Enum TorchStrategy 

pub enum TorchStrategy {
    Cuda {
        os: Os,
        driver_version: Version,
        source: TorchSource,
    },
    Amd {
        os: Os,
        gpu_architecture: AmdGpuArchitecture,
        source: TorchSource,
    },
    Xpu {
        os: Os,
        source: TorchSource,
    },
    Backend {
        backend: TorchBackend,
        source: TorchSource,
    },
}

The strategy to use when determining the appropriate PyTorch index.

Variants

Cuda

Select the appropriate PyTorch index based on the operating system and CUDA driver version (e.g., 550.144.03).

Fields

os: Os

driver_version: Version

source: TorchSource

Amd

Select the appropriate PyTorch index based on the operating system and AMD GPU architecture (e.g., gfx1100).

Fields

os: Os

gpu_architecture: AmdGpuArchitecture

source: TorchSource

Xpu

Select the appropriate PyTorch index based on the operating system and Intel GPU presence.

Fields

os: Os

source: TorchSource

Backend

Use the specified PyTorch index.

Fields

backend: TorchBackend

source: TorchSource

Implementations

impl TorchStrategy

pub fn from_mode( mode: TorchMode, source: TorchSource, os: &Os, ) -> Result<Self, AcceleratorError>

Determine the TorchStrategy from the given TorchMode, Os, and Accelerator.

pub fn applies_to(&self, package_name: &PackageName) -> bool

Returns true if the TorchStrategy applies to the given PackageName.

pub fn has_system_dependency(&self, package_name: &PackageName) -> bool

Returns true if the given PackageName has a system dependency (e.g., CUDA or ROCm).

For example, triton is hosted on the PyTorch indexes, but does not have a system dependency on the associated CUDA version (i.e., the triton on the cu128 index doesn’t depend on CUDA 12.8).

pub fn index_urls(&self) -> impl Iterator<Item = &IndexUrl>

Return the appropriate index URLs for the given TorchStrategy.

Trait Implementations

impl Clone for TorchStrategy

fn clone(&self) -> TorchStrategy

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for TorchStrategy

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl PartialEq for TorchStrategy

fn eq(&self, other: &TorchStrategy) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl Eq for TorchStrategy

impl StructuralPartialEq for TorchStrategy