File size: 1,976 Bytes
b66f552 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 | # Copyright (c) 2023-2025, Songlin Yang, Yu Zhang
import torch
from fla.ops.linear_attn.utils import normalize_output
from fla.ops.simple_gla import fused_chunk_simple_gla
@torch.compiler.disable
def fused_chunk_linear_attn(
q: torch.Tensor,
k: torch.Tensor,
v: torch.Tensor,
scale: float | None = None,
initial_state: torch.Tensor | None = None,
output_final_state: bool = False,
normalize: bool = True,
cu_seqlens: torch.LongTensor | None = None,
) -> tuple[torch.Tensor, torch.Tensor]:
r"""
Args:
q (torch.Tensor):
queries of shape `[B, T, H, K]`.
k (torch.Tensor):
keys of shape `[B, T, H, K]`.
v (torch.Tensor):
values of shape `[B, T, H, V]`.
scale (Optional[float]):
Scale factor for linear attention scores.
If not provided, it will default to `1 / sqrt(K)`. Default: `None`.
initial_state (Optional[torch.Tensor]):
Initial state of shape `[B, H, K, V]`. Default: `None`.
output_final_state (Optional[bool]):
Whether to output the final state of shape `[B, H, K, V]`. Default: `False`.
normalize (bool):
Whether to normalize the output. Default: `True`.
cu_seqlens (torch.LongTensor):
Cumulative sequence lengths of shape `[N+1]` used for variable-length training,
consistent with the FlashAttention API.
Returns:
o (torch.Tensor):
Outputs of shape `[B, T, H, V]`.
final_state (torch.Tensor):
Final state of shape `[B, H, K, V]` if `output_final_state=True` else `None`
"""
o, final_state = fused_chunk_simple_gla(
q=q,
k=k,
v=v,
scale=scale,
initial_state=initial_state,
output_final_state=output_final_state,
cu_seqlens=cu_seqlens,
)
if normalize:
o = normalize_output(q * scale, k, o)
return o, final_state
|