Introduction

Tengsuan is a GPU infrastructure platform that lets you:

Run scripts on any GPU with a single command and stream the logs back
Spin up isolated GPU sandboxes with root SSH in seconds
Keep datasets and checkpoints on persistent volumes
Inject credentials with encrypted secrets
Organise everything into apps, environments and team workspaces

Everything is metered per second and paid from prepaid credits. There is no configuration to write: GPU type, image, mounts and secrets are flags.

A complete example

# train.py
import torch, os
print(torch.cuda.get_device_name())
print("HF token present:", "HF_TOKEN" in os.environ)
torch.save({"ok": True}, "/mnt/checkpoints/step0.pt")

$ ts run train.py --gpu h100 --image pytorch --secret huggingface --volume checkpoints

The file is uploaded, an H100 host is allocated with the PyTorch image, your Hugging Face secret is injected as environment variables, the checkpoints volume is mounted at /mnt/checkpoints, logs stream to your terminal, and the run appears under the train app in the console.