A juvenil question

[dont-know-why]

Thanks for ur work !!!

when I run the command :
python ../../tools/test-ms-f.py --model models/model-f_iter_20000.caffemodel --images list/val_id.txt --dir /xxx/xxx/xxx --output DSRG_final_output --gpu 0 --smooth true
I got a issue :
Check failed: error == cudaSuccess (2 vs. 0) out of memory
it turns out that the error happens on the line "net.forward()"
I am using nvidia gtx960 4 GB.
This problem didn't occur during the training process
How could I find the batch_size to decrease the using of memory?