Error for 8 GPUs

[wanderine]

I can train a 3D GAN using 4 GPUs, but when I try to use 8 GPUs I get this error

File "/scratch/local/MITPGAN3D/trainer.py", line 158, in step_fn  *
    average_samples = utils.random_weight_sample(reals, fakes, self.dimensionality)
File "/scratch/local/MITPGAN3D/utils.py", line 17, in random_weight_sample  *
    return (weight * reals) + ((1 - weight) * fakes)

ValueError: Dimensions must be equal, but are 32 and 64 for '{{node mul_3}} = Mul[T=DT_FLOAT](sub, model_9/activation_14/Tanh)' with input shapes: [32,1,1,1,1], [64,8,8,8,1].

Any idea?

[satra]

@wanderine - i would suggest using the latest nobrainer release. we should be deprecating this repo soon. there have been quite a few changes in the code base and the nobrainer code is the most appropriate.