Orchestrate Kubernetes InferenceService workloads by reconciling CRDs into Deployments or LeaderWorkerSets.
npx skills add https://github.com/moreh-dev/mif --skill guide-odin
Please help me install this Agent Skill. Skill: guide-odin Source: https://github.com/moreh-dev/mif/tree/main/skills/guide-odin Command: npx skills add https://github.com/moreh-dev/mif --skill guide-odin
Odin orchestrates and reconciles InferenceService workloads in Kubernetes, enabling scalable, repeatable deployment of AI inference workloads via a reusable, template-driven configuration system.
Run Odin in your cluster to deploy your first InferenceService using the vLLM presets and templates.
High-intent search queries and answers about installing and using this skill.
To deploy a vLLM InferenceService on Kubernetes, Odin reconciles your custom resources into Deployments or LeaderWorkerSets using reusable templates and presets for scalable AI inference workloads.
Deployment and LeaderWorkerSet topologies in Odin dictate workload layout for inference. LeaderWorkerSets support multi-node setups and parallelism patterns, while standard Deployments fit simpler single-node configurations.
Odin resolves InferenceServiceTemplate references by layering runtime-bases and presets, applying variable substitution and merge semantics to cleanly separate platform configurations from model-specific settings.
Yes, InferenceService supports tensor, pipeline, data, and expert parallelism styles. Odin reconciles these configurations to correctly provision the underlying Kubernetes workload topology and worker templates.
Odin InferenceService reconciliation failures often stem from invalid template references, incorrect merge semantics, or webhook validation errors during runtime-base selection and workload topology configuration.
Yes, LeaderWorkerSet is required for multi-node vLLM serving and PD-disaggregation workflows. Odin uses this topology to manage complex worker layouts and rollout strategies for distributed inference.
truongnat...
automateyournetworkGenerate Visio network diagrams from CDP/LLDP data and upload to SharePoint.
hsliuustc0106Distribute large models across GPUs or nodes with tensor and pipeline parallelism.
ayesha-aziz123Generate Helm chart scaffolds for Kubernetes frontend and backend deployments.
