suppress a warning

README.md

@@ -10,11 +10,9 @@ The main novelty seems to be an extra layer of indirection with the prior networ
 
 This model is SOTA for text-to-image for now.
 
-It may also explore an extension of using <a href="https://huggingface.co/spaces/multimodalart/latentdiffusion">latent diffusion</a> in the decoder from Rombach et al.
-
 Please join <a href="https://discord.gg/xBPBXfcFHd"><img alt="Join us on Discord" src="https://img.shields.io/discord/823813159592001537?color=5865F2&logo=discord&logoColor=white"></a> if you are interested in helping out with the replication
 
-There was enough interest for a Jax version. It will be completed after the Pytorch version shows signs of life on my toy tasks. <a href="https://github.com/lucidrains/dalle2-jax">Placeholder repository</a>. I will also eventually extend this to <a href="https://github.com/lucidrains/dalle2-video">text to video</a>, once the repository is in a good place.
+There was enough interest for a <a href="https://github.com/lucidrains/dalle2-jax">Jax version</a>. I will also eventually extend this to <a href="https://github.com/lucidrains/dalle2-video">text to video</a>, once the repository is in a good place.
 
 ## Install
 
@@ -248,13 +246,6 @@ loss = decoder(images, unet_number = 2)
 loss.backward()
 
 # do the above for many steps for both unets
-
-# then it will learn to generate images based on the CLIP image embeddings
-
-# chaining the unets from lowest resolution to highest resolution (thus cascading)
-
-mock_image_embed = torch.randn(1, 512).cuda()
-images = decoder.sample(mock_image_embed) # (1, 3, 512, 512)
 ```
 
 Finally, to generate the DALL-E2 images from text. Insert the trained `DiffusionPrior` as well as the `Decoder` (which wraps `CLIP`, the causal transformer, and unet(s))
@@ -357,7 +348,8 @@ decoder = Decoder(
     image_sizes = (128, 256),
     clip = clip,
     timesteps = 100,
-    cond_drop_prob = 0.2
+    cond_drop_prob = 0.2,
+    condition_on_text_encodings = False  # set this to True if you wish to condition on text during training and sampling
 ).cuda()
 
 for unet_number in (1, 2):
@@ -385,7 +377,245 @@ You can also train the decoder on images of greater than the size (say 512x512)
 
 For the layperson, no worries, training will all be automated into a CLI tool, at least for small scale training.
 
-## CLI Usage (work in progress)
+## Training on Preprocessed CLIP Embeddings
+
+It is likely, when scaling up, that you would first preprocess your images and text into corresponding embeddings before training the prior network. You can do so easily by simply passing in `image_embed`, `text_embed`, and optionally `text_encodings` and `text_mask`
+
+Working example below
+
+```python
+import torch
+from dalle2_pytorch import DiffusionPriorNetwork, DiffusionPrior, CLIP
+
+# get trained CLIP from step one
+
+clip = CLIP(
+    dim_text = 512,
+    dim_image = 512,
+    dim_latent = 512,
+    num_text_tokens = 49408,
+    text_enc_depth = 6,
+    text_seq_len = 256,
+    text_heads = 8,
+    visual_enc_depth = 6,
+    visual_image_size = 256,
+    visual_patch_size = 32,
+    visual_heads = 8,
+).cuda()
+
+# setup prior network, which contains an autoregressive transformer
+
+prior_network = DiffusionPriorNetwork(
+    dim = 512,
+    depth = 6,
+    dim_head = 64,
+    heads = 8
+).cuda()
+
+# diffusion prior network, which contains the CLIP and network (with transformer) above
+
+diffusion_prior = DiffusionPrior(
+    net = prior_network,
+    clip = clip,
+    timesteps = 100,
+    cond_drop_prob = 0.2,
+    condition_on_text_encodings = False  # this probably should be true, but just to get Laion started
+).cuda()
+
+# mock data
+
+text = torch.randint(0, 49408, (4, 256)).cuda()
+images = torch.randn(4, 3, 256, 256).cuda()
+
+# precompute the text and image embeddings
+# here using the diffusion prior class, but could be done with CLIP alone
+
+clip_image_embeds = diffusion_prior.get_image_embed(images)
+clip_text_embeds = diffusion_prior.get_text_cond(text).get('text_embed')
+
+# feed text and images into diffusion prior network
+
+loss = diffusion_prior(
+    text_embed = clip_text_embeds,
+    image_embed = clip_image_embeds
+)
+
+loss.backward()
+
+# do the above for many many many steps
+# now the diffusion prior can generate image embeddings from the text embeddings
+```
+
+You can also completely go `CLIP`-less, in which case you will need to pass in the `image_embed_dim` into the `DiffusionPrior` on initialization
+
+```python
+import torch
+from dalle2_pytorch import DiffusionPriorNetwork, DiffusionPrior
+
+# setup prior network, which contains an autoregressive transformer
+
+prior_network = DiffusionPriorNetwork(
+    dim = 512,
+    depth = 6,
+    dim_head = 64,
+    heads = 8
+).cuda()
+
+# diffusion prior network, which contains the CLIP and network (with transformer) above
+
+diffusion_prior = DiffusionPrior(
+    net = prior_network,
+    image_embed_dim = 512,               # this needs to be set
+    timesteps = 100,
+    cond_drop_prob = 0.2,
+    condition_on_text_encodings = False  # this probably should be true, but just to get Laion started
+).cuda()
+
+# mock data
+
+text = torch.randint(0, 49408, (4, 256)).cuda()
+images = torch.randn(4, 3, 256, 256).cuda()
+
+# precompute the text and image embeddings
+# here using the diffusion prior class, but could be done with CLIP alone
+
+clip_image_embeds = torch.randn(4, 512).cuda()
+clip_text_embeds = torch.randn(4, 512).cuda()
+
+# feed text and images into diffusion prior network
+
+loss = diffusion_prior(
+    text_embed = clip_text_embeds,
+    image_embed = clip_image_embeds
+)
+
+loss.backward()
+
+# do the above for many many many steps
+# now the diffusion prior can generate image embeddings from the text embeddings
+```
+
+## Experimental
+
+### DALL-E2 with Latent Diffusion
+
+This repository decides to take the next step and offer DALL-E2 combined with <a href="https://huggingface.co/spaces/multimodalart/latentdiffusion">latent diffusion</a>, from Rombach et al.
+
+You can use it as follows. Latent diffusion can be limited to just the first U-Net in the cascade, or to any number you wish.
+
+```python
+import torch
+from dalle2_pytorch import Unet, Decoder, CLIP, VQGanVAE
+
+# trained clip from step 1
+
+clip = CLIP(
+    dim_text = 512,
+    dim_image = 512,
+    dim_latent = 512,
+    num_text_tokens = 49408,
+    text_enc_depth = 1,
+    text_seq_len = 256,
+    text_heads = 8,
+    visual_enc_depth = 1,
+    visual_image_size = 256,
+    visual_patch_size = 32,
+    visual_heads = 8
+)
+
+# 3 unets for the decoder (a la cascading DDPM)
+
+# first two unets are doing latent diffusion
+# vqgan-vae must be trained before hand
+
+vae1 = VQGanVAE(
+    dim = 32,
+    image_size = 256,
+    layers = 3,
+    layer_mults = (1, 2, 4)
+)
+
+vae2 = VQGanVAE(
+    dim = 32,
+    image_size = 512,
+    layers = 3,
+    layer_mults = (1, 2, 4)
+)
+
+unet1 = Unet(
+    dim = 32,
+    image_embed_dim = 512,
+    cond_dim = 128,
+    channels = 3,
+    sparse_attn = True,
+    sparse_attn_window = 2,
+    dim_mults = (1, 2, 4, 8)
+)
+
+unet2 = Unet(
+    dim = 32,
+    image_embed_dim = 512,
+    channels = 3,
+    dim_mults = (1, 2, 4, 8, 16),
+    cond_on_image_embeds = True,
+    cond_on_text_encodings = False
+)
+
+unet3 = Unet(
+    dim = 32,
+    image_embed_dim = 512,
+    channels = 3,
+    dim_mults = (1, 2, 4, 8, 16),
+    cond_on_image_embeds = True,
+    cond_on_text_encodings = False,
+    attend_at_middle = False
+)
+
+# decoder, which contains the unet(s) and clip
+
+decoder = Decoder(
+    clip = clip,
+    vae = (vae1, vae2),                # latent diffusion for unet1 (vae1) and unet2 (vae2), but not for the last unet3
+    unet = (unet1, unet2, unet3),      # insert unets in order of low resolution to highest resolution (you can have as many stages as you want here)
+    image_sizes = (256, 512, 1024),    # resolutions, 256 for first unet, 512 for second, 1024 for third
+    timesteps = 100,
+    cond_drop_prob = 0.2
+).cuda()
+
+# mock images (get a lot of this)
+
+images = torch.randn(1, 3, 1024, 1024).cuda()
+
+# feed images into decoder, specifying which unet you want to train
+# each unet can be trained separately, which is one of the benefits of the cascading DDPM scheme
+
+with decoder.one_unet_in_gpu(1):
+    loss = decoder(images, unet_number = 1)
+    loss.backward()
+
+with decoder.one_unet_in_gpu(2):
+    loss = decoder(images, unet_number = 2)
+    loss.backward()
+
+with decoder.one_unet_in_gpu(3):
+    loss = decoder(images, unet_number = 3)
+    loss.backward()
+
+# do the above for many steps for both unets
+
+# then it will learn to generate images based on the CLIP image embeddings
+
+# chaining the unets from lowest resolution to highest resolution (thus cascading)
+
+mock_image_embed = torch.randn(1, 512).cuda()
+images = decoder.sample(mock_image_embed) # (1, 3, 1024, 1024)
+```
+
+## Training wrapper (wip)
+
+Offer training wrappers
+
+## CLI (wip)
 
