Metis

An AI Inference Platform for running AI models as production services.

Launch serverless endpoints with ready-to-use inference engines, build inference infrastructure with workloads, templates, and volumes, and evaluate endpoint performance with Benchmark. Administrators operate resources with Kueue scheduling and cluster monitoring.

Model training is done in Maxis, and AI assets such as models and datasets are managed in Signum Hub.

Components

AI Inference

Deploy models as serverless endpoints (vLLM and more), and operate your inference environment with workloads, templates, volumes, and Benchmark.

→

AI Inference Admin

View all endpoints and workloads from an administrator perspective, and manage Kueue scheduling, monitoring, and projects.

→

Metis Container

Manage workloads, networking, storage, and policies on a shared Kubernetes environment.

→

Quick Start

If you're new to Metis, try these recommended guides!

1. Deploying a Serverless Endpoint in AI Inference

a. Go to Serverless in AI Inference.
b. Select the model and inference engine to deploy, and configure resources.
c. Create the endpoint and check that its status becomes running.
d. Send a request to the issued endpoint and check the response.

2. Deploying Workloads in AI Inference Using Hub Assets

a. Check available models, datasets, and packages in Hub.
b. Go to Infrastructure in AI Inference and create a workload.
c. Select resources from Hub along with required volumes, images, and execution templates.
d. Deploy the workload and monitor execution status and logs.