pytorch
DO NOT CLOSE IT / GitHub Workflow Runner Determinator
In the FIRST comment you can specify the configuration for each experiment.
The format is:
- Above the line break, you have yaml formatted text listing all experiments and their configurations
- Below the line break, you list all the people who should be opted into any experiment, followed by a comma separated list of all experiments they should be opted into.
- If you comment out a person (by prefixing their line with
#) then that line will have no effect
Anyone opted in will always have the given experiment enabled. For everyone else, they will have a rollout_perc chance of having the experiment enabled on their workflow. Note that the experiment is always enabled/disabled at the workflow level, not the job level.
This configuration is parsed by _runner-determinator.yml workflow on pytorch/pytorch.
experiments:
lf:
rollout_perc: 95
all_branches: true
Opted-in:
@ZainRizvi,lf @wdvr,lf @atalman,lf @seemethere,lf @botmethere,lf @clee2000,lf @huydhn,lf @izaitsevfb,lf @jeanschmidt,lf @PaliC,lf @kit1980,lf @malfet,lf @zxiiro,lf
No, pls don't close the issue. It will be used for dynamic rollout while we are migrating from our runners to the linux foundation runners.
Removed myself from the list, as LF runners queue is huge.. Also, why everyone need to be mentioned twice?
Removed myself from the list, as LF runners queue is huge.. Also, why everyone need to be mentioned twice?
PR pytorch/pytorch#129246 changed it so that we are using @ prefixed usernames. So I think the non-@ prefix ones no longer need to be there.
Looks like pytorch/pytorch#129462 introduced a regression where it no longer properly checked the @ prefixed usernames. I have a fix on that but can't open a PR as it depends on pytorch/pytorch#129500 getting merged first. Just waiting for PR checks to complete and I'll try to get it in and open a PR for the @ fix.