 ```bash
 $ dream 'sharing a sunset at the summit of mount everest with my dog'
@@ -393,9 +623,7 @@ $ dream 'sharing a sunset at the summit of mount everest with my dog'
 
 Once built, images will be saved to the same directory the command is invoked
 
-## Training wrapper (wip)
-
-Offer training wrappers
+<a href="https://github.com/lucidrains/big-sleep">template</a>
 
 ## Training CLI (wip)
 
@@ -412,11 +640,18 @@ Offer training wrappers
 - [x] add efficient attention in unet
 - [x] be able to finely customize what to condition on (text, image embed) for specific unet in the cascade (super resolution ddpms near the end may not need too much conditioning)
 - [x] offload unets not being trained on to CPU for memory efficiency (for training each resolution unets separately)
-- [ ] build out latent diffusion architecture, with the vq-reg variant (vqgan-vae), make it completely optional and compatible with cascading ddpms
+- [x] build out latent diffusion architecture, with the vq-reg variant (vqgan-vae), make it completely optional and compatible with cascading ddpms
+- [x] for decoder, allow ability to customize objective (predict epsilon vs x0), in case latent diffusion does better with prediction of x0
+- [x] use attention-based upsampling https://arxiv.org/abs/2112.11435
+- [x] use inheritance just this once for sharing logic between decoder and prior network ddpms
+- [ ] abstract interface for CLIP adapter class, so other CLIPs can be brought in
 - [ ] become an expert with unets, cleanup unet code, make it fully configurable, port all learnings over to https://github.com/lucidrains/x-unet
 - [ ] copy the cascading ddpm code to a separate repo (perhaps https://github.com/lucidrains/denoising-diffusion-pytorch) as the main contribution of dalle2 really is just the prior network
+- [ ] transcribe code to Jax, which lowers the activation energy for distributed training, given access to TPUs
 - [ ] train on a toy task, offer in colab
 - [ ] extend diffusion head to use diffusion-gan (potentially using lightweight-gan) to speed up inference
+- [ ] bring in tools to train vqgan-vae
+- [ ] bring in vit-vqgan https://arxiv.org/abs/2110.04627 for the latent diffusion
 
 ## Citations
 
@@ -448,23 +683,12 @@ Offer training wrappers
 
 ```bibtex
 @inproceedings{Liu2022ACF,
-    title   = {A ConvNet for the 2020s},
+    title   = {A ConvNet for the 2020https://arxiv.org/abs/2112.11435s},
     author  = {Zhuang Liu and Hanzi Mao and Chaozheng Wu and Christoph Feichtenhofer and Trevor Darrell and Saining Xie},
     year    = {2022}
 }
 ```
 
