Open emeryberger opened 2 years ago
Just tried running this and things didn't work for steps 3 onwards. I may be doing something dumb. Regardless really keen to try this out to share profiles with people I'm helping to debug PyTorch inference slowdowns
Hi @emeryberger just flagging this issue again, still eager to try out easy profiling from a notebook for demos. Maybe this is related to how colab supports python 3.7 by default not 3.8?
Describe the bug Running Scalene fails on Google Colab unless
--cli
is passed as an argument.To Reproduce Steps to reproduce the behavior:
!pip install scalene
)%load_ext scalene
)%scrun torchtest()
%scrun --cli torchtest()
There also seems to be an issue where the saved files for each cell are getting deleted so they can't appear in the profile; this is arguably a separate issue but noting it here.
Expected behavior Scalene should auto detect whether it would fail (because it cannot connect to localhost), and fail over gracefully to use the CLI.