Open pskl opened 9 months ago
Cyberes
commented
9 months ago I played around with that and found it was simpler to just re-download everything rather than deal with the complexity of Jupyter kernels and venvs.
The pip_cache_dir option accomplishes half of your suggestion, though. This makes pip cache the downloads so you only have to re-install, not re-download and re-install.
pskl
commented
9 months ago Why is this closed?
Cyberes
commented
9 months ago Would you rather leave it open?
pskl
commented
7 months ago Hmmm, ah sorry at the time of my last comment I somehow didnt see your response, my bad.
And btw I'm a big fan of your work in general! Great stuff!
Maybe just to bounce ideas around lets keep it open: please detail what you have tried so far and whether you think its work investigating further.
I might give it a go because it would speed up my pipeline dramatically (this is the most time consuming step right now).
Since libraries are installed in the persistent storage space, is there a way to not have to redownload every dependency on each VM creation because its by far the slowest step. Talking about the "Install requirements and download repositories" step here.
Maybe an additional new code step used for "rebooting a previous environment with the persisted libraries and spin up Gradio" in one cell would be useful?