Roelof van Dijk

@unofficialquant The downloaded models are stored in a tempfile based on a hash of the url or directly read (`fetch` and `fetch_as_file`). https://github.com/tinygrad/tinygrad/blob/e2f6b09ffde3de184421c1dff164be1edee95cca/extra/utils.py#L16 If those are cached, you probably need...

> The issue with `tempfile` is it creates the directory at `/var/folders`. I've tested earlier on and GitHub action is not able to cache anything from that directory. Would have...

Maybe consider a pre-built docker image for running the CUDA test to avoid the apt installs?

Minor semantic nit (if I am not mistaken): If 200ms->100ms is 100% speedup (2x as fast), 99.58ms->79.66ms is already a **25%** speedup. A 20% speedup would be 82.98ms (99.58ms/120). I'd...