add weight standardization behind feature flag, which may potentially work well with group norm

Author: lucidrains

README.md

@@ -1264,4 +1264,14 @@ For detailed information on training the diffusion prior, please refer to the [d
 }
 ```
 
+```bibtex
+@article{Qiao2019WeightS,
+    title   = {Weight Standardization},
+    author  = {Siyuan Qiao and Huiyu Wang and Chenxi Liu and Wei Shen and Alan Loddon Yuille},
+    journal = {ArXiv},
+    year    = {2019},
+    volume  = {abs/1903.10520}
+}
+```
+
 *Creating noise from data is easy; creating data from noise is generative modeling.* - <a href="https://arxiv.org/abs/2011.13456">Yang Song's paper</a>

dalle2_pytorch/dalle2_pytorch.py

@@ -1451,6 +1451,26 @@ def Downsample(dim, *, dim_out = None):
     dim_out = default(dim_out, dim)
     return nn.Conv2d(dim, dim_out, 4, 2, 1)
 
+class WeightStandardizedConv2d(nn.Conv2d):
+    """
+    https://arxiv.org/abs/1903.10520
+    weight standardization purportedly works synergistically with group normalization
+    """
+    def forward(self, x):
+        eps = 1e-5 if x.dtype == torch.float32 else 1e-3
+
+        weight = self.weight
+        flattened_weights = rearrange(weight, 'o ... -> o (...)')
+
+        mean = reduce(weight, 'o ... -> o 1 1 1', 'mean')
+
+        var = torch.var(flattened_weights, dim = -1, unbiased = False)
+        var = rearrange(var, 'o -> o 1 1 1')
+
+        weight = (weight - mean) * (var + eps).rsqrt()
+
+        return F.conv2d(x, weight, self.bias, self.stride, self.padding, self.dilation, self.groups)
+
 class SinusoidalPosEmb(nn.Module):
     def __init__(self, dim):
         super().__init__()
@@ -1469,10 +1489,13 @@ def __init__(
         self,
         dim,
         dim_out,
-        groups = 8
+        groups = 8,
+        weight_standardization = False
     ):
         super().__init__()
-        self.project = nn.Conv2d(dim, dim_out, 3, padding = 1)
+        conv_klass = nn.Conv2d if not weight_standardization else WeightStandardizedConv2d
+
+        self.project = conv_klass(dim, dim_out, 3, padding = 1)
         self.norm = nn.GroupNorm(groups, dim_out)
         self.act = nn.SiLU()
 
@@ -1496,6 +1519,7 @@ def __init__(
         cond_dim = None,
         time_cond_dim = None,
         groups = 8,
+        weight_standardization = False,
         cosine_sim_cross_attn = False
     ):
         super().__init__()
@@ -1521,8 +1545,8 @@ def __init__(
                 )
             )
 
-        self.block1 = Block(dim, dim_out, groups = groups)
-        self.block2 = Block(dim_out, dim_out, groups = groups)
+        self.block1 = Block(dim, dim_out, groups = groups, weight_standardization = weight_standardization)
+        self.block2 = Block(dim_out, dim_out, groups = groups, weight_standardization = weight_standardization)
         self.res_conv = nn.Conv2d(dim, dim_out, 1) if dim != dim_out else nn.Identity()
 
     def forward(self, x, time_emb = None, cond = None):
@@ -1747,6 +1771,7 @@ def __init__(
         init_dim = None,
         init_conv_kernel_size = 7,
         resnet_groups = 8,
+        resnet_weight_standardization = False,
         num_resnet_blocks = 2,
         init_cross_embed = True,
         init_cross_embed_kernel_sizes = (3, 7, 15),
@@ -1894,7 +1919,7 @@ def __init__(
 
         # prepare resnet klass
 
-        resnet_block = partial(ResnetBlock, cosine_sim_cross_attn = cosine_sim_cross_attn)
+        resnet_block = partial(ResnetBlock, cosine_sim_cross_attn = cosine_sim_cross_attn, weight_standardization = resnet_weight_standardization)
 
         # give memory efficient unet an initial resnet block
 

dalle2_pytorch/version.py

@@ -1 +1 @@
-__version__ = '1.6.5'
+__version__ = '1.7.0'