fix amp issue for https://github.com/lucidrains/DALLE2-pytorch/issues/82

README.md

@@ -1003,6 +1003,7 @@ Once built, images will be saved to the same directory the command is invoked
 - [x] pull logic for training diffusion prior into a class DiffusionPriorTrainer, for eventual script based + CLI based training
 - [x] make sure the cascading ddpm in the repository can be trained unconditionally, offer a one-line CLI tool for training on a folder of images
 - [x] bring in cross-scale embedding from iclr paper https://github.com/lucidrains/vit-pytorch/blob/main/vit_pytorch/crossformer.py#L14
+- [x] cross embed layers for downsampling, as an option
 - [ ] become an expert with unets, cleanup unet code, make it fully configurable, port all learnings over to https://github.com/lucidrains/x-unet (test out unet² in ddpm repo) - consider https://github.com/lucidrains/uformer-pytorch attention-based unet
 - [ ] transcribe code to Jax, which lowers the activation energy for distributed training, given access to TPUs
 - [ ] train on a toy task, offer in colab

dalle2_pytorch/dalle2_pytorch.py

@@ -41,9 +41,6 @@ def exists(val):
 def identity(t, *args, **kwargs):
     return t
 
-def is_odd(n):
-    return (n % 2) == 1
-
 def default(val, d):
     if exists(val):
         return val
@@ -306,7 +303,7 @@ def cosine_beta_schedule(timesteps, s = 0.008):
     as proposed in https://openreview.net/forum?id=-NEXDKk8gZ
     """
     steps = timesteps + 1
-    x = torch.linspace(0, timesteps, steps)
+    x = torch.linspace(0, timesteps, steps, dtype = torch.float64)
     alphas_cumprod = torch.cos(((x / timesteps) + s) / (1 + s) * torch.pi * 0.5) ** 2
     alphas_cumprod = alphas_cumprod / alphas_cumprod[0]
     betas = 1 - (alphas_cumprod[1:] / alphas_cumprod[:-1])
@@ -317,21 +314,21 @@ def linear_beta_schedule(timesteps):
     scale = 1000 / timesteps
     beta_start = scale * 0.0001
     beta_end = scale * 0.02
-    return torch.linspace(beta_start, beta_end, timesteps)
+    return torch.linspace(beta_start, beta_end, timesteps, dtype = torch.float64)
 
 
 def quadratic_beta_schedule(timesteps):
     scale = 1000 / timesteps
     beta_start = scale * 0.0001
     beta_end = scale * 0.02
-    return torch.linspace(beta_start**2, beta_end**2, timesteps) ** 2
+    return torch.linspace(beta_start**2, beta_end**2, timesteps, dtype = torch.float64) ** 2
 
 
 def sigmoid_beta_schedule(timesteps):
     scale = 1000 / timesteps
     beta_start = scale * 0.0001
     beta_end = scale * 0.02
-    betas = torch.linspace(-6, 6, timesteps)
+    betas = torch.linspace(-6, 6, timesteps, dtype = torch.float64)
     return torch.sigmoid(betas) * (beta_end - beta_start) + beta_start
 
 
@@ -371,17 +368,21 @@ class BaseGaussianDiffusion(nn.Module):
         self.loss_type = loss_type
         self.loss_fn = loss_fn
 
-        self.register_buffer('betas', betas)
-        self.register_buffer('alphas_cumprod', alphas_cumprod)
-        self.register_buffer('alphas_cumprod_prev', alphas_cumprod_prev)
+        # register buffer helper function to cast double back to float
+
+        register_buffer = lambda name, val: self.register_buffer(name, val.to(torch.float32))
+
+        register_buffer('betas', betas)
+        register_buffer('alphas_cumprod', alphas_cumprod)
+        register_buffer('alphas_cumprod_prev', alphas_cumprod_prev)
 
         # calculations for diffusion q(x_t | x_{t-1}) and others
 
-        self.register_buffer('sqrt_alphas_cumprod', torch.sqrt(alphas_cumprod))
-        self.register_buffer('sqrt_one_minus_alphas_cumprod', torch.sqrt(1. - alphas_cumprod))
-        self.register_buffer('log_one_minus_alphas_cumprod', torch.log(1. - alphas_cumprod))
-        self.register_buffer('sqrt_recip_alphas_cumprod', torch.sqrt(1. / alphas_cumprod))
-        self.register_buffer('sqrt_recipm1_alphas_cumprod', torch.sqrt(1. / alphas_cumprod - 1))
+        register_buffer('sqrt_alphas_cumprod', torch.sqrt(alphas_cumprod))
+        register_buffer('sqrt_one_minus_alphas_cumprod', torch.sqrt(1. - alphas_cumprod))
+        register_buffer('log_one_minus_alphas_cumprod', torch.log(1. - alphas_cumprod))
+        register_buffer('sqrt_recip_alphas_cumprod', torch.sqrt(1. / alphas_cumprod))
+        register_buffer('sqrt_recipm1_alphas_cumprod', torch.sqrt(1. / alphas_cumprod - 1))
 
         # calculations for posterior q(x_{t-1} | x_t, x_0)
 
@@ -389,13 +390,13 @@ class BaseGaussianDiffusion(nn.Module):
 
