Standalone tree split from LLMRL/projects/kda. Includes Triton dt_bias backward fix, train_k3 --preset 0.5b, SFT, Docker runtime, and tests.
344 B
344 B
Vendored FLA KDA kernels
Subset of flash-linear-attention
used by kda.ops backend="triton".
- License: MIT (see
LICENSE) - Upstream version tag in
__init__.py - Import path is
kda._fla.*, notfla.* - Not included: context parallel, Ascend, TileLang,
flash_kda, non-KDA ops