For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /get-started.md).
Mojo function
warp_sums
def warp_sums(x: Array[SIMD[.bfloat16, 8], Int(4)], wide: Bool) -> SIMD[.float32, 4]
The row's per-warp sums of squares as rms_norm's block tier forms them: under SM-count rows (wide, 1024 threads) one per piece, else one over the thread's pieces in element 0. Every lane holds them.
Returns: