I installed latest build of theano and while importing, it is stuck at "importing theano" forever. I am using a cluster of GPUs. On the cluster there are TitanX and TitanXP GPUs. on parts with TitanX, under the same setting, theano is working fine. However, when I submit the job to nodes containing TitanXP, the code gets stuck at importing theano forever and never proceeds further. This rules out the fact that source of problem could be CUDA version or flags. Anyway following are the details of my setting.
I have cuda: 8.0 cudnn: 5.1. I have following flags: theano_flags['force_device'] = 'True' theano_flags['dnn.enabled'] = 'True' theano_flags['floatX'] = 'float32' theano_flags['assert_no_cpu_op'] = 'warn' theano_flags['allow_gc'] = 'True' theano_flags['gpuarray.preallocate'] = '0.80' theano_flags['dnn.conv.algo_bwd_filter'] = 'deterministic' # Necessary for deterministic training runs. theano_flags['dnn.conv.algo_bwd_data'] = 'deterministic' # Necessary for deterministic training runs. I use environment variable to set theano_flags[device]='cuda'+os.environ['GPU_ID'] so, it is also not a problem of assigning right GPU. Also, I have used other TitanXp with tensorflow and pytorch with same CUDA drivers. So, that is also not source of problem either. Has anyone encountered similar issue before? -- --- 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.
