switching between CPU and GPU when resuming about matconvnet

[vedaldi]

Hi, the problem is that the intermediate models are saved either in GPU or CPU format, depending on the initial setting. The solution would be to use vl_simplenn_move() right after loading the intermediate model. We should add this to a future version.

On 8 Jan 2015, at 18:05, guanggz [email protected] wrote:

When i interrupt the example code, say cnn_mnist with opts.train.useGpu = false, and change it to GPU mode by doing opts.train.useGpu = true, it gives errors like

resuming by loading epoch 59
training: epoch 60: processing batch 1 of 600 ...Error using vl_nnconv
DATA and FILTERS are not both CPU or GPU arrays.
Also, if I complete run can_mnist in GPU mode, it will skip to run again in CPU mode.
Any idea that I can switch btw CPU and GPU freely and start from the beginning as I wish?

Thanks in advance.

—
Reply to this email directly or view it on GitHub #37.

Hi Andrea, thanks for quick reply.
It is a great function for notebooks to carry on with unfinished jobs. But as for now, could you suggest some quick fix/hints for this problem? thx.
On 08 Jan 2015, at 19:22, Andrea Vedaldi [email protected] wrote:

Hi, the problem is that the intermediate models are saved either in GPU or CPU format, depending on the initial setting. The solution would be to use vl_simplenn_move() right after loading the intermediate model. We should add this to a future version.

On 8 Jan 2015, at 18:05, guanggz [email protected] wrote:

When i interrupt the example code, say cnn_mnist with opts.train.useGpu = false, and change it to GPU mode by doing opts.train.useGpu = true, it gives errors like

resuming by loading epoch 59
training: epoch 60: processing batch 1 of 600 ...Error using vl_nnconv
DATA and FILTERS are not both CPU or GPU arrays.
Also, if I complete run can_mnist in GPU mode, it will skip to run again in CPU mode.
Any idea that I can switch btw CPU and GPU freely and start from the beginning as I wish?

Thanks in advance.

—
Reply to this email directly or view it on GitHub #37.

—
Reply to this email directly or view it on GitHub.

[vedaldi]

Hi, yes of course. Simply use vl_simplenn_move() to convert the model to GPU or CPU as needed right after the load instruction in cnn_Train.m

On 8 Jan 2015, at 18:33, guanggz [email protected] wrote:

The solution would be to use vl_simplenn_move() right after loading the intermediate model.

Thanks! good to go again :D
On 08 Jan 2015, at 19:40, Andrea Vedaldi [email protected] wrote:

vl_simplenn_move()