@@ -118,18 +118,16 @@ class Net(nn.Sequential):
118118
119119 def __init__ (self , n_classes : int = 10 ):
120120 super ().__init__ (
121- nn .LazyConv2d (
122- 32 , 3 , 1
123- ), # NOTE: `in_channels` is determined in the first forward pass
121+ # NOTE: `in_channels` is determined in the first forward pass
122+ nn .LazyConv2d (32 , 3 , 1 ),
124123 nn .ReLU (),
125124 nn .Conv2d (32 , 64 , 3 , 1 ),
126125 nn .ReLU (),
127126 nn .MaxPool2d (2 ),
128127 nn .Dropout2d (0.25 ),
129128 nn .Flatten (),
130- nn .LazyLinear (
131- 128
132- ), # NOTE: `in_features` is determined in the first forward pass
129+ # NOTE: `in_features` is determined in the first forward pass
130+ nn .LazyLinear (128 ),
133131 nn .ReLU (),
134132 nn .Dropout (0.5 ),
135133 nn .Linear (128 , n_classes ),
@@ -174,12 +172,10 @@ def test_epoch(model: Net, device: torch.device, test_loader: DataLoader) -> flo
174172 for data , target in test_loader :
175173 data , target = data .to (device ), target .to (device )
176174 output = model (data )
177- test_loss += F .nll_loss (
178- output , target , reduction = "sum"
179- ).item () # sum up batch loss
180- pred = output .argmax (
181- dim = 1 , keepdim = True
182- ) # get the index of the max log-probability
175+ # sum up batch loss
176+ test_loss += F .nll_loss (output , target , reduction = "sum" ).item ()
177+ # get the index of the max log-probability
178+ pred = output .argmax (dim = 1 , keepdim = True )
183179 correct += pred .eq (target .view_as (pred )).sum ().item ()
184180
185181 test_loss /= num_batches
@@ -379,4 +375,7 @@ def main(**kwargs):
379375 },
380376)
381377fig .show ()
378+ fig .write_image ("../../docs/src/_static/warm_start_thumbnail.png" )
382379fig
380+
381+ # sphinx_gallery_thumbnail_path = '_static/warm_start_thumbnail.png'
0 commit comments