eric612 / MobileNet-YOLO

A caffe implementation of MobileNet-YOLO detection network
Other
865 stars 442 forks source link

how to use multiple gpus for training[Training Issue] #233

Open guagua11 opened 4 years ago

guagua11 commented 4 years ago
     My trian machine has two tesla k80 gpus,so i edit the CMakeList.txt and turn on th nccl&cudnn,also i edit the train_yovlov3_lite.sh with the --gpu 0,1,2,3(k80 has two gpu cores),and then build .now when i go training ,it breaked on the "solver.cpp:203] Creating test net (#0) specified by test_net file: models/yolov3/mobilenet_yolov3_lite_test.prototxt",just fall into this operate without any response.

with --gpu 0,it works, so how can i ust mulite gpus for this training? thank you !