anthony-maio

Triton Memory Efficient Patterns

Name: Triton Memory Efficient Patterns
Author: anthony-maio

AI community

SOURCE https://github.com/anthony-maio/triton-skills/blob/main/skills/triton-memory-efficient-patterns.md

Description

--- name: triton-memory-efficient-patterns description: Teach an AI agent to minimize GPU memory via seed-based PRNG, fusion, and recomputation in Triton kernels. --- # Memory-efficient Triton kernels: seed PRNG, fusion, and recomputation > **Targets:** Triton >= 2.1, SM70+/CDNA2+ Overview This guide describes patterns for minimizing GPU memory footprint in Triton kernels: Philox seed-based PRNG (generate dropout masks on-the-fly), activation checkpointing via recomputation, fused elementwise

Installation

Installs to ~/.claude/skills/anthony-maio-triton-skills-triton-memory-efficient-patterns/SKILL.md

Terminal

mkdir -p ~/.claude/skills/anthony-maio-triton-skills-triton-memory-efficient-patterns && curl -fsSL https://raw.githubusercontent.com/anthony-maio/triton-skills/HEAD/skills/triton-memory-efficient-patterns.md -o ~/.claude/skills/anthony-maio-triton-skills-triton-memory-efficient-patterns/SKILL.md

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