unsloth-studio — Unsloth managed UI
A managed web UI for Unsloth fine-tuning — config, launch, and monitor runs from the browser, no Python required. Built on top of the official unsloth/unsloth:stable image.
Image tag
docker.io/manvarharsh/unsloth-studio:cuda12
What’s in this image
- Base:
unsloth/unsloth:stable(upstream Unsloth image) - Unsloth Studio web UI
- All Unsloth dependencies
- JupyterHub and Jupyter Notebook (port 8888)
- OpenSSH server
Default ports
| Port | Service |
|---|---|
| 22 | SSH |
| 8000 | JupyterHub / Unsloth Studio |
| 8888 | Jupyter Notebook |
Use cases
- Point-and-click LoRA / QLoRA fine-tunes
- Browser-driven dataset prep and run management
- Quick iteration without writing training scripts
- Pair with the Unsloth notebook workflow on Podstack
Environment variables
| Variable | Description |
|---|---|
ENABLE_SSH | Enable SSH server |
ENABLE_UNSLOTH_STUDIO | Start the Studio UI (defaults to true) |
PODSTACK_API_URL | Backend URL for token validation |
SSH_PUBLIC_KEY | Public key for SSH |
Persistence
Mount at /data. The Studio writes runs, configs, and outputs under /data/.
See also
- Conventions and shared environment variables
- unsloth — CLI/notebook variant
- Accessing output files in the browser