New issue

Have a question about this project? # for a free GitHub account to open an issue and contact its maintainers and the community.

#

By clicking “#”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? # to your account

Jump to bottom

June 2017 release #359

Merged

drnikolaev merged 1 commit into NVIDIA:caffe-0.16 from drnikolaev:caffe-0.16.2-pr

Jun 23, 2017

drnikolaev commented Jun 22, 2017

Enhancements and bug fixes:

CUDA 9 support
Data Cache & Shuffle
Global Gradient Scaling for 16-bit float math support
Better Python support and error handling
Improved determinism across different number of GPUs
Fixed caffe-0.16 converges slower and produces lower accuracy (compared to caffe-0.15) #347 algorithm selection return wrong result: needs 140632G working space. #353 deadlock when train on multiple GPU and restore the state from previous training #255


          June 2017 release

2a4581e

drnikolaev merged commit fd77927 into NVIDIA:caffe-0.16

drnikolaev deleted the caffe-0.16.2-pr branch

June 23, 2017 01:04

# for free to join this conversation on GitHub. Already have an account? # to comment

Labels

None yet