Any idea on why downloading speed is abysmal 200 B/s

```
python download.py mnist celebA
```


It is clear that the download speed "hits" some kind of limitation on 200 B/s.

Any idea ?

Start by reproducing `python download.py mnist celebA` and inspect `download.py` for the download path and any speed-limiting behavior. Compare the reported rate for both datasets; done means identifying the source of the roughly 200 B/s limit and documenting or correcting it.

Written by the indexing model from the issue text.