fastvideo.v1.attention.backends.sage_attn

fastvideo.v1.attention.backends.sage_attn#

Module Contents#

Classes#

SageAttentionBackend

SageAttentionImpl

Data#

logger

API#

class fastvideo.v1.attention.backends.sage_attn.SageAttentionBackend[source]#

Bases: fastvideo.v1.attention.backends.abstract.AttentionBackend

accept_output_buffer: bool[source]#

True

static get_impl_cls() → Type[fastvideo.v1.attention.backends.sage_attn.SageAttentionImpl][source]#
static get_name() → str[source]#
static get_supported_head_sizes() → List[int][source]#
class fastvideo.v1.attention.backends.sage_attn.SageAttentionImpl(num_heads: int, head_size: int, causal: bool, softmax_scale: float, num_kv_heads: Optional[int] = None, prefix: str = '', **extra_impl_args)[source]#

Bases: fastvideo.v1.attention.backends.abstract.AttentionImpl

forward(query: torch.Tensor, key: torch.Tensor, value: torch.Tensor, attn_metadata: fastvideo.v1.attention.backends.abstract.AttentionMetadata) → torch.Tensor[source]#
fastvideo.v1.attention.backends.sage_attn.logger[source]#

‘init_logger(…)’

Contents