         # above: equal to 1. / (1. / (1. - alpha_cumprod_tm1) + alpha_t / beta_t)
 
-        self.register_buffer('posterior_variance', posterior_variance)
+        register_buffer('posterior_variance', posterior_variance)
 
         # below: log calculation clipped because the posterior variance is 0 at the beginning of the diffusion chain
 
-        self.register_buffer('posterior_log_variance_clipped', torch.log(posterior_variance.clamp(min =1e-20)))
-        self.register_buffer('posterior_mean_coef1', betas * torch.sqrt(alphas_cumprod_prev) / (1. - alphas_cumprod))
-        self.register_buffer('posterior_mean_coef2', (1. - alphas_cumprod_prev) * torch.sqrt(alphas) / (1. - alphas_cumprod))
+        register_buffer('posterior_log_variance_clipped', torch.log(posterior_variance.clamp(min =1e-20)))
+        register_buffer('posterior_mean_coef1', betas * torch.sqrt(alphas_cumprod_prev) / (1. - alphas_cumprod))
+        register_buffer('posterior_mean_coef2', (1. - alphas_cumprod_prev) * torch.sqrt(alphas) / (1. - alphas_cumprod))
 
     def q_mean_variance(self, x_start, t):
         mean = extract(self.sqrt_alphas_cumprod, t, x_start.shape) * x_start
@@ -1235,12 +1236,13 @@ class CrossEmbedLayer(nn.Module):
     def __init__(
         self,
         dim_in,
-        dim_out,
         kernel_sizes,
+        dim_out = None,
         stride = 2
     ):
         super().__init__()
-        assert all([*map(is_odd, kernel_sizes)])
+        assert all([*map(lambda t: (t % 2) == (stride % 2), kernel_sizes)])
+        dim_out = default(dim_out, dim_in)
 
         kernel_sizes = sorted(kernel_sizes)
         num_scales = len(kernel_sizes)
@@ -1282,6 +1284,8 @@ class Unet(nn.Module):
         init_conv_kernel_size = 7,
         resnet_groups = 8,
         init_cross_embed_kernel_sizes = (3, 7, 15),
+        cross_embed_downsample = False,
+        cross_embed_downsample_kernel_sizes = (2, 4),
         **kwargs
     ):
         super().__init__()
@@ -1302,7 +1306,7 @@ class Unet(nn.Module):
         init_channels = channels if not lowres_cond else channels * 2 # in cascading diffusion, one concats the low resolution image, blurred, for conditioning the higher resolution synthesis
         init_dim = default(init_dim, dim // 3 * 2)
 
-        self.init_conv = CrossEmbedLayer(init_channels, init_dim, init_cross_embed_kernel_sizes, stride = 1)
+        self.init_conv = CrossEmbedLayer(init_channels, dim_out = init_dim, kernel_sizes = init_cross_embed_kernel_sizes, stride = 1)
 
         dims = [init_dim, *map(lambda m: dim * m, dim_mults)]
         in_out = list(zip(dims[:-1], dims[1:]))
@@ -1362,6 +1366,12 @@ class Unet(nn.Module):
 
         assert len(resnet_groups) == len(in_out)
 
+        # downsample klass
+
+        downsample_klass = Downsample
+        if cross_embed_downsample:
+            downsample_klass = partial(CrossEmbedLayer, kernel_sizes = cross_embed_downsample_kernel_sizes)
+
         # layers
 
         self.downs = nn.ModuleList([])
@@ -1377,7 +1387,7 @@ class Unet(nn.Module):
                 ResnetBlock(dim_in, dim_out, time_cond_dim = time_cond_dim, groups = groups),
                 Residual(LinearAttention(dim_out, **attn_kwargs)) if sparse_attn else nn.Identity(),
                 ResnetBlock(dim_out, dim_out, cond_dim = layer_cond_dim, time_cond_dim = time_cond_dim, groups = groups),
-                Downsample(dim_out) if not is_last else nn.Identity()
+                downsample_klass(dim_out) if not is_last else nn.Identity()
             ]))
 
         mid_dim = dims[-1]
@@ -1482,11 +1492,12 @@ class Unet(nn.Module):
 
         if self.cond_on_image_embeds:
             image_tokens = self.image_to_cond(image_embed)
+            null_image_embed = self.null_image_embed.to(image_tokens.dtype) # for some reason pytorch AMP not working
 
             image_tokens = torch.where(
                 image_keep_mask,
                 image_tokens,
-                self.null_image_embed
+                null_image_embed
             )
 
         # take care of text encodings (optional)
@@ -1510,10 +1521,12 @@ class Unet(nn.Module):
                 text_mask = rearrange(text_mask, 'b n -> b n 1')
                 text_keep_mask = text_mask & text_keep_mask
 
+            null_text_embed = self.null_text_embed.to(text_tokens.dtype) # for some reason pytorch AMP not working
+
             text_tokens = torch.where(
                 text_keep_mask,
                 text_tokens,
-                self.null_text_embed
+                null_text_embed
             )
 
         # main conditioning tokens (c)

setup.py

@@ -10,7 +10,7 @@ setup(
       'dream = dalle2_pytorch.cli:dream'
     ],
   },
-  version = '0.2.7',
+  version = '0.2.10',
   license='MIT',
   description = 'DALL-E 2',
   author = 'Phil Wang',