always use sandwich norm for attention layer

dalle2_pytorch/dalle2_pytorch.py

@@ -1,7 +1,7 @@
 import math
 from tqdm import tqdm
 from inspect import isfunction
-from functools import partial
+from functools import partial, wraps
 from contextlib import contextmanager
 from collections import namedtuple
 from pathlib import Path
@@ -45,6 +45,14 @@ def exists(val):
 def identity(t, *args, **kwargs):
     return t
 
+def maybe(fn):
+    @wraps(fn)
+    def inner(x):
+        if not exists(x):
+            return x
+        return fn(x)
+    return inner
+
 def default(val, d):
     if exists(val):
         return val
@@ -606,7 +614,6 @@ class Attention(nn.Module):
         heads = 8,
         dropout = 0.,
         causal = False,
-        post_norm = False,
         rotary_emb = None
     ):
         super().__init__()
@@ -616,7 +623,6 @@ class Attention(nn.Module):
 
         self.causal = causal
         self.norm = LayerNorm(dim)
-        self.post_norm = LayerNorm(dim)     # sandwich norm from Coqview paper + Normformer
         self.dropout = nn.Dropout(dropout)
 
         self.null_kv = nn.Parameter(torch.randn(2, dim_head))
@@ -627,7 +633,7 @@ class Attention(nn.Module):
 
         self.to_out = nn.Sequential(
             nn.Linear(inner_dim, dim, bias = False),
-            LayerNorm(dim) if post_norm else nn.Identity()
+            LayerNorm(dim)
         )
 
     def forward(self, x, mask = None, attn_bias = None):
@@ -684,8 +690,7 @@ class Attention(nn.Module):
         out = einsum('b h i j, b j d -> b h i d', attn, v)
 
         out = rearrange(out, 'b h n d -> b n (h d)')
-        out = self.to_out(out)
-        return self.post_norm(out)
+        return self.to_out(out)
 
 class CausalTransformer(nn.Module):
     def __init__(
@@ -711,7 +716,7 @@ class CausalTransformer(nn.Module):
         self.layers = nn.ModuleList([])
         for _ in range(depth):
             self.layers.append(nn.ModuleList([
-                Attention(dim = dim, causal = True, dim_head = dim_head, heads = heads, dropout = attn_dropout, post_norm = normformer, rotary_emb = rotary_emb),
+                Attention(dim = dim, causal = True, dim_head = dim_head, heads = heads, dropout = attn_dropout, rotary_emb = rotary_emb),
                 FeedForward(dim = dim, mult = ff_mult, dropout = ff_dropout, post_activation_norm = normformer)
             ]))
 
@@ -1173,7 +1178,11 @@ class CrossAttention(nn.Module):
         self.null_kv = nn.Parameter(torch.randn(2, dim_head))
         self.to_q = nn.Linear(dim, inner_dim, bias = False)
         self.to_kv = nn.Linear(context_dim, inner_dim * 2, bias = False)
-        self.to_out = nn.Linear(inner_dim, dim, bias = False)
+
+        self.to_out = nn.Sequential(
+            nn.Linear(inner_dim, dim, bias = False),
+            LayerNorm(dim)
+        )
 
     def forward(self, x, context, mask = None):
         b, n, device = *x.shape[:2], x.device
@@ -1844,6 +1853,8 @@ class Decoder(BaseGaussianDiffusion):
         b = shape[0]
         img = torch.randn(shape, device = device)
 
+        lowres_cond_img = maybe(normalize_neg_one_to_one)(lowres_cond_img)
+
         for i in tqdm(reversed(range(0, self.num_timesteps)), desc = 'sampling loop time step', total = self.num_timesteps):
             img = self.p_sample(
                 unet,
@@ -1868,9 +1879,7 @@ class Decoder(BaseGaussianDiffusion):
         # normalize to [-1, 1]
 
         x_start = normalize_neg_one_to_one(x_start)
-
-        if exists(lowres_cond_img):
-            lowres_cond_img = normalize_neg_one_to_one(lowres_cond_img)
+        lowres_cond_img = maybe(normalize_neg_one_to_one)(lowres_cond_img)
 
         # get x_t
 

setup.py

@@ -10,7 +10,7 @@ setup(
       'dream = dalle2_pytorch.cli:dream'
     ],
   },
-  version = '0.2.16',
+  version = '0.2.19',
   license='MIT',
   description = 'DALL-E 2',
   author = 'Phil Wang',