perf(sm120): skip split-K when the reduction is slower than the single-token GEMM - #77

ormandj wants to merge 1 commit into

perf(sm120): skip split-K when the reduction is slower than the single-token GEMM#77
ormandj wants to merge 1 commit into
sgl-project:devfrom
ormandj:fix/sm120-splitk-tiny-m

sm120: skip split-K for single-token decode shapes