I am trying to save the weights of my neuralnet implemented in keras 1.2 using
model.save_weights('weights.h5')
but this is giving me 'disk quota exceeded' error message. I am guessing from the error message that it should take (implied by 'total write size') less than 70MB space, but it is actually taking (implied by 'bytes actually written') an unreasonable amount of space (18446744073709551615 bytes). I am expecting the size closer to 70MB since my network is just a one-layer 4096-unit SimpleRNN. I am running it on my college's supercomputer which uses slurm, and I am not sure if it also might be a slurm problem.
Here is the error message:
IOError: Can't prepare for writing data (File write failed: file descriptor = 3,
errno = 122, error message = 'disk quota exceeded', buf = 0x2b72e0202010,
total write size = 67108864, bytes this sub-write = 67108864, bytes actually
written = 18446744073709551615, offset = 2720)
Exception RuntimeError: RuntimeError("Can't decrement id ref count(Unable to extend
file properly, errno = 122, error message = 'disk quota exceeded')",) in
'h5py._objects.ObjectID.__dealloc__' ignored /var/spool/slurm/job06721/
slurm_script: line 12: 10763 Segmentation fault python keras_test.py