fast ci with cpu-based wheels

[skshetry]

• [ ] ❗ I have followed the Contributing to DVCLive guide.

• [ ] 📖 If this PR requires documentation updates, I have created a separate PR (or issue, at least) in dvc.org and linked it here.

Thank you for the contribution - we'll try to review it as soon as possible. 🙏

[codecov-commenter]

Codecov Report

Base: 94.85% // Head: 94.85% // No change to project coverage :thumbsup:

Coverage data is based on head (13df01f) compared to base (2d61715). Patch has no changes to coverable lines.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #329   +/-   ##
=======================================
  Coverage   94.85%   94.85%           
=======================================
  Files          36       36           
  Lines        1770     1770           
  Branches      165      165           
=======================================
  Hits         1679     1679           
  Misses         66       66           
  Partials       25       25           

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

:umbrella: View full report at Codecov.
:loudspeaker: Do you have feedback about the report comment? Let us know in this issue.

[daavoo]

Hard to say because is not deterministic to compare across different workflow runs, but it doesn't look like there is much difference in speed?

[skshetry]

It does seem to save 2-3mins in CI.

Maybe more so locally (torch is alone 1.3Gb compared to 170Mb CPU based wheel, similar difference in size with tensorflow-cpu).

In CI, most of the time is being spent on installation of the wheel rather than download, so if we could cache venv, it’d have been much faster.