Mmdetection: How to train the model on other gpu_id?

Created on 24 Mar 2019  路  2Comments  路  Source: open-mmlab/mmdetection

model = MMDataParallel(model, device_ids=range(cfg.gpus)).cuda()
When i try to train the model on gpu_id=3,it is still on the gpu_id=0

picture Markovcom

Most helpful comment

You can specify CUDA_VISIBLE_DEVICES like CUDA_VISIBLE_DEVICES=2,3 ./tools/dist_train.sh xxxx.

picture hellock on 4 Apr 2019
❤4

All 2 comments

You can specify CUDA_VISIBLE_DEVICES like CUDA_VISIBLE_DEVICES=2,3 ./tools/dist_train.sh xxxx.

hellock picture hellock on 4 Apr 2019
❤4

You can specify CUDA_VISIBLE_DEVICES like CUDA_VISIBLE_DEVICES=2,3 ./tools/dist_train.sh xxxx.

You can specify CUDA_VISIBLE_DEVICES like CUDA_VISIBLE_DEVICES=2,3 ./tools/dist_train.sh xxxx.

Only need to modify dist_train.sh?

BlueBlueFF picture BlueBlueFF on 16 Jun 2019
Was this page helpful?
0 / 5 - 0 ratings

Related issues

namheegordonkim picture namheegordonkim  路  3Comments
hust-kevin picture hust-kevin  路  3Comments
liugaolian picture liugaolian  路  3Comments
FrankXinqi picture FrankXinqi  路  3Comments
Youngkl0726 picture Youngkl0726  路  3Comments