Skip to main content

MSP Overview

Choose a maintained path below, verify each result in the Console, and open the detailed task page when you need exact fields, commands, or troubleshooting.

Deploy Your First Model
Model Serving Platform

End-to-end platform for deploying, training, and managing AI models on your own GPU clusters.

Docs 2.0 · Model Serving Platform

What You Can Do

  • Deploy models from Hugging Face or custom sources to GPU clusters
  • Fine-tune models with SFT, DPO, or CPT methods
  • Manage GPU clusters and monitor resource utilization
  • Configure API keys, providers, storage, and alerts

Who This Is For

ML Engineers · Platform Administrators · DevOps Teams

Choose Your Path

01Recommended

Deploy a Model

Deploy pre-trained or fine-tuned models to production

  1. Upload model or import from Hugging Face
  2. Configure GPU resources
  3. Create deployment
  4. Monitor serving status

Result: Model running and serving inference requests

Start Deploying
02

Train a Model

Fine-tune models with your own datasets

  1. Prepare dataset
  2. Choose fine-tuning method (SFT/DPO/CPT)
  3. Launch training job
  4. Evaluate results

Result: Fine-tuned model ready for deployment

Start Training
03

Manage Clusters

Register and monitor GPU infrastructure

  1. Register K8s cluster
  2. Configure resource groups
  3. Monitor node health
  4. Set up alerts

Result: Healthy cluster serving workloads with observability

Manage Clusters

Before You Start

  • An active Smart Studio account with MSP access
  • At least one registered GPU cluster (for deployment/training)
  • Basic understanding of model serving concepts

Core Journey

1Upload model or import from Hugging Face2Configure GPU resources3Create deployment4Monitor serving status

Expected Result

Your AI models deployed, trained, and monitored on managed GPU infrastructure with full observability.

Recommended Next Step

Deploy Your First Model