dela
584f7e9e73
Standalone tree split from LLMRL/projects/kda. Includes Triton dt_bias backward fix, train_k3 --preset 0.5b, SFT, Docker runtime, and tests.
6 lines
153 B
Python
6 lines
153 B
Python
"""Vendored FLA WY recompute used by the chunk KDA backward."""
|
|
|
|
from kda._fla.ops.kda.wy_fast import recompute_w_u_fwd
|
|
|
|
__all__ = ["recompute_w_u_fwd"]
|