Triton Fused Normalizations banner
anthony-maio anthony-maio

Triton Fused Normalizations

Development community

Description

--- name: triton-fused-normalizations description: Implement fused LayerNorm, RMSNorm, and GroupNorm kernels (forward & backward) in Triton with single-pass statistics and two-stage reductions. --- # Fused Normalization Kernels in Triton (LayerNorm, RMSNorm, GroupNorm) > **Targets:** Triton >= 2.1, SM70+/CDNA2+ ## Overview Fused normalization kernels compute statistics and apply normalization in a single pass, avoiding extra HBM round-trips. Map one program per normalization "row" (per token

Installation

Installs to ~/.claude/skills/anthony-maio-triton-skills-triton-fused-normalizations/SKILL.md

Terminal
mkdir -p ~/.claude/skills/anthony-maio-triton-skills-triton-fused-normalizations && curl -fsSL https://raw.githubusercontent.com/anthony-maio/triton-skills/HEAD/skills/triton-fused-normalizations.md -o ~/.claude/skills/anthony-maio-triton-skills-triton-fused-normalizations/SKILL.md

Restart Claude Code, or start a new session, for it to be picked up.

Full documentation available on GitHub

View Source Repository