-```bibtex
-@misc{zhang2019root,
-    title   = {Root Mean Square Layer Normalization},
-    author  = {Biao Zhang and Rico Sennrich},
-    year    = {2019},
-    eprint  = {1910.07467},
-    archivePrefix = {arXiv},
-    primaryClass = {cs.LG}
-}
-```
-
 ```bibtex
 @inproceedings{Tu2022MaxViTMV,
     title   = {MaxViT: Multi-Axis Vision Transformer},

dalle2_pytorch/attention.py

@@ -0,0 +1,125 @@
+import torch
+from torch import nn, einsum
+import torch.nn.functional as F
+
+from einops import rearrange, repeat
+
+class LayerNormChan(nn.Module):
+    def __init__(
+        self,
+        dim,
+        eps = 1e-5
+    ):
+        super().__init__()
+        self.eps = eps
+        self.gamma = nn.Parameter(torch.ones(1, dim, 1, 1))
+
+    def forward(self, x):
+        var = torch.var(x, dim = 1, unbiased = False, keepdim = True)
+        mean = torch.mean(x, dim = 1, keepdim = True)
+        return (x - mean) / (var + self.eps).sqrt() * self.gamma
+
+# attention-based upsampling
+# from https://arxiv.org/abs/2112.11435
+
+class QueryAndAttend(nn.Module):
+    def __init__(
+        self,
+        *,
+        dim,
+        num_queries = 1,
+        dim_head = 32,
+        heads = 8,
+        window_size = 3
+    ):
+        super().__init__()
+        self.scale = dim_head ** -0.5
+        inner_dim = dim_head * heads
+        self.heads = heads
+        self.dim_head = dim_head
+        self.window_size = window_size
+        self.num_queries = num_queries
+
+        self.rel_pos_bias = nn.Parameter(torch.randn(heads, num_queries, window_size * window_size, 1, 1))
+
+        self.queries = nn.Parameter(torch.randn(heads, num_queries, dim_head))
+        self.to_kv = nn.Conv2d(dim, dim_head * 2, 1, bias = False)
+        self.to_out = nn.Conv2d(inner_dim, dim, 1, bias = False)
+
+    def forward(self, x):
+        """
+        einstein notation
+        b - batch
+        h - heads
+        l - num queries
+        d - head dimension
+        x - height
+        y - width
+        j - source sequence for attending to (kernel size squared in this case)
+        """
+
+        wsz, heads, dim_head, num_queries = self.window_size, self.heads, self.dim_head, self.num_queries
+        batch, _, height, width = x.shape
+
+        is_one_query = self.num_queries == 1
+
+        # queries, keys, values
+
+        q = self.queries * self.scale
+        k, v = self.to_kv(x).chunk(2, dim = 1)
+
+        # similarities
+
+        sim = einsum('h l d, b d x y -> b h l x y', q, k)
+        sim = rearrange(sim, 'b ... x y -> b (...) x y')
+
+        # unfold the similarity scores, with float(-inf) as padding value
+
+        mask_value = -torch.finfo(sim.dtype).max
+        sim = F.pad(sim, ((wsz // 2,) * 4), value = mask_value)
+        sim = F.unfold(sim, kernel_size = wsz)
+        sim = rearrange(sim, 'b (h l j) (x y) -> b h l j x y', h = heads, l = num_queries, x = height, y = width)
+
+        # rel pos bias
+
+        sim = sim + self.rel_pos_bias
+
+        # numerically stable attention
+
+        sim = sim - sim.amax(dim = -3, keepdim = True).detach()
+        attn = sim.softmax(dim = -3)
+
+        # unfold values
+
+        v = F.pad(v, ((wsz // 2,) * 4), value = 0.)
+        v = F.unfold(v, kernel_size = wsz)
+        v = rearrange(v, 'b (d j) (x y) -> b d j x y', d = dim_head, x = height, y = width)
+
+        # aggregate values
+
+        out = einsum('b h l j x y, b d j x y -> b l h d x y', attn, v)
+
+        # combine heads
+
+        out = rearrange(out, 'b l h d x y -> (b l) (h d) x y')
+        out = self.to_out(out)
+        out = rearrange(out, '(b l) d x y -> b l d x y', b = batch)
+
+        # return original input if one query
+
+        if is_one_query:
+            out = rearrange(out, 'b 1 ... -> b ...')
+
+        return out
+
+class QueryAttnUpsample(nn.Module):
+    def __init__(self, dim, **kwargs):
+        super().__init__()
+        self.norm = LayerNormChan(dim)
+        self.qna = QueryAndAttend(dim = dim, num_queries = 4, **kwargs)
+
+    def forward(self, x):
+        x = self.norm(x)
+        out = self.qna(x)
+        out = rearrange(out, 'b (w1 w2) c h w -> b c (h w1) (w w2)', w1 = 2, w2 = 2)
+        return out

dalle2_pytorch/cli.py

@@ -1,9 +1,51 @@
 import click
+import torch
+import torchvision.transforms as T
+from pathlib import Path
+
+from dalle2_pytorch import DALLE2, Decoder, DiffusionPrior
+
+def safeget(dictionary, keys, default = None):
+    return reduce(lambda d, key: d.get(key, default) if isinstance(d, dict) else default, keys.split('.'), dictionary)
+
+def simple_slugify(text, max_length = 255):
+    return text.replace("-", "_").replace(",", "").replace(" ", "_").replace("|", "--").strip('-_')[:max_length]
+
+def get_pkg_version():
+    from pkg_resources import get_distribution
+    return get_distribution('dalle2_pytorch').version
 
 def main():
     pass
 
 @click.command()
+@click.option('--model', default = './dalle2.pt', help = 'path to trained DALL-E2 model')
+@click.option('--cond_scale', default = 2, help = 'conditioning scale (classifier free guidance) in decoder')
 @click.argument('text')
-def dream(text):
-    return 'not ready yet'
+def dream(
+    model,
+    cond_scale,
+    text
+):
+    model_path = Path(model)
+    full_model_path = str(model_path.resolve())
+    assert model_path.exists(), f'model not found at {full_model_path}'
+    loaded = torch.load(str(model_path))
+
+    version = safeget(loaded, 'version')
+    print(f'loading DALL-E2 from {full_model_path}, saved at version {version} - current package version is {get_pkg_version()}')
+
+    prior_init_params = safeget(loaded, 'init_params.prior')
+    decoder_init_params = safeget(loaded, 'init_params.decoder')
+    model_params = safeget(loaded, 'model_params')
+
+    prior = DiffusionPrior(**prior_init_params)
+    decoder = Decoder(**decoder_init_params)
+
+    dalle2 = DALLE2(prior, decoder)
+    dalle2.load_state_dict(model_params)
+
+    image = dalle2(text, cond_scale = cond_scale)
+
+    pil_image = T.ToPILImage()(image)
+    return pil_image.save(f'./{simple_slugify(text)}.png')

dalle2_pytorch/dalle2_pytorch.py

@@ -16,6 +16,8 @@ from einops_exts.torch import EinopsToAndFrom
 from kornia.filters import gaussian_blur2d
 
 from dalle2_pytorch.tokenizer import tokenizer
+from dalle2_pytorch.vqgan_vae import NullVQGanVAE, VQGanVAE
+from dalle2_pytorch.attention import QueryAttnUpsample
 
 # use x-clip
 
@@ -48,6 +50,12 @@ def is_list_str(x):
         return False
     return all([type(el) == str for el in x])
 
+def pad_tuple_to_length(t, length, fillvalue = None):
+    remain_length = length - len(t)
+    if remain_length <= 0:
+        return t
+    return (*t, *((fillvalue,) * remain_length))
+
 # for controlling freezing of CLIP
 
 def set_module_requires_grad_(module, requires_grad):
@@ -76,7 +84,7 @@ def resize_image_to(t, image_size, mode = 'bilinear'): # take a look at https://
     if orig_image_size == shape:
         return t
 
-    return F.interpolate(t, size = shape, mode = mode)
+    return F.interpolate(t, size = shape, mode = mode, align_corners = False)
 
 # classifier free guidance functions
 
@@ -135,6 +143,92 @@ def sigmoid_beta_schedule(timesteps):
     return torch.sigmoid(betas) * (beta_end - beta_start) + beta_start
 
 
+class BaseGaussianDiffusion(nn.Module):
+    def __init__(self, *, beta_schedule, timesteps, loss_type):
+        super().__init__()
+
+        if beta_schedule == "cosine":
+            betas = cosine_beta_schedule(timesteps)
