Conversation
|
Mh, not sure if I agree. The whole idea of rtpt is, that others can see the ETA for processes running on the machines to get a feeling for when a certain GPU might be available. Which use case do you have in mind? |
|
At the moment, in my opinion, a huge problem is that if you are using Jupyter notebooks on the GPUs there is no "easy" way to name the process. With this change, you could just instantiate rtpt at the beginning of the notebook and others could contact you if your process is blocking a GPU for an extensive amount of time. |
|
I guess we could introduce an "interactive" mode that requires a special description and removes the need for calling |
|
Yeah, an interactive mode sounds like a better idea. Maybe instead of an ETA we could even add a timestamp when the Jupyter script was started to check if it is still actively in use or not. |
|
I will have a look at this at some time. |
|
Yes, this sounds much better :) we could just use the same functionality of the ETA stuff to measure the time since init to display "since:xxx" and adapt the eta case with "eta:xxx". |
Sometimes it is not possible to approximate the runtime of the script. In these cases, it would be nice to display the ETA since this will clutter the process name with unnecessary information.
To prevent users from using this library without displaying the eta by default, a new flag
no_iterationswas introduced. If set to true, the ETA string will not be displayed. To prevent users from accidentally disabling the ETA string, a RuntimeError is thrown ifstepis called whileno_iterationsis set to true.