Open brando90 opened 3 years ago
I am not familiar with the PyTorch profiler but Nvidia Nsight Systems would probably be useful. This article is focused on model serving but covers how to use it: https://paulbridger.com/posts/video-analytics-pipeline-tuning/
Profiling the code of Torchmeta is a great idea! I don't have much experience with this (other than using pyinstrument occasionally), so any input on that is welcome!
I wanted to help profile torchmeta.
I was wondering what is the recommended too (to be able to figure out if it's the models or the dataloader that is causing the slow down + have it work even if GPUs are present).
I know only of:
https://pytorch.org/tutorials/recipes/recipes/profiler.html
python -m cProfile -s cumtime main.py > profile.txt
Are these the tools to use for profiling full pytorch code?