🐛 Describe the bug

Not sure if this is intentional, but it makes it difficult for the responses to play well with caching, e.g. this prompted the astral-sh/uv#4967 where it's unclear how to resolve.

FWIW,

% curl -I https://download.pytorch.org/whl/cpu | grep cache-control
cache-control: no-cache,no-store,must-revalidate

and

% curl -I https://pypi.org/simple/ | grep cache-control
cache-control: max-age=600, public

In this particular instance this is manifesting as a problem for me because requests to https://download.pytorch.org/whl/cpu sometimes sporadically fail (e.g. with 503), so being able to reliably cache it would mitigate this considerably.

Versions

Fairly certain this issue is independent of my runtime and reproducible anywhere.

cc @ezyang @gchanan @zou3519 @kadeng @msaroufim @seemethere @malfet @osalpekar @atalman