Apologies if somehow these are covered in different places. On the latest ubuntu and cuda-theano image, after about an hour of tweaking cnmem to no avail, I discovered my 820m is simply too old to support cudnn. Then I ran
THEANO_FLAGS='floatX=float32,device=cpu,lib.cnmem=1' theano-test (assuming cnmem is simply ignored here) and got, at the very end *Ran 430 tests in 185.914s* *FAILED (SKIP=18, errors=88, failures=12)* If I understand docker recipes correctly, they could be configured flexibly to always run the latest-best versions of everything, but perhaps cuda-theano has been left a bit behind? I am guessing the absence of GPU only produced the 18 skips, but I could be wrong. Are the docker images perhaps only to be thought of "as is" with no further interest from the community? I am appending the rest of the error messages, thanks! http://pastebin.com/vkXWZbnE -- --- You received this message because you are subscribed to the Google Groups "theano-users" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
