skills.rest
alon3153
alon3153COMMUNITY
Explore all skills by this author →
GitHub

google-cloud-solution-guided-gke-ai-migration

Migrates existing AI inference workloads to self-hosted GKE serving using gcloud and kubectl.

Updated May 11, 2026
Search Tags:#migration#kubernetes#gpu#gcloud#gke#vllm#ai inference
One-click install
npx skills add https://github.com/alon3153/upe-social-publisher --skill google-cloud-solution-guided-gke-ai-migration-alon3153
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: google-cloud-solution-guided-gke-ai-migration
Source: https://github.com/alon3153/upe-social-publisher/tree/main/.agents/skills/google-cloud-solution-guided-gke-ai-migration
Command: npx skills add https://github.com/alon3153/upe-social-publisher --skill google-cloud-solution-guided-gke-ai-migration-alon3153

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes assets (resource) components.

What problem does it solve? Moving an existing AI inference workload from Cloud Run, the Gemini API, or Gemini Enterprise Agent Platform onto self-hosted GKE inference involves hardware sizing, model staging, manifest authoring, and traffic cutover decisions that are easy to get wrong. This Skill guides that migration through a structured four-phase workflow so nothing is missed. ## Core Features & Use Cases - Guided 4-Phase Migration: Discovery of the source workload via read-only gcloud inspection, solution design with a deterministic VRAM sizing formula, implementation with generated Kubernetes manifests, and validation with cutover and rollback guidance. - Opinionated Golden Path: Defaults to vLLM serving, Custom Compute Classes for accelerator provisioning, Cloud Storage FUSE or Managed Lustre for model staging, GKE Workload Identity, and Gateway API exposure. - Use Case: A team running a Gemma 2 9B model on Cloud Run wants to cut inference costs by self-hosting on GKE with L4 GPUs. The Skill inspects the Cloud Run service, calculates VRAM needs, generates vLLM Deployment, ComputeClass, storage, and Gateway manifests, and walks through staging, rollout verification, and traffic cutover. ## Quick Start Ask the assistant to migrate your existing Cloud Run or Gemini API inference workload to self-hosted GKE inference and answer the discovery questions it asks.

Frequently Asked Questions about google-cloud-solution-guided-gke-ai-migration

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

FAQPage Schema
How do I migrate a Cloud Run AI workload to GKE inference?▼

Follow the four-phase workflow: discover the existing Cloud Run configuration with read-only gcloud commands, design the target architecture with VRAM sizing, apply generated vLLM and storage manifests with kubectl, then validate the endpoint and cut over traffic.

How do I calculate GPU VRAM requirements for vLLM on GKE?▼

Use the deterministic formula: (parameters times 2 divided by quantization factor, plus KV cache overhead) times 1.2 safety margin. Compare the result against full card memory, such as 24 GB for an NVIDIA L4, to select the accelerator.

Should I use this migration skill or gke-inference for a new deployment?▼

Use this skill only when an existing workload on Cloud Run, Gemini API, Agent Platform, or a VM must be migrated. For brand new GKE inference deployments with no source workload, use the gke-inference skill instead.

Can I automate the GKE migration with Gemini Cloud Assist MCP?▼

This skill covers manual, architect-guided migration only. If you want automated infrastructure analysis or resource mutation, the workflow stops and directs you to the Gemini Cloud Assist MCP server documentation and setup codelab.

Why does my vLLM pod fail to mount the GCS model bucket?▼

Every pod mounting a Cloud Storage FUSE volume needs the gke-gcsfuse/volumes annotation and a Kubernetes ServiceAccount bound via Workload Identity to a Google service account with roles/storage.objectUser on the bucket. Check both before other troubleshooting.

How should Hugging Face tokens be handled for gated models on GKE?▼

Never write token values into manifests. Create the secret directly with kubectl create secret generic hf-secret before applying workloads, reference it via env.valueFrom.secretKeyRef, and revoke any token that was pasted into a conversation.

Related Skills

Explore More Software Engineering Skills→
googlegoogle
19.1k

google-cloud-solution-guided-gke-ai-migration

Migrates existing AI inference workloads to self-hosted GKE inference using gcloud and kubectl.

Official
Advanced
wangx7wangx7

gke-scaling

Configure GKE autoscaling with HPA, VPA, and Node Auto-Provisioning.

Community
Intermediate
handsomelong922handsomelong922

serving-llms-vllm

Deploy scalable LLM serving with vLLM for OpenAI-compatible endpoints.

Community
Advanced
googlegoogle
17.1k

gke-inference

Automate AI/ML inference deployment and optimization on Google Kubernetes Engine.

Official
Advanced