FreeToken brings datacenter-scale model serving to your desktop. Run massive models locally, fast and efficiently.
fix(fla): stop l2norm recompiling per token count
Xiaoze Fanf7c31e92dbf2
diff --git a/python/freetoken/kernel/fla/l2norm.py b/python/freetoken/kernel/fla/l2norm.pyindex b7153120..f6ebd6cd 100644--- a/python/freetoken/kernel/fla/l2norm.py+++ b/python/freetoken/kernel/fla/l2norm.py@@ -51,13 +51,12 @@ def l2norm_fwd_kernel1(# ],# key=["D", "NB"],# )[email protected](do_not_specialize=["T"])def l2norm_fwd_kernel(x,y,eps,- NB: tl.constexpr,- T: tl.constexpr,+ T,D: tl.constexpr,BT: tl.constexpr,BD: tl.constexpr,@@ -91,7 +90,6 @@ def l2norm_fwd(raise RuntimeError("This layer doesn't support feature dim >= 64KB.")if D <= 512:- NB = triton.cdiv(T, 2048)def grid(meta):return (triton.cdiv(T, meta["BT"]),)@@ -100,7 +98,6 @@ def l2norm_fwd(x,y,eps,- NB=NB,T=T,D=D,BD=BD,