Multi-GPU Training Issue

[ShenZheng2000]

Hi, authors! I encounter an issue when training with 8 GPUs. The dataset has been created, and the model has been initialized. However, there is a TypeError shown as below. The training did go well with single GPU training, though.

Could you help me solving this issue? Thanks!

[RL-arch]

I also have the issue with multi GPU training, the error I got was: "runtimeerror: non-empty 3d or 4d (batch mode) tensor expected for input, but got: [ torch.cuda.floattensor{0,256,64,64} ]"

[jpzhai]

Hi, I have the same problem as you, did you solve it?

[ShenZheng2000]

I found a solution by modifying train.py.

Replace this

            model.optimize_parameters()   # calculate loss functions, get gradients, update network weights

with this

            model.set_input(data)
            model.optimize_parameters()   # calculate loss functions, get gradients, update network weights