Initial K3 snapshot: 0.5B KDA/MLA/MoE train path

Standalone tree split from LLMRL/projects/kda. Includes Triton dt_bias backward fix, train_k3 --preset 0.5b, SFT, Docker runtime, and tests.
2026-08-25 14:43:17 +08:00
commit 584f7e9e73
140 changed files with 21592 additions and 0 deletions
@@ -0,0 +1,9 @@
+# Vendored FLA KDA kernels
+
+Subset of [flash-linear-attention](https://github.com/fla-org/flash-linear-attention)
+used by `kda.ops` `backend="triton"`.
+
+- License: MIT (see `LICENSE`)
+- Upstream version tag in `__init__.py`
+- Import path is `kda._fla.*`, not `fla.*`
+- Not included: context parallel, Ascend, TileLang, `flash_kda`, non-KDA ops