Use the VAE's device.

Fixes #37

tiled_vae.py

@@ -352,7 +352,7 @@ def perfcount(fn):
     def wrapper(*args, **kwargs):
         ts = time()
 
-        if torch.cuda.is_available():
+        if torch.cuda.is_available() and devices.device not in ['cpu', devices.cpu]:
             torch.cuda.reset_peak_memory_stats(devices.device)
         devices.torch_gc()
         gc.collect()
@@ -361,7 +361,7 @@ def perfcount(fn):
 
         devices.torch_gc()
         gc.collect()
-        if torch.cuda.is_available():
+        if torch.cuda.is_available() and devices.device not in ['cpu', devices.cpu]:
             vram = torch.cuda.max_memory_allocated(devices.device) / 2**20
             print(f'[Tiled VAE]: Done in {time() - ts:.3f}s, max VRAM alloc {vram:.3f} MB')
         else:
@@ -417,8 +417,6 @@ class GroupNormParam:
         pixels = torch.tensor(self.pixel_list, dtype=torch.float32, device=devices.device) / max_value
         sum_pixels = torch.sum(pixels)
         pixels = pixels.unsqueeze(1) / sum_pixels
-        # var = torch.sum(var * pixels.to(var.device), dim=0)
-        # mean = torch.sum(mean * pixels.to(var.device), dim=0)
         var = torch.sum(var * pixels, dim=0)
         mean = torch.sum(mean * pixels, dim=0)
         return lambda x:  custom_group_norm(x, 32, mean, var, self.weight, self.bias)
@@ -761,6 +759,7 @@ class TiledVAE:
         fast = kwargs['fast'] if 'fast' in kwargs else args[3]
         color_fix = kwargs['color_fix'] if 'color_fix' in kwargs else False
         is_decoder = self.is_decoder
+        devices.device = _vae.device
 
         # for shorthand
         vae = _vae.first_stage_model