Deploy and Host TorchCode on Railway

TorchCode is a self-hosted practice platform for PyTorch: implement softmax, attention, GPT-2 blocks and more from scratch in Jupyter notebooks and get instant auto-grading. No GPU is needed.

About Hosting TorchCode

This template runs the official ghcr.io/duoan/torchcode:latest image (JupyterLab with the problem notebooks and the grader) with a persistent volume at /app/data. JupyterLab can run code on the server, so it is protected with a generated access token: open the service Variables tab, copy JUPYTER_TOKEN, then open https:///?token=. The blank notebooks are reset to their templates on every restart, so keep a copy of work you want to keep.

Common Use Cases

Prepare for ML engineer interviews by coding operators from scratch
Run a shared practice environment for a study group or team
Teach PyTorch fundamentals with graded exercises

Dependencies for TorchCode Hosting

One service with a volume; CPU only, no external database

Deployment Dependencies

Why Deploy TorchCode on Railway?

Railway gives you a public HTTPS URL, a persistent volume and automatic restarts with no server to maintain.