BinaryNet icon indicating copy to clipboard operation
BinaryNet copied to clipboard
verified publisher icon MatthieuCourbariaux

speed of training

Open adl2db opened this issue 8 years ago • 2 comments

I'm running the mnist example using a K80 Tesla Chip but I'm finding each epoch is running around 130mins. Wondering if there's any Theano configuration or something that could help with run time.

Thank you.

adl2db avatar Jul 25 '17 15:07 adl2db

Same problem here, it would take me 50 days to train MNIST for 1000 iterations. Is it possible to get trained parameters anywhere and load them? Thank you :)

ahmygit avatar Feb 14 '18 07:02 ahmygit

Back then, on a GTX Titan Black, each epoch took about 22 seconds. The whole training lasted about 6h.

Unfortunately, I did not keep the trained parameters.

MatthieuCourbariaux avatar Feb 18 '18 06:02 MatthieuCourbariaux