+        elif beta_schedule == "linear":
+            betas = linear_beta_schedule(timesteps)
+        elif beta_schedule == "quadratic":
+            betas = quadratic_beta_schedule(timesteps)
+        elif beta_schedule == "jsd":
+            betas = 1.0 / torch.linspace(timesteps, 1, timesteps)
+        elif beta_schedule == "sigmoid":
+            betas = sigmoid_beta_schedule(timesteps)
+        else:
+            raise NotImplementedError()
+
+        alphas = 1. - betas
+        alphas_cumprod = torch.cumprod(alphas, axis = 0)
+        alphas_cumprod_prev = F.pad(alphas_cumprod[:-1], (1, 0), value = 1.)
+
+        timesteps, = betas.shape
+        self.num_timesteps = int(timesteps)
+        self.loss_type = loss_type
+
+        self.register_buffer('betas', betas)
+        self.register_buffer('alphas_cumprod', alphas_cumprod)
+        self.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))
+
+        # calculations for posterior q(x_{t-1} | x_t, x_0)
+
+        posterior_variance = betas * (1. - alphas_cumprod_prev) / (1. - alphas_cumprod)
+
+        # above: equal to 1. / (1. / (1. - alpha_cumprod_tm1) + alpha_t / beta_t)
+
+        self.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))
+
+    def q_mean_variance(self, x_start, t):
+        mean = extract(self.sqrt_alphas_cumprod, t, x_start.shape) * x_start
+        variance = extract(1. - self.alphas_cumprod, t, x_start.shape)
+        log_variance = extract(self.log_one_minus_alphas_cumprod, t, x_start.shape)
+        return mean, variance, log_variance
+
+    def q_posterior(self, x_start, x_t, t):
+        posterior_mean = (
+            extract(self.posterior_mean_coef1, t, x_t.shape) * x_start +
+            extract(self.posterior_mean_coef2, t, x_t.shape) * x_t
+        )
+        posterior_variance = extract(self.posterior_variance, t, x_t.shape)
+        posterior_log_variance_clipped = extract(self.posterior_log_variance_clipped, t, x_t.shape)
+        return posterior_mean, posterior_variance, posterior_log_variance_clipped
+
+    def q_sample(self, x_start, t, noise=None):
+        noise = default(noise, lambda: torch.randn_like(x_start))
+
+        return (
+            extract(self.sqrt_alphas_cumprod, t, x_start.shape) * x_start +
+            extract(self.sqrt_one_minus_alphas_cumprod, t, x_start.shape) * noise
+        )
+
+    def predict_start_from_noise(self, x_t, t, noise):
+        return (
+            extract(self.sqrt_recip_alphas_cumprod, t, x_t.shape) * x_t -
+            extract(self.sqrt_recipm1_alphas_cumprod, t, x_t.shape) * noise
+        )
+
+    def sample(self, *args, **kwargs):
+        raise NotImplementedError
+
+    def forward(self, *args, **kwargs):
+        raise NotImplementedError
+
 # diffusion prior
 
 class LayerNorm(nn.Module):
@@ -413,25 +507,41 @@ class DiffusionPriorNetwork(nn.Module):
         image_embed,
         diffusion_timesteps,
         *,
-        text_encodings,
         text_embed,
+        text_encodings = None,
         mask = None,
         cond_drop_prob = 0.2
     ):
-        batch, text_enc_len, device = image_embed.shape[0], text_encodings.shape[-2], image_embed.device
+        batch, dim, device, dtype = *image_embed.shape, image_embed.device, image_embed.dtype
 
         # in section 2.2, last paragraph
         # "... consisting of encoded text, CLIP text embedding, diffusion timestep embedding, noised CLIP image embedding, final embedding for prediction"
 
         text_embed, image_embed = rearrange_many((text_embed, image_embed), 'b d -> b 1 d')
 
+        # make text encodings optional
+        # although the paper seems to suggest it is present <--
+
+        if not exists(text_encodings):
+            text_encodings = torch.empty((batch, 0, dim), device = device, dtype = dtype)
+
+        if not exists(mask):
+            mask = torch.ones((batch, text_encodings.shape[-2]), device = device, dtype = torch.bool)
+
+        # classifier free guidance
+
+        cond_prob_mask = prob_mask_like((batch,), cond_drop_prob, device = device)
+        cond_prob_mask = rearrange(cond_prob_mask, 'b -> b 1')
+
+        mask &= cond_prob_mask
+
+        # whether text embedding is masked or not depends on the classifier free guidance conditional masking
+
+        mask = torch.cat((mask, cond_prob_mask), dim = 1)
+
         # whether text embedding is used for conditioning depends on whether text encodings are available for attention (for classifier free guidance, even though it seems from the paper it was not used in the prior ddpm, as the objective is different)
         # but let's just do it right
 
-        if exists(mask):
-            not_all_masked_out = mask.any(dim = -1)
-            mask = torch.cat((mask, rearrange(not_all_masked_out, 'b -> b 1')), dim = 1)
-
         if exists(mask):
             mask = F.pad(mask, (0, 2), value = True) # extend mask for text embedding, noised image embedding, time step embedding, and learned query
 
@@ -447,16 +557,6 @@ class DiffusionPriorNetwork(nn.Module):
             learned_queries
         ), dim = -2)
 
-        # mask if it doesn't exist
-
-        if not exists(mask):
-            mask = torch.ones((batch, text_enc_len), device = device, dtype = torch.bool)
-
-        # classifier free guidance
-
-        cond_prob_mask = prob_mask_like((batch,), cond_drop_prob, device = device)
-        mask &= rearrange(cond_prob_mask, 'b -> b 1')
-
         # attend
 
         tokens = self.causal_transformer(tokens, mask = mask)
@@ -467,122 +567,80 @@ class DiffusionPriorNetwork(nn.Module):
 
         return pred_image_embed
 
-class DiffusionPrior(nn.Module):
+class DiffusionPrior(BaseGaussianDiffusion):
     def __init__(
         self,
         net,
         *,
-        clip,
+        clip = None,
+        image_embed_dim = None,
+        image_size = None,
+        image_channels = 3,
         timesteps = 1000,
         cond_drop_prob = 0.2,
         loss_type = "l1",
-        predict_x0 = True,
+        predict_x_start = True,
         beta_schedule = "cosine",
+        condition_on_text_encodings = True, # the paper suggests this is needed, but you can turn it off for your CLIP preprocessed text embed -> image embed training
     ):
-        super().__init__()
-        assert isinstance(clip, CLIP)
-        freeze_model_and_make_eval_(clip)
-        self.clip = clip
+        super().__init__(
+            beta_schedule = beta_schedule,
+            timesteps = timesteps,
+            loss_type = loss_type
+        )
+
+        if exists(clip):
+            assert isinstance(clip, CLIP)
+            freeze_model_and_make_eval_(clip)
+            self.clip = clip
+        else:
+            assert exists(image_embed_dim), 'latent dimension must be given, if training prior network without CLIP given'
+            self.clip = None
 
         self.net = net
-        self.image_embed_dim = clip.dim_latent
-        self.channels = clip.image_channels
-        self.image_size = clip.image_size
-        self.cond_drop_prob = cond_drop_prob
+        self.image_embed_dim = default(image_embed_dim, lambda: clip.dim_latent)
+        self.channels = default(image_channels, lambda: clip.image_channels)
 
-        self.predict_x0 = predict_x0
+        self.cond_drop_prob = cond_drop_prob
+        self.condition_on_text_encodings = condition_on_text_encodings
+
+        self.predict_x_start = predict_x_start
         # in paper, they do not predict the noise, but predict x0 directly for image embedding, claiming empirically better results. I'll just offer both.
 
