skilld
All skills
microsoft avatar

/airunway-aks-setup

@d0c21c0 official
by microsoftmicrosoft/github-copilot-for-azure253 stars
206

Set up AI Runway on AKS — from bare cluster to running model. Covers cluster verification, controller install, GPU assessment, provider setup, and first deployment. WHEN: "setup AI Runway", "onboard AKS cluster", "install AI Runway", "airunway setup", "deploy model to AKS", "GPU inference on AKS", "KAITO setup on AKS", "run LLM on AKS", "vLLM on AKS", "set up model serving on AKS", "AI Runway controller".

Ask your Agent

Use this Skill: https://skilld.dev/gh/microsoft/github-copilot-for-azure/airunway-aks-setup

This session only. Nothing lands on disk.

referencesgpu-profiles.md

≈642 tokens on demand. Your agent reads this file only when SKILL.md points to it.

GPU Compatibility Reference

This reference is used by the airunway-aks-setup skill during Step 3 — GPU Assessment to match detected GPU hardware to known compatibility profiles and surface constraints before model deployment.

GPU Profiles

GPU Model VRAM (GB) bfloat16 float16 Attention Backends Compute Capability Notes
T4 16 No Yes XFORMERS 7.5 No bfloat16 — must use float16 or float32.
V100 (16 GB) 16 No Yes XFORMERS 7.0 Limited flash attention. Prefer xformers.
V100 (32 GB) 32 No Yes XFORMERS 7.0 Same dtype constraints as 16 GB variant.
A10 24 Yes Yes FLASH_ATTN, XFORMERS 8.6 Single-slot Ampere GPU.
A10G 24 Yes Yes FLASH_ATTN, XFORMERS 8.6 Good general-purpose GPU. Common in AWS.
L4 24 Yes Yes FLASH_ATTN, XFORMERS 8.9 Inference-optimized. Common in GCP and Azure.
L40S 48 Yes Yes FLASH_ATTN, TRITON_ATTN, XFORMERS 8.9 Ada Lovelace. Growing availability on Azure.
A100 (40 GB) 40 Yes Yes FLASH_ATTN, TRITON_ATTN, XFORMERS 8.0 High-performance training and inference.
A100 (80 GB) 80 Yes Yes FLASH_ATTN, TRITON_ATTN, XFORMERS 8.0 Recommended for large models (70B+).
H100 80 Yes Yes FLASH_ATTN, TRITON_ATTN, XFORMERS 9.0 Highest single-GPU performance.
H200 (SXM) 141 Yes Yes FLASH_ATTN, TRITON_ATTN, XFORMERS 9.0 Maximum memory (HBM3e).
H20 96 Yes Yes FLASH_ATTN, TRITON_ATTN, XFORMERS 9.0 China-market H100 derivative. 96 GB HBM3.

Attention Backends

Backend Description Min Compute Capability
FLASH_ATTN FlashAttention-2 — fastest, lowest memory 8.0 (Ampere+)
TRITON_ATTN Triton-based attention — good performance 8.0 (Ampere+)
XFORMERS Memory-efficient attention — works on older GPUs 7.0 (Volta+)

Compatibility Warnings

Surface these warnings when the following GPUs are detected:

T4

Warning: T4 GPUs do not support bfloat16. You must configure --dtype float16 in serving arguments. Failure to do so causes errors or silent dtype casting.

V100

Warning: V100 GPUs do not support bfloat16 and have limited flash attention support. Use xformers backend and --dtype float16.

Model Sizing & Recommendations

For VRAM sizing estimates and starter model recommendations, see model-sizing.md.

Source: SKILL.md on GitHub

Third-party checks

No alerts5mo3 checks · Risk SAFE
  • Gen Agent Trust Hub5mo

    This skill facilitates the setup of AI Runway on Azure Kubernetes Service (AKS) using standard administrative tools. It incorporates security best practices, particularly regarding the handling of sensitive HuggingFace tokens during deployment. No security considerations were identified.

  • Socket5mo

    No alerts

  • Snyk5mo

    Risk: LOW · No issues

Provenance

Signed by skilld at d0c21c0. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 20 hours ago.

Activeupdated 5 months ago

Capability

metadata
{
  "author": "Microsoft",
  "version": "0.0.0-placeholder"
}
argument-hint
[skip-to-step N]

Topics

  • kubernetes
  • aks
  • azure
  • llm
  • model-serving
  • gpu
  • inference
  • kaito
  • vllm
  • mlops

README badge

README badge for microsoft/github-copilot-for-azure/airunway-aks-setup

What it does

Installs and configures AI Runway on an existing AKS cluster, walking through controller setup, GPU assessment, inference provider selection, and model deployment. Targets users deploying large language models or other GPU-accelerated workloads to Kubernetes via KAITO, Dynamo, or KubeRay.

Generated from the current SKILL.md.

Frequently asked

Do I need an existing AKS cluster to use this skill?
Yes. This skill assumes you already have an AKS cluster. If you don't, the skill will hand off to the azure-kubernetes skill to provision one first, then return here.
What inference providers does this skill support?
The skill covers KAITO, Dynamo, and KubeRay as inference provider options. Step 4 recommends and installs the appropriate provider for your setup.
Does this skill work with CPU-only clusters?
Yes, CPU-only inference is acceptable, though the skill includes GPU assessment and provider setup primarily designed for GPU workloads. A bare cluster without GPU resources is still supported.
What happens if I already have part of AI Runway set up?
You can use the `skip-to-step N` argument to resume from a specific phase. The skill will report the current state and skip already-completed steps.
What are the cost implications of using this skill?
GPU node pools incur significant charges — A100-80GB can cost $3–5+/hr. The skill will confirm you understand these costs before provisioning GPU resources.

Generated from the current SKILL.md. These answers refresh after source changes.