lower default initial learning rate to what Jonathan Ho had in his original repo

dalle2_pytorch/dalle2_pytorch.py

@@ -614,7 +614,6 @@ class Attention(nn.Module):
         heads = 8,
         dropout = 0.,
         causal = False,
-        post_norm = False,
         rotary_emb = None
     ):
         super().__init__()
@@ -624,7 +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))
@@ -635,7 +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):
@@ -692,8 +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.to_out(out)
-        return self.post_norm(out)
 
 class CausalTransformer(nn.Module):
     def __init__(
@@ -719,7 +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)
             ]))
 

dalle2_pytorch/optimizer.py

@@ -7,7 +7,7 @@ def separate_weight_decayable_params(params):
 
 def get_optimizer(
     params,
-    lr = 3e-4,
+    lr = 2e-5,
     wd = 1e-2,
     betas = (0.9, 0.999),
     filter_by_requires_grad = False

dalle2_pytorch/train.py

@@ -221,7 +221,7 @@ class DecoderTrainer(nn.Module):
         self,
         decoder,
         use_ema = True,
-        lr = 3e-4,
+        lr = 2e-5,
         wd = 1e-2,
         max_grad_norm = None,
         amp = False,

setup.py

@@ -10,7 +10,7 @@ setup(
       'dream = dalle2_pytorch.cli:dream'
     ],
   },
-  version = '0.2.18',
+  version = '0.2.20',
   license='MIT',
   description = 'DALL-E 2',
   author = 'Phil Wang',