-        if beta_schedule == "cosine":
-            betas = cosine_beta_schedule(timesteps)
-        elif beta_schedule == "linear":
-            betas = linear_beta_schedule(timesteps)
-        elif beta_schedule == "quadratic":
-            betas = quadratic_beta_schedule(timesteps)
-        elif beta_schedule == "jsd":
-            betas = 1.0 / torch.linspace(timesteps, 1, timesteps)
-        elif beta_schedule == "sigmoid":
-            betas = sigmoid_beta_schedule(timesteps)
-        else:
-            raise NotImplementedError()
-
-        alphas = 1. - betas
-        alphas_cumprod = torch.cumprod(alphas, axis = 0)
-        alphas_cumprod_prev = F.pad(alphas_cumprod[:-1], (1, 0), value = 1.)
-
-        timesteps, = betas.shape
-        self.num_timesteps = int(timesteps)
-        self.loss_type = loss_type
-
-        self.register_buffer('betas', betas)
-        self.register_buffer('alphas_cumprod', alphas_cumprod)
-        self.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))
-
-        # calculations for posterior q(x_{t-1} | x_t, x_0)
-
-        posterior_variance = betas * (1. - alphas_cumprod_prev) / (1. - alphas_cumprod)
-
-        # above: equal to 1. / (1. / (1. - alpha_cumprod_tm1) + alpha_t / beta_t)
-
-        self.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))
-
+    @torch.no_grad()
     def get_image_embed(self, image):
+        assert exists(self.clip)
+
         image_encoding = self.clip.visual_transformer(image)
         image_cls = image_encoding[:, 0]
         image_embed = self.clip.to_visual_latent(image_cls)
         return l2norm(image_embed)
 
+    @torch.no_grad()
     def get_text_cond(self, text):
+        assert exists(self.clip)
+
         text_encodings = self.clip.text_transformer(text)
         text_cls, text_encodings = text_encodings[:, 0], text_encodings[:, 1:]
         text_embed = self.clip.to_text_latent(text_cls)
         text_embed = l2norm(text_embed)
+
+        if not self.condition_on_text_encodings:            
+            return dict(text_embed = text_embed)
+
         return dict(text_encodings = text_encodings, text_embed = text_embed, mask = text != 0)
 
-    def q_mean_variance(self, x_start, t):
-        mean = extract(self.sqrt_alphas_cumprod, t, x_start.shape) * x_start
-        variance = extract(1. - self.alphas_cumprod, t, x_start.shape)
-        log_variance = extract(self.log_one_minus_alphas_cumprod, t, x_start.shape)
-        return mean, variance, log_variance
-
-    def predict_start_from_noise(self, x_t, t, noise):
-        return (
-            extract(self.sqrt_recip_alphas_cumprod, t, x_t.shape) * x_t -
-            extract(self.sqrt_recipm1_alphas_cumprod, t, x_t.shape) * noise
-        )
-
-    def q_posterior(self, x_start, x_t, t):
-        posterior_mean = (
-            extract(self.posterior_mean_coef1, t, x_t.shape) * x_start +
-            extract(self.posterior_mean_coef2, t, x_t.shape) * x_t
-        )
-        posterior_variance = extract(self.posterior_variance, t, x_t.shape)
-        posterior_log_variance_clipped = extract(self.posterior_log_variance_clipped, t, x_t.shape)
-        return posterior_mean, posterior_variance, posterior_log_variance_clipped
-
     def p_mean_variance(self, x, t, text_cond, clip_denoised: bool):
-        if self.predict_x0:
-            x_recon = self.net(x, t, **text_cond)
+        pred = self.net(x, t, **text_cond)
+
+        if self.predict_x_start:
+            x_recon = pred
             # not 100% sure of this above line - for any spectators, let me know in the github issues (or through a pull request) if you know how to correctly do this
             # i'll be rereading https://arxiv.org/abs/2111.14822, where i think a similar approach is taken
         else:
-            x_recon = self.predict_start_from_noise(x, t = t, noise = self.net(x, t, **text_cond))
+            x_recon = self.predict_start_from_noise(x, t = t, noise = pred)
 
-        if clip_denoised:
+        if clip_denoised and not self.predict_x_start:
             x_recon.clamp_(-1., 1.)
 
         model_mean, posterior_variance, posterior_log_variance = self.q_posterior(x_start=x_recon, x_t=x, t=t)
@@ -608,14 +666,6 @@ class DiffusionPrior(nn.Module):
             img = self.p_sample(img, torch.full((b,), i, device = device, dtype = torch.long), text_cond = text_cond)
         return img
 
-    def q_sample(self, x_start, t, noise=None):
-        noise = default(noise, lambda: torch.randn_like(x_start))
-
-        return (
-            extract(self.sqrt_alphas_cumprod, t, x_start.shape) * x_start +
-            extract(self.sqrt_one_minus_alphas_cumprod, t, x_start.shape) * noise
-        )
-
     def p_losses(self, image_embed, t, text_cond, noise = None):
         noise = default(noise, lambda: torch.randn_like(image_embed))
 
@@ -628,7 +678,7 @@ class DiffusionPrior(nn.Module):
             **text_cond
         )
 
-        to_predict = noise if not self.predict_x0 else image_embed
+        to_predict = noise if not self.predict_x_start else image_embed
 
         if self.loss_type == 'l1':
             loss = F.l1_loss(to_predict, x_recon)
@@ -667,13 +717,41 @@ class DiffusionPrior(nn.Module):
         top_image_embeds = image_embeds.gather(1, top_sim_indices)
         return rearrange(top_image_embeds, 'b 1 d -> b d')
 
-    def forward(self, text, image, *args, **kwargs):
-        b, device, img_size, = image.shape[0], image.device, self.image_size
-        check_shape(image, 'b c h w', h = img_size, w = img_size, c = self.channels)
+    def forward(
+        self,
+        text = None,
+        image = None,
+        text_embed = None,      # allow for training on preprocessed CLIP text and image embeddings
+        image_embed = None,
+        text_encodings = None,  # as well as CLIP text encodings
+        text_mask = None,       # text mask <- may eventually opt for the learned padding tokens technique from DALL-E1 to reduce complexity
+        *args,
+        **kwargs
+    ):
+        assert exists(text) ^ exists(text_embed), 'either text or text embedding must be supplied'
+        assert exists(image) ^ exists(image_embed), 'either text or text embedding must be supplied'
+        assert not (self.condition_on_text_encodings and (not exists(text_encodings) and not exists(text))), 'text encodings must be present if you specified you wish to condition on it on initialization'
 
-        times = torch.randint(0, self.num_timesteps, (b,), device = device, dtype = torch.long)
-        image_embed = self.get_image_embed(image)
-        text_cond = self.get_text_cond(text)
+        if exists(image):
+            image_embed = self.get_image_embed(image)
+
+        # calculate text conditionings, based on what is passed in
+
+        if exists(text):
+            text_cond = self.get_text_cond(text)
+        else:
+            text_cond = dict(
+                text_embed = text_embed,
+                text_encodings = text_encodings,
+                mask = text_mask
+            )
+
+        # timestep conditioning from ddpm
+
+        batch, device = image_embed.shape[0], image_embed.device
+        times = torch.randint(0, self.num_timesteps, (batch,), device = device, dtype = torch.long)
+
+        # calculate forward loss
 
         loss = self.p_losses(image_embed, times, text_cond = text_cond, *args, **kwargs)
         return loss
@@ -841,6 +919,7 @@ class Unet(nn.Module):
         sparse_attn_window = 8,  # window size for sparse attention
