vllm-ascend-operator-fusion

Analyze vLLM-Ascend operator graphs to identify fusion opportunities and define fused-operator interfaces.

258|48|Updated Jun 22, 2020
One-click install
npx skills add https://github.com/mindspore-ai/akg --skill vllm-ascend-operator-fusion
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: vllm-ascend-operator-fusion
Source: https://github.com/mindspore-ai/akg/tree/main/akg_agents/workspace/.opencode/skills/vllm-ascend-operator-fusion
Command: npx skills add https://github.com/mindspore-ai/akg --skill vllm-ascend-operator-fusion

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Analyses and maps operator paths in vLLM-Ascend models to identify fusion opportunities and specify new fused-operator interfaces that guide downstream generation and code updates.

Core Features & Use Cases

  • Fusion opportunity discovery: Identify fusion patterns across operator graphs and backends.
  • Interface specification: Propose concrete input/output contracts for fused kernels.
  • Guidance for changes: Outline necessary file modifications and assets to enable fusion kernel generation.

Quick Start

Provide a fusion analysis for a given vLLM-Ascend model path and outline the required file changes and new fused-operator interfaces.

Frequently Asked Questions about vllm-ascend-operator-fusion

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I identify operator fusion opportunities in vLLM-Ascend models?

To define new fused-operator interfaces, specify concrete input and output contracts for fused kernels, outline necessary file modifications, and provide clear patch points to enable downstream fusion kernel generation.

What do I need to provide to start analyzing vLLM-Ascend operator paths for kernel fusion?

To start analyzing vLLM-Ascend operator paths for kernel fusion, you need to provide model execution graphs, operator patterns, and patch points to map paths and specify new fused-operator interfaces.

Does vLLM-Ascend operator fusion analysis support all model backends?

vLLM-Ascend operator fusion analysis applies to models supported by vLLM-Ascend across backends, ensuring broad identification of fusion opportunities and interface specification for various execution graphs.

What is the best way to plan kernel fusion and downstream code updates for vLLM-Ascend?

The best way to plan kernel fusion and downstream code updates for vLLM-Ascend is to map operator paths, define interface contracts, and specify file changes to guide downstream fusion kernel generation.

Why do I need to specify interface contracts for fused kernels in vLLM-Ascend?

You need to specify interface contracts for fused kernels in vLLM-Ascend to establish concrete input and output boundaries, which are required to guide downstream code updates and enable fusion kernel generation.