-
Notifications
You must be signed in to change notification settings - Fork 1.2k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
LiveTask: write output path at start/end #6815
Conversation
@daavoo that might be a better idea, now its a bit clunky. Ideally I think it should be shown the same way as progress bar, in one place, but I am afraid that it is impossible since every framework logs something from time to time. |
I think having it show up by default at least once is probably useful so users can just click on it rather than have to find the path and open it themselves. What do you mean about showing it like a progress bar? Is there any instance where the output will change between steps? |
@dberenbaum I mean that ideally, instead of printing URL in new line, we would be using \r to go back to line beginning and write the URL again in same line, but its impossible to implement, because we always log something during training, resulting in new lines with URL
Probably not |
Thanks, @pared. What's the value of overwriting if the info doesn't change between steps? |
@dberenbaum there is none, the main point was that ideally, we would only get a single path log, but usually, training produces some other logs, hiding static site URL somewhere in the beginning. Producing it every step makes it even more ugly, so I think let's log it once. |
Good point, this is what I was missing. Maybe it would be nice to print once more at the end?
Sounds good for now. |
@daavoo I guess we can put the path into the debug for now. We can prepare another change for logging in the beginning and/or in the end. |
So the user will not be shown the link to the output at all by default? Does the html open by default? |
I will open a new P.R. addind the automatic open today |
@pared Why put the path into debug by default rather than printing once? Is it simply cleaner to implement? I think it seems useful to print the path somewhere. |
@dberenbaum makes sense, now that I think about it it probably does not make sense to put it to debug because it will only clutter debug then, instead of |
825f308
to
91679d2
Compare
Writing at the end of each step (create_summary) can mess up with some ML Framework progress bars, I moved the message to the monitor task, as the path won't change during training. Fixes iterative/dvclive#177
91679d2
to
9326289
Compare
Closes iterative/dvclive#177
Logging at the end of each step can mess up with some ML Framework progress bars, I moved the message to the instantiation of the monitor task, as the path won't change during training.
@pared Do you consider necessary logging at each
create_summary
call? If so, we could maybe keep the call but atdebug
level