train.py takes too long when i run locally on jupyter lab (45-46sec per epoch) but faster on colab(3-4sec per epoch)

[SuryanshuBasa]

While coding the assignment on 05_going_modular locally, the train.py script whej running from CLI takes up 45sec per epoch where as its significantly faster on colab. I have an RTX 3060 6GB. The functions for individual .py scripts are the same form the course. Looking for any help
Thank you :)

[LuluW8071]

Could be due to many reasons:

Feature	Google Colab	Your Device
CPU	Intel Xeon [Large Cache Memory]	Desktop chip (e.g., Intel Core or AMD Ryzen)
GPU	Tesla T4 (12GB VRAM)	NVIDIA RTX 3060 (6GB VRAM)
Software Environment	Linux	Windows
I/O Performance	Potentially faster due to access to Google Drive and optimized cloud storage solutions	Depends on local storage and network setup

Maybe in your case one solution would be increase num_workers in your dataloader script to half of your cpu threads and run the train script again. Also remember to check task manager to see other applications resource consumption. Maybe that is slowing down your training progress.

[SuryanshuBasa]

Thanks for the info, the table makes it clear :)