AutoDL

A Chinese pay-by-the-hour GPU cloud platform, popular with students for running experiments.

AutoDL is a Chinese GPU-rental cloud platform: users pick a GPU model and a base image on its website, spin up a GPU-backed container instance, and are billed by usage time, accessing it through JupyterLab or SSH. The platform ships pre-built images with common environments such as PyTorch and CUDA already installed, plus a data disk, file storage, and network acceleration aimed at academic sites. Students without a lab server, or without a strong enough local GPU, often use it to train and fine-tune models and run simulations. Remember to shut the instance down when you’re done, since it keeps billing otherwise; data should be kept in a persistent directory, since it’s lost once the instance is released.

ExampleRenting an RTX 4090 and booting the official PyTorch image, then SSHing in to clone the LeRobot code and fine-tune an ACT policy.

Also called

AutoDL GPU Cloud

Related

Secure Shell (SSH) · Docker · Conda · Package Mirror Sources in China (pip / conda / apt / Docker mirrors) · Common Training and Inference GPUs (RTX 4090 / A100 / H100 / B200) · Hugging Face Mirror (hf-mirror.com)

Sources

AutoDL 官网 (Chinese)
AutoDL 帮助文档 (Chinese)