         attend_at_middle = True, # whether to have a layer of attention at the bottleneck (can turn off for higher resolution in cascading DDPM, before bringing in efficient attention)
         cond_on_text_encodings = False,
+        max_text_len = 256,
         cond_on_image_embeds = False,
     ):
         super().__init__()
@@ -891,7 +970,7 @@ class Unet(nn.Module):
         # for classifier free guidance
 
         self.null_image_embed = nn.Parameter(torch.randn(1, num_image_tokens, cond_dim))
-        self.null_text_embed = nn.Parameter(torch.randn(1, 1, cond_dim))
+        self.null_text_embed = nn.Parameter(torch.randn(1, max_text_len, cond_dim))
 
         # attention related params
 
@@ -940,11 +1019,16 @@ class Unet(nn.Module):
 
     # if the current settings for the unet are not correct
     # for cascading DDPM, then reinit the unet with the right settings
-    def force_lowres_cond(self, lowres_cond):
-        if lowres_cond == self.lowres_cond:
+    def cast_model_parameters(
+        self,
+        *,
+        lowres_cond,
+        channels
+    ):
+        if lowres_cond == self.lowres_cond and channels == self.channels:
             return self
 
-        updated_kwargs = {**self._locals, 'lowres_cond': lowres_cond}
+        updated_kwargs = {**self._locals, 'lowres_cond': lowres_cond, 'channels': channels}
         return self.__class__(**updated_kwargs)
 
     def forward_with_cond_scale(
@@ -1014,7 +1098,7 @@ class Unet(nn.Module):
             text_tokens = torch.where(
                 cond_prob_mask,
                 text_tokens,
-                self.null_text_embed
+                self.null_text_embed[:, :text_tokens.shape[1]]
             )
 
         # main conditioning tokens (c)
@@ -1094,37 +1178,65 @@ class LowresConditioner(nn.Module):
 
         return cond_fmap
 
-class Decoder(nn.Module):
+class Decoder(BaseGaussianDiffusion):
     def __init__(
         self,
         unet,
         *,
         clip,
+        vae = tuple(),
         timesteps = 1000,
         cond_drop_prob = 0.2,
         loss_type = 'l1',
         beta_schedule = 'cosine',
+        predict_x_start = False,
+        predict_x_start_for_latent_diffusion = False,
         image_sizes = None,                         # for cascading ddpm, image size at each stage
         lowres_cond_upsample_mode = 'bilinear',     # cascading ddpm - low resolution upsample mode
         lowres_downsample_first = True,             # cascading ddpm - resizes to lower resolution, then to next conditional resolution + blur
         blur_sigma = 0.1,                           # cascading ddpm - blur sigma
         blur_kernel_size = 3,                       # cascading ddpm - blur kernel size
+        condition_on_text_encodings = False,        # the paper suggested that this didn't do much in the decoder, but i'm allowing the option for experimentation
     ):
-        super().__init__()
+        super().__init__(
+            beta_schedule = beta_schedule,
+            timesteps = timesteps,
+            loss_type = loss_type
+        )
+
         assert isinstance(clip, CLIP)
         freeze_model_and_make_eval_(clip)
         self.clip = clip
         self.clip_image_size = clip.image_size
         self.channels = clip.image_channels
 
+        self.condition_on_text_encodings = condition_on_text_encodings
+
         # automatically take care of ensuring that first unet is unconditional
         # while the rest of the unets are conditioned on the low resolution image produced by previous unet
 
+        unets = cast_tuple(unet)
+        vaes = pad_tuple_to_length(cast_tuple(vae), len(unets), fillvalue = NullVQGanVAE(channels = self.channels))
+
         self.unets = nn.ModuleList([])
-        for ind, one_unet in enumerate(cast_tuple(unet)):
+        self.vaes = nn.ModuleList([])
+
+        for ind, (one_unet, one_vae) in enumerate(zip(unets, vaes)):
+            assert isinstance(one_unet, Unet)
+            assert isinstance(one_vae, (VQGanVAE, NullVQGanVAE))
+
             is_first = ind == 0
-            one_unet = one_unet.force_lowres_cond(not is_first)
+            latent_dim = one_vae.encoded_dim if exists(one_vae) else None
+
+            unet_channels = default(latent_dim, self.channels)
+
+            one_unet = one_unet.cast_model_parameters(
+                lowres_cond = not is_first,
+                channels = unet_channels
+            )
+
             self.unets.append(one_unet)
+            self.vaes.append(one_vae.copy_for_eval())
 
         # unet image sizes
 
@@ -1135,6 +1247,10 @@ class Decoder(nn.Module):
         self.image_sizes = image_sizes
         self.sample_channels = cast_tuple(self.channels, len(image_sizes))
 
+        # predict x0 config
+
+        self.predict_x_start = cast_tuple(predict_x_start, len(unets)) if not predict_x_start_for_latent_diffusion else tuple(map(lambda t: isinstance(t, VQGanVAE), self.vaes))
+
         # cascading ddpm related stuff
 
         lowres_conditions = tuple(map(lambda t: t.lowres_cond, self.unets))
@@ -1151,55 +1267,6 @@ class Decoder(nn.Module):
 
         self.cond_drop_prob = cond_drop_prob
 
-        # noise schedule
-
-        if beta_schedule == "cosine":
-            betas = cosine_beta_schedule(timesteps)
-        elif beta_schedule == "linear":
-            betas = linear_beta_schedule(timesteps)
-        elif beta_schedule == "quadratic":
-            betas = quadratic_beta_schedule(timesteps)
-        elif beta_schedule == "jsd":
-            betas = 1.0 / torch.linspace(timesteps, 1, timesteps)
-        elif beta_schedule == "sigmoid":
-            betas = sigmoid_beta_schedule(timesteps)
-        else:
-            raise NotImplementedError()
-
-        alphas = 1. - betas
-        alphas_cumprod = torch.cumprod(alphas, axis = 0)
-        alphas_cumprod_prev = F.pad(alphas_cumprod[:-1], (1, 0), value = 1.)
-
-        timesteps, = betas.shape
-        self.num_timesteps = int(timesteps)
-        self.loss_type = loss_type
-
-        self.register_buffer('betas', betas)
-        self.register_buffer('alphas_cumprod', alphas_cumprod)
-        self.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))
-
-        # calculations for posterior q(x_{t-1} | x_t, x_0)
-
-        posterior_variance = betas * (1. - alphas_cumprod_prev) / (1. - alphas_cumprod)
-
-        # above: equal to 1. / (1. / (1. - alpha_cumprod_tm1) + alpha_t / beta_t)
-
-        self.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))
-
     def get_unet(self, unet_number):
         assert 0 < unet_number <= len(self.unets)
         index = unet_number - 1
@@ -1219,10 +1286,12 @@ class Decoder(nn.Module):
         yield
         unet.cpu()
 
+    @torch.no_grad()
     def get_text_encodings(self, text):
         text_encodings = self.clip.text_transformer(text)
         return text_encodings[:, 1:]
 
+    @torch.no_grad()
     def get_image_embed(self, image):
         image = resize_image_to(image, self.clip_image_size)
         image_encoding = self.clip.visual_transformer(image)
@@ -1230,67 +1299,51 @@ class Decoder(nn.Module):
         image_embed = self.clip.to_visual_latent(image_cls)
         return l2norm(image_embed)
 
-    def q_mean_variance(self, x_start, t):
-        mean = extract(self.sqrt_alphas_cumprod, t, x_start.shape) * x_start
-        variance = extract(1. - self.alphas_cumprod, t, x_start.shape)
-        log_variance = extract(self.log_one_minus_alphas_cumprod, t, x_start.shape)
