take care of backwards within trainer classes for diffusion prior and decoder, readying to take care of gradient accumulation as well (plus, unsure if loss should be backwards within autocast block)

README.md

@@ -775,7 +775,6 @@ decoder_trainer = DecoderTrainer(
 
 for unet_number in (1, 2):
     loss = decoder_trainer(images, text = text, unet_number = unet_number)  # use the decoder_trainer forward
-    loss.backward()
 
     decoder_trainer.update(unet_number) # update the specific unet as well as its exponential moving average
 
@@ -839,7 +838,6 @@ diffusion_prior_trainer = DiffusionPriorTrainer(
 )
 
 loss = diffusion_prior_trainer(text, images)
-loss.backward()
 diffusion_prior_trainer.update()  # this will update the optimizer as well as the exponential moving averaged diffusion prior
 
 # after much of the above three lines in a loop
@@ -1017,6 +1015,7 @@ Once built, images will be saved to the same directory the command is invoked
 - [ ] offer save / load methods on the trainer classes to automatically take care of state dicts for scalers / optimizers / saving versions and checking for breaking changes
 - [ ] bring in skip-layer excitatons (from lightweight gan paper) to see if it helps for either decoder of unet or vqgan-vae training
 - [ ] decoder needs one day worth of refactor for tech debt
+- [ ] allow for unet to be able to condition non-cross attention style as well
 
 ## Citations
 

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.post_norm(out)
+        return self.to_out(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)
             ]))
 
@@ -1166,6 +1163,7 @@ class CrossAttention(nn.Module):
         dim_head = 64,
         heads = 8,
         dropout = 0.,
+        norm_context = False
     ):
         super().__init__()
         self.scale = dim_head ** -0.5
@@ -1175,7 +1173,7 @@ class CrossAttention(nn.Module):
         context_dim = default(context_dim, dim)
 
         self.norm = LayerNorm(dim)
-        self.norm_context = LayerNorm(context_dim)
+        self.norm_context = LayerNorm(context_dim) if norm_context else nn.Identity()
         self.dropout = nn.Dropout(dropout)
 
         self.null_kv = nn.Parameter(torch.randn(2, dim_head))
@@ -1381,6 +1379,9 @@ class Unet(nn.Module):
             Rearrange('b (n d) -> b n d', n = num_image_tokens)
         ) if image_embed_dim != cond_dim else nn.Identity()
 
+        self.norm_cond = nn.LayerNorm(cond_dim)
+        self.norm_mid_cond = nn.LayerNorm(cond_dim)
+
         # text encoding conditioning (optional)
 
         self.text_to_cond = None
@@ -1596,6 +1597,11 @@ class Unet(nn.Module):
 
         mid_c = c if not exists(text_tokens) else torch.cat((c, text_tokens), dim = -2)
 
+        # normalize conditioning tokens
+
+        c = self.norm_cond(c)
+        mid_c = self.norm_mid_cond(mid_c)
+
         # go through the layers of the unet, down and up
 
         hiddens = []

dalle2_pytorch/optimizer.py

@@ -7,16 +7,17 @@ def separate_weight_decayable_params(params):
 
 def get_optimizer(
     params,
-    lr = 3e-4,
+    lr = 2e-5,
     wd = 1e-2,
     betas = (0.9, 0.999),
+    eps = 1e-8,
     filter_by_requires_grad = False
 ):
     if filter_by_requires_grad:
         params = list(filter(lambda t: t.requires_grad, params))
 
     if wd == 0:
-        return Adam(params, lr = lr, betas = betas)
+        return Adam(params, lr = lr, betas = betas, eps = eps)
 
     params = set(params)
     wd_params, no_wd_params = separate_weight_decayable_params(params)
@@ -26,4 +27,4 @@ def get_optimizer(
         {'params': list(no_wd_params), 'weight_decay': 0},
     ]
 
-    return AdamW(param_groups, lr = lr, weight_decay = wd, betas = betas)
+    return AdamW(param_groups, lr = lr, weight_decay = wd, betas = betas, eps = eps)

dalle2_pytorch/train.py

@@ -90,7 +90,7 @@ class EMA(nn.Module):
     def __init__(
         self,
         model,
-        beta = 0.99,
+        beta = 0.9999,
         update_after_step = 1000,
         update_every = 10,
     ):
@@ -147,6 +147,7 @@ class DiffusionPriorTrainer(nn.Module):
         use_ema = True,
         lr = 3e-4,
         wd = 1e-2,
+        eps = 1e-6,
         max_grad_norm = None,
         amp = False,
         **kwargs
@@ -173,6 +174,7 @@ class DiffusionPriorTrainer(nn.Module):
             diffusion_prior.parameters(),
             lr = lr,
             wd = wd,
+            eps = eps,
             **kwargs
         )
 
@@ -212,7 +214,9 @@ class DiffusionPriorTrainer(nn.Module):
     ):
         with autocast(enabled = self.amp):
             loss = self.diffusion_prior(*args, **kwargs)
-        return self.scaler.scale(loss / divisor)
+            scaled_loss = self.scaler.scale(loss / divisor)
+            scaled_loss.backward()
+        return loss.item()
 
 # decoder trainer
 
@@ -221,8 +225,9 @@ class DecoderTrainer(nn.Module):
         self,
         decoder,
         use_ema = True,
-        lr = 3e-4,
+        lr = 2e-5,
         wd = 1e-2,
+        eps = 1e-8,
         max_grad_norm = None,
         amp = False,
         **kwargs
@@ -247,13 +252,14 @@ class DecoderTrainer(nn.Module):
         # be able to finely customize learning rate, weight decay
         # per unet
 
-        lr, wd = map(partial(cast_tuple, length = self.num_unets), (lr, wd))
+        lr, wd, eps = map(partial(cast_tuple, length = self.num_unets), (lr, wd, eps))
 
-        for ind, (unet, unet_lr, unet_wd) in enumerate(zip(self.decoder.unets, lr, wd)):
+        for ind, (unet, unet_lr, unet_wd, unet_eps) in enumerate(zip(self.decoder.unets, lr, wd, eps)):
             optimizer = get_optimizer(
                 unet.parameters(),
                 lr = unet_lr,
                 wd = unet_wd,
+                eps = unet_eps,
                 **kwargs
             )
 
@@ -326,4 +332,6 @@ class DecoderTrainer(nn.Module):
     ):
         with autocast(enabled = self.amp):
             loss = self.decoder(x, unet_number = unet_number, **kwargs)
-        return self.scale(loss / divisor, unet_number = unet_number)
+            scaled_loss = self.scale(loss / divisor, unet_number = unet_number)
+            scaled_loss.backward()
+        return loss.item()

setup.py

@@ -10,7 +10,7 @@ setup(
       'dream = dalle2_pytorch.cli:dream'
     ],
   },
-  version = '0.2.18',
+  version = '0.2.24',
   license='MIT',
   description = 'DALL-E 2',
   author = 'Phil Wang',