triton-ascend-attention

Community

Optimize attention ops on Ascend with Triton.

Authorxchang1121
Version1.0.0
Installs0

System Documentation

What problem does it solve?

Optimizes Transformer-style attention workloads on Ascend by applying Triton-based kernel optimizations, including QKV tiling, online softmax, and masking strategies.

Core Features & Use Cases

  • QKV tiling and block-wise computation to reduce memory footprint and latency in attention operations.
  • Online softmax and masking techniques for efficient causal and masked attention on large sequences.
  • Flash Attention tiling strategies to boost throughput for multi-head attention on Ascend hardware.

Quick Start

Run the Triton-Ascend optimized attention workflow on Ascend hardware for Transformer-based models.

Dependency Matrix

Required Modules

None required

Components

Standard package

💻 Claude Code Installation

Recommended: Let Claude install automatically. Simply copy and paste the text below to Claude Code.

Please help me install this Skill:
Name: triton-ascend-attention
Download link: https://github.com/xchang1121/AutoResearch-CC-hook/archive/main.zip#triton-ascend-attention

Please download this .zip file, extract it, and install it in the .claude/skills/ directory.
View Source Repository

Agent Skills Search Helper

Install a tiny helper to your Agent, search and equip skill from 471,000+ vetted skills library on demand.