profile.TensorRTProfile.add_input_shape(
    path: aitune.torch.module.forward_signature.ForwardInputPath,
    min_shape: tuple[int, ...],
    opt_shape: tuple[int, ...],
    max_shape: tuple[int, ...]
) -> aitune.torch.backend.tensorrt.tensorrt_profile.TensorRTProfile

Add a shape binding to the profile.

Parameters:

path

ForwardInputPath

Forward parameter path identifying the input tensor

min_shape

tuple[int, ...]

The minimum shape the profile will support

opt_shape

tuple[int, ...]

The shape for which TensorRT will tune the engine

max_shape

tuple[int, ...]

The maximum shape the profile will support

Returns: TensorRTProfile

The profile object for chaining

aitune.torch.backend.tensorrt.tensorrt_profile.TensorRTProfile.from_dict(
    data: dict
) -> aitune.torch.backend.tensorrt.tensorrt_profile.TensorRTProfile

classmethod

Create TensorRTProfile from dictionary.

aitune.torch.backend.tensorrt.tensorrt_profile.TensorRTProfile.profile_to_dict(
    profile: polygraphy.backend.trt.Profile
) -> dict

classmethod

Convert Polygraphy Profile to dictionary.

aitune.torch.backend.tensorrt.tensorrt_profile.TensorRTProfile.to_dict() -> dict

Convert TensorRTProfile to dictionary.

aitune.torch.backend.tensorrt.tensorrt_profile.logger = logging.getLogger(__name__)