Download the 40GB dataset to where the disk is

The corpus you need is 40GB, hosted on a university mirror. You're on hotel Wi-Fi.

You don't download it. You tell the container to: hit the ?download endpoint on the container's files URL with the source link, and the container fetches it directly: source to server, never touching your laptop. ?downloads shows progress while you order dinner. The file lands on the disk where you'll actually process it, on a connection that isn't a hotel access point.

The old way moved 40GB twice: mirror to laptop, laptop to server. The new way moves it once, and your machine moved kilobytes.