fastvideo.v1.platforms.interface

classmethod get_attn_backend_cls(selected_backend: Optional[fastvideo.v1.platforms.interface.AttentionBackendEnum], head_size: int, dtype: torch.dtype) → str[source]#: Get the attention backend class of a device.

abstract classmethod get_current_memory_usage(device: Optional[torch.types.Device] = None) → float[source]#: Return the memory usage in bytes.

classmethod get_device_capability(device_id: int = 0) → Optional[fastvideo.v1.platforms.interface.DeviceCapability][source]#: Stateless version of :func:torch.cuda.get_device_capability.

classmethod get_device_communicator_cls() → str[source]#: Get device specific communicator class for distributed communication.

abstract classmethod get_device_name(device_id: int = 0) → str[source]#: Get the name of a device.

abstract classmethod get_device_total_memory(device_id: int = 0) → int[source]#: Get the total memory of a device in bytes.

abstract classmethod get_device_uuid(device_id: int = 0) → str[source]#: Get the uuid of a device, e.g. the PCI bus ID.

classmethod has_device_capability(capability: Union[Tuple[int, int], int], device_id: int = 0) → bool[source]#

Test whether this platform is compatible with a device capability.

The capability argument can either be:

A tuple (major, minor).
An integer <major><minor>. (See :meth:DeviceCapability.to_int)

classmethod inference_mode()[source]#

A device-specific wrapper of torch.inference_mode.

This wrapper is recommended because some hardware backends such as TPU do not support torch.inference_mode. In such a case, they will fall back to torch.no_grad by overriding this method.

abstract classmethod is_async_output_supported(enforce_eager: Optional[bool]) → bool[source]#: Check if the current platform supports async output.

is_cpu() → bool[source]#

is_cuda() → bool[source]#

is_cuda_alike() → bool[source]#: Stateless version of :func:torch.cuda.is_available.

is_out_of_tree() → bool[source]#

is_rocm() → bool[source]#

is_tpu() → bool[source]#

classmethod seed_everything(seed: Optional[int] = None) → None[source]#

Set the seed of each random module. torch.manual_seed will set seed on all devices.

Loosely based on: https://github.com/Lightning-AI/pytorch-lightning/blob/2.4.0/src/lightning/fabric/utilities/seed.py#L20

simple_compile_backend: str[source]#: ‘inductor’

supported_quantization: list[str][source]#: []

classmethod verify_model_arch(model_arch: str) → None[source]#

Verify whether the current platform supports the specified model architecture.

This will raise an Error or Warning based on the model support on the current platform.
By default all models are considered supported.

classmethod verify_quantization(quant: str) → None[source]#: Verify whether the quantization is supported by the current platform.

class fastvideo.v1.platforms.interface.PlatformEnum[source]#

Bases: enum.Enum

CPU[source]#: ‘auto(…)’

CUDA[source]#: ‘auto(…)’

OOT[source]#: ‘auto(…)’

ROCM[source]#: ‘auto(…)’

TPU[source]#: ‘auto(…)’

UNSPECIFIED[source]#: ‘auto(…)’

class fastvideo.v1.platforms.interface.UnspecifiedPlatform[source]#

Bases: fastvideo.v1.platforms.interface.Platform

device_type = <Multiline-String>[source]#

fastvideo.v1.platforms.interface.logger[source]#: ‘init_logger(…)’

fastvideo.v1.platforms.interface

Contents

fastvideo.v1.platforms.interface#

Module Contents#

Classes#

Data#

API#

`fastvideo.v1.platforms.interface`#