-        return mean, variance, log_variance
+    def p_mean_variance(self, unet, x, t, image_embed, text_encodings = None, lowres_cond_img = None, clip_denoised = True, predict_x_start = False, cond_scale = 1.):
+        pred = unet.forward_with_cond_scale(x, t, image_embed = image_embed, text_encodings = text_encodings, cond_scale = cond_scale, lowres_cond_img = lowres_cond_img)
 
-    def predict_start_from_noise(self, x_t, t, noise):
-        return (
-            extract(self.sqrt_recip_alphas_cumprod, t, x_t.shape) * x_t -
-            extract(self.sqrt_recipm1_alphas_cumprod, t, x_t.shape) * noise
-        )
+        if predict_x_start:
+            x_recon = pred
+        else:
+            x_recon = self.predict_start_from_noise(x, t = t, noise = pred)
 
-    def q_posterior(self, x_start, x_t, t):
-        posterior_mean = (
-            extract(self.posterior_mean_coef1, t, x_t.shape) * x_start +
-            extract(self.posterior_mean_coef2, t, x_t.shape) * x_t
-        )
-        posterior_variance = extract(self.posterior_variance, t, x_t.shape)
-        posterior_log_variance_clipped = extract(self.posterior_log_variance_clipped, t, x_t.shape)
-        return posterior_mean, posterior_variance, posterior_log_variance_clipped
-
-    def p_mean_variance(self, unet, x, t, image_embed, text_encodings = None, lowres_cond_img = None, clip_denoised = True, cond_scale = 1.):
-        pred_noise = unet.forward_with_cond_scale(x, t, image_embed = image_embed, text_encodings = text_encodings, cond_scale = cond_scale, lowres_cond_img = lowres_cond_img)
-        x_recon = self.predict_start_from_noise(x, t = t, noise = pred_noise)
-
-        if clip_denoised:
+        if clip_denoised and not predict_x_start:
             x_recon.clamp_(-1., 1.)
 
         model_mean, posterior_variance, posterior_log_variance = self.q_posterior(x_start=x_recon, x_t=x, t=t)
         return model_mean, posterior_variance, posterior_log_variance
 
     @torch.no_grad()
-    def p_sample(self, unet, x, t, image_embed, text_encodings = None, cond_scale = 1., lowres_cond_img = None, clip_denoised = True, repeat_noise = False):
+    def p_sample(self, unet, x, t, image_embed, text_encodings = None, cond_scale = 1., lowres_cond_img = None, predict_x_start = False, clip_denoised = True, repeat_noise = False):
         b, *_, device = *x.shape, x.device
-        model_mean, _, model_log_variance = self.p_mean_variance(unet, x = x, t = t, image_embed = image_embed, text_encodings = text_encodings, cond_scale = cond_scale, lowres_cond_img = lowres_cond_img, clip_denoised = clip_denoised)
+        model_mean, _, model_log_variance = self.p_mean_variance(unet, x = x, t = t, image_embed = image_embed, text_encodings = text_encodings, cond_scale = cond_scale, lowres_cond_img = lowres_cond_img, clip_denoised = clip_denoised, predict_x_start = predict_x_start)
         noise = noise_like(x.shape, device, repeat_noise)
         # no noise when t == 0
         nonzero_mask = (1 - (t == 0).float()).reshape(b, *((1,) * (len(x.shape) - 1)))
         return model_mean + nonzero_mask * (0.5 * model_log_variance).exp() * noise
 
     @torch.no_grad()
-    def p_sample_loop(self, unet, shape, image_embed, lowres_cond_img = None, text_encodings = None, cond_scale = 1):
+    def p_sample_loop(self, unet, shape, image_embed, predict_x_start = False, lowres_cond_img = None, text_encodings = None, cond_scale = 1):
         device = self.betas.device
 
         b = shape[0]
         img = torch.randn(shape, device = device)
 
         for i in tqdm(reversed(range(0, self.num_timesteps)), desc = 'sampling loop time step', total = self.num_timesteps):
-            img = self.p_sample(unet, img, torch.full((b,), i, device = device, dtype = torch.long), image_embed = image_embed, text_encodings = text_encodings, cond_scale = cond_scale, lowres_cond_img = lowres_cond_img)
+            img = self.p_sample(
+                unet,
+                img,
+                torch.full((b,), i, device = device, dtype = torch.long),
+                image_embed = image_embed,
+                text_encodings = text_encodings,
+                cond_scale = cond_scale,
+                lowres_cond_img = lowres_cond_img,
+                predict_x_start = predict_x_start
+            )
 
         return img
 
-    def q_sample(self, x_start, t, noise=None):
-        noise = default(noise, lambda: torch.randn_like(x_start))
-
-        return (
-            extract(self.sqrt_alphas_cumprod, t, x_start.shape) * x_start +
-            extract(self.sqrt_one_minus_alphas_cumprod, t, x_start.shape) * noise
-        )
-
-    def p_losses(self, unet, x_start, t, *, image_embed, lowres_cond_img = None, text_encodings = None, noise = None):
+    def p_losses(self, unet, x_start, t, *, image_embed, lowres_cond_img = None, text_encodings = None, predict_x_start = False, noise = None):
         noise = default(noise, lambda: torch.randn_like(x_start))
 
         x_noisy = self.q_sample(x_start = x_start, t = t, noise = noise)
@@ -1304,12 +1357,14 @@ class Decoder(nn.Module):
             cond_drop_prob = self.cond_drop_prob
         )
 
+        target = noise if not predict_x_start else x_start
+
         if self.loss_type == 'l1':
-            loss = F.l1_loss(noise, x_recon)
+            loss = F.l1_loss(target, x_recon)
         elif self.loss_type == 'l2':
-            loss = F.mse_loss(noise, x_recon)
+            loss = F.mse_loss(target, x_recon)
         elif self.loss_type == "huber":
-            loss = F.smooth_l1_loss(noise, x_recon)
+            loss = F.smooth_l1_loss(target, x_recon)
         else:
             raise NotImplementedError()
 
@@ -1322,27 +1377,46 @@ class Decoder(nn.Module):
 
         text_encodings = self.get_text_encodings(text) if exists(text) else None
 
+        assert not (self.condition_on_text_encodings and not exists(text_encodings)), 'text or text encodings must be passed into decoder if specified'
+
         img = None
 
-        for unet, channel, image_size in tqdm(zip(self.unets, self.sample_channels, self.image_sizes)):
+        for unet, vae, channel, image_size, predict_x_start in tqdm(zip(self.unets, self.vaes, self.sample_channels, self.image_sizes, self.predict_x_start)):
             with self.one_unet_in_gpu(unet = unet):
-                lowres_cond_img = self.to_lowres_cond(
-                    img,
-                    target_image_size = image_size
-                ) if unet.lowres_cond else None
+                lowres_cond_img = None
+                shape = (batch_size, channel, image_size, image_size)
+
+                if unet.lowres_cond:
+                    lowres_cond_img = self.to_lowres_cond(img, target_image_size = image_size)
+
+                image_size = vae.get_encoded_fmap_size(image_size)
+                shape = (batch_size, vae.encoded_dim, image_size, image_size)
+
+                if exists(lowres_cond_img):
+                    lowres_cond_img = vae.encode(lowres_cond_img)
 
                 img = self.p_sample_loop(
                     unet,
-                    (batch_size, channel, image_size, image_size),
+                    shape,
                     image_embed = image_embed,
                     text_encodings = text_encodings,
                     cond_scale = cond_scale,
+                    predict_x_start = predict_x_start,
                     lowres_cond_img = lowres_cond_img
                 )
 
+                img = vae.decode(img)
+
         return img
 
