Is it possible to generate using CPU?

[xiankgx]

with open('ffhq.pkl', 'rb') as f:
    G = pickle.load(f)['G_ema'].cuda()  # torch.nn.Module
z = torch.randn([1, G.z_dim]).cuda()    # latent codes
c = None                                # class labels (not used in this example)
img = G(z, c)                           # NCHW, float32, dynamic range [-1, +1]

I trained a custom model and was able to generate images using the above code. However, when I switch the device to CPU, I am faced with the following error:
RuntimeError: "slow_conv_transpose2d_out_cpu" not implemented for 'Half'

How can I fix it?

[JCBrouwer]

You can convert the generator to float and then run with force_fp32=True:

G = G.float()  # convert weights that are half precision to fp32
img = G(z, c, force_fp32=True)  # necessary to make sure intermediate casts to half aren't run

If you want to avoid having to type that every time you can also override the forward method like this:

import functools
G.forward = functools.partial(G.forward, force_fp32=True)

img = G(z, c)

[gtnbssn]

Look here for how to modify generate.py accordingly:

#54 (comment)