Exposing iter_size parameter in DIGITS interface to enable larger training batches #660

mfernezir · 2016-03-27T00:37:00Z

As mentioned in NVIDIA/caffe#16, 0.14 Caffe now has the iter_size option which enables delayed batching and hence larger training batches. It would be great to have this option in DIGITS. Or alternatively, option to manually edit the solver file before training like suggested in #565.

lukeyeager added enhancement caffe labels Mar 28, 2016

drozdvadym mentioned this issue Apr 6, 2016

F660 iter size parameter #675

Closed

lukeyeager mentioned this issue May 16, 2016

Expose iter_size solver option #744

Merged

lukeyeager closed this as completed in #744 May 18, 2016

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Exposing iter_size parameter in DIGITS interface to enable larger training batches #660

Exposing iter_size parameter in DIGITS interface to enable larger training batches #660

mfernezir commented Mar 27, 2016

Exposing iter_size parameter in DIGITS interface to enable larger training batches #660

Exposing iter_size parameter in DIGITS interface to enable larger training batches #660

Comments

mfernezir commented Mar 27, 2016