Skip to content

Fine Tuning as a Service using Rafay and Unsloth Studio

Fine-tuning large language models used to be an exercise reserved for teams with deep MLOps expertise and bespoke infrastructure. With Unsloth Studio — an open-source web UI for training and running LLMs — the barrier to entry has dropped considerably.

But packaging Unsloth Studio into a repeatable, self-service experience that neo clouds and enterprise can offer their end users? That still requires thoughtful orchestration.

In this post, we walk through how to deliver Unsloth Studio as a one-click, app-store-style experience using Rafay's App Marketplace. By the end, you'll understand how to create an Unsloth Studio App SKU, configure it for end users, test it, and share it across customer organizations — all without requiring your users to know anything about Kubernetes, Docker, or GPU scheduling.

Unsloth Studio in Rafay

Why Unsloth Studio?

Unsloth Studio is a web-based interface for running and fine-tuning open-source language models like Qwen, DeepSeek, Llama, Gemma, and hundreds more. It supports features such as model downloading, LoRA-based fine-tuning with significantly reduced VRAM requirements, GGUF model export, tool calling, and interactive chat — all from a browser.

The official Docker image (unsloth/unsloth) bundles everything needed: the Unsloth training framework, Jupyter Lab, and the Studio web UI. Running it standalone is straightforward:

docker run -d -e JUPYTER_PASSWORD="mypassword" \
  -p 8888:8888 -p 8000:8000 -p 2222:22 \
  -v $(pwd)/work:/workspace/work \
  --gpus all \
  unsloth/unsloth

The Studio's web interface is accessible on port 8000, while Jupyter Lab remains available on port 8888. But running a one-off Docker container is very different from delivering a governed, multi-tenant, self-service experience to dozens or hundreds of AI practitioners across your organization. That's where Rafay comes in.

The Rafay App Marketplace

The Rafay Platform includes an App Marketplace that lets platform administrators take any Docker-based application and turn it into a self-service SKU. The end user experience is comparable to an app store: users browse available apps, click deploy, and get a fully provisioned instance — complete with a URL, TLS termination, and GPU access — in under a minute.

Under the hood, the platform handles container orchestration on Kubernetes, namespace isolation, ingress configuration, and lifecycle management. Administrators define the experience once in Rafay's PaaS Studio, and end users consume it without touching any infrastructure.

Unsloth Studio in Rafay

Important

Navigate to our detailed guide that captures "step-by-step" instructions for details on prerequisites, requirements etc

The End User Experience

From the end user's perspective, the entire workflow looks like this:

  1. Log in to the Rafay Self Service Portal
  2. Browse the app marketplace and find the Unsloth Studio card.
  3. Click the card, provide an instance name (and optionally select GPU count if the admin has enabled that override).

Unsloth Studio Config

  1. Click Deploy and wait ~60 seconds for the Unsloth Studio instance to come up.
  2. Click the Web URL and start fine-tuning models.

No YAML files. No Kubernetes knowledge. No Docker commands. No GPU driver troubleshooting. The platform team has abstracted all of that complexity away

Unsloth Studio Config

Conclusion

The combination of Unsloth Studio and Rafay's App Marketplace provides a powerful pattern for democratizing LLM fine-tuning within an organization. Platform teams get the governance, multi-tenancy, and lifecycle controls they need. AI practitioners get a frictionless, self-service experience that lets them focus on models and data rather than infrastructure.

By packaging Unsloth Studio as an App, you're not just deploying a container — you're building an internal AI platform capability that scales with your organization's needs.