Standalone tree split from LLMRL/projects/kda. Includes Triton dt_bias backward fix, train_k3 --preset 0.5b, SFT, Docker runtime, and tests.
8 lines
274 B
Python
8 lines
274 B
Python
"""Local Triton KDA kernels vendored from FLA chunk_{fwd,intra,bwd,wy,gate}."""
|
|
|
|
from .chunk import ChunkKDAFunction, chunk_kda
|
|
from .chunk_fwd import chunk_kda_fwd
|
|
from .gate import kda_gate_fwd
|
|
|
|
__all__ = ["ChunkKDAFunction", "chunk_kda", "chunk_kda_fwd", "kda_gate_fwd"]
|