-    def forward(self, image, text = None, image_embed = None, text_encodings = None, unet_number = None):
+    def forward(
+        self,
+        image,
+        text = None,
+        image_embed = None,
+        text_encodings = None,
+        unet_number = None
+    ):
         assert not (len(self.unets) > 1 and not exists(unet_number)), f'you must specify which unet you want trained, from a range of 1 to {len(self.unets)}, if you are training cascading DDPM (multiple unets)'
         unet_number = default(unet_number, 1)
         unet_index = unet_number - 1
@@ -1350,6 +1424,8 @@ class Decoder(nn.Module):
         unet = self.get_unet(unet_number)
 
         target_image_size = self.image_sizes[unet_index]
+        vae = self.vaes[unet_index]
+        predict_x_start = self.predict_x_start[unet_index]
 
         b, c, h, w, device, = *image.shape, image.device
 
@@ -1363,9 +1439,19 @@ class Decoder(nn.Module):
 
         text_encodings = self.get_text_encodings(text) if exists(text) and not exists(text_encodings) else None
 
+        assert not (self.condition_on_text_encodings and not exists(text_encodings)), 'text or text encodings must be passed into decoder if specified'
+
         lowres_cond_img = self.to_lowres_cond(image, target_image_size = target_image_size, downsample_image_size = self.image_sizes[unet_index - 1]) if unet_number > 1 else None
-        ddpm_image = resize_image_to(image, target_image_size)
-        return self.p_losses(unet, ddpm_image, times, image_embed = image_embed, text_encodings = text_encodings, lowres_cond_img = lowres_cond_img)
+        image = resize_image_to(image, target_image_size)
+
+        vae.eval()
+        with torch.no_grad():
+            image = vae.encode(image)
+
+            if exists(lowres_cond_img):
+                lowres_cond_img = vae.encode(lowres_cond_img)
+
+        return self.p_losses(unet, image, times, image_embed = image_embed, text_encodings = text_encodings, lowres_cond_img = lowres_cond_img, predict_x_start = predict_x_start)
 
 # main class
 
@@ -1382,7 +1468,9 @@ class DALLE2(nn.Module):
         assert isinstance(decoder, Decoder)
         self.prior = prior
         self.decoder = decoder
+
         self.prior_num_samples = prior_num_samples
+        self.decoder_need_text_cond = self.decoder.condition_on_text_encodings
 
     @torch.no_grad()
     @eval_decorator
@@ -1392,11 +1480,18 @@ class DALLE2(nn.Module):
         cond_scale = 1.
     ):
         device = next(self.parameters()).device
+        one_text = isinstance(text, str) or (not is_list_str(text) and text.shape[0] == 1)
 
         if isinstance(text, str) or is_list_str(text):
             text = [text] if not isinstance(text, (list, tuple)) else text
             text = tokenizer.tokenize(text).to(device)
 
         image_embed = self.prior.sample(text, num_samples_per_batch = self.prior_num_samples)
-        images = self.decoder.sample(image_embed, cond_scale = cond_scale)
+
+        text_cond = text if self.decoder_need_text_cond else None
+        images = self.decoder.sample(image_embed, text = text_cond, cond_scale = cond_scale)
+
+        if one_text:
+            return images[0]
+
         return images

dalle2_pytorch/train.py

@@ -0,0 +1,53 @@
+import copy
+import torch
+from torch import nn
+
+# exponential moving average wrapper
+
+class EMA(nn.Module):
+    def __init__(
+        self,
+        model,
+        beta = 0.99,
+        ema_update_after_step = 1000,
+        ema_update_every = 10,
+    ):
+        super().__init__()
+        self.beta = beta
+        self.online_model = model
+        self.ema_model = copy.deepcopy(model)
+
+        self.ema_update_after_step = ema_update_after_step # only start EMA after this step number, starting at 0
+        self.ema_update_every = ema_update_every
+
+        self.register_buffer('initted', torch.Tensor([False]))
+        self.register_buffer('step', torch.tensor([0.]))
+
+    def update(self):
+        self.step += 1
+
+        if self.step <= self.ema_update_after_step or (self.step % self.ema_update_every) != 0:
+            return
+
+        if not self.initted:
+            self.ema_model.state_dict(self.online_model.state_dict())
+            self.initted.data.copy_(torch.Tensor([True]))
+
+        self.update_moving_average(self.ema_model, self.online_model)
+
+    def update_moving_average(ma_model, current_model):
+        def calculate_ema(beta, old, new):
+            if not exists(old):
+                return new
+            return old * beta + (1 - beta) * new
+
+        for current_params, ma_params in zip(current_model.parameters(), ma_model.parameters()):
+            old_weight, up_weight = ma_params.data, current_params.data
+            ma_params.data = calculate_ema(self.beta, old_weight, up_weight)
+
+        for current_buffer, ma_buffer in zip(current_model.buffers(), ma_model.buffers()):
+            new_buffer_value = calculate_ema(self.beta, ma_buffer, current_buffer)
+            ma_buffer.copy_(new_buffer_value)
+
+    def __call__(self, *args, **kwargs):
+        return self.ema_model(*args, **kwargs)

dalle2_pytorch/vqgan_vae.py

@@ -13,6 +13,8 @@ import torchvision
 
 from einops import rearrange, reduce, repeat
 
+from dalle2_pytorch.attention import QueryAttnUpsample
+
 # constants
 
 MList = nn.ModuleList
@@ -243,6 +245,7 @@ class ResBlock(nn.Module):
     def forward(self, x):
         return self.net(x) + x
 
+# vqgan attention layer
 class VQGanAttention(nn.Module):
     def __init__(
         self,
@@ -287,6 +290,28 @@ class VQGanAttention(nn.Module):
 
         return out + residual
 
+class NullVQGanVAE(nn.Module):
+    def __init__(
+        self,
+        *,
+        channels
+    ):
+        super().__init__()
+        self.encoded_dim = channels
+        self.layers = 0
+
+    def get_encoded_fmap_size(self, size):
+        return size
+
+    def copy_for_eval(self):
+        return self
+
+    def encode(self, x):
+        return x
+
+    def decode(self, x):
+        return x
+
 class VQGanVAE(nn.Module):
     def __init__(
         self,
@@ -294,7 +319,7 @@ class VQGanVAE(nn.Module):
         dim,
         image_size,
         channels = 3,
-        num_layers = 4,
+        layers = 4,
         layer_mults = None,
         l2_recon_loss = False,
         use_hinge_loss = True,
@@ -321,37 +346,39 @@ class VQGanVAE(nn.Module):
 
         self.image_size = image_size
         self.channels = channels
-        self.num_layers = num_layers
-        self.fmap_size = image_size // (num_layers ** 2)
+        self.layers = layers
+        self.fmap_size = image_size // (layers ** 2)
         self.codebook_size = vq_codebook_size
 
         self.encoders = MList([])
         self.decoders = MList([])
 
-        layer_mults = default(layer_mults, list(map(lambda t: 2 ** t, range(num_layers))))
-        assert len(layer_mults) == num_layers, 'layer multipliers must be equal to designated number of layers'
+        layer_mults = default(layer_mults, list(map(lambda t: 2 ** t, range(layers))))
+        assert len(layer_mults) == layers, 'layer multipliers must be equal to designated number of layers'
 
         layer_dims = [dim * mult for mult in layer_mults]
         dims = (dim, *layer_dims)
         codebook_dim = layer_dims[-1]
 
+        self.encoded_dim = dims[-1]
+
         dim_pairs = zip(dims[:-1], dims[1:])
 
         append = lambda arr, t: arr.append(t)
         prepend = lambda arr, t: arr.insert(0, t)
 
         if not isinstance(num_resnet_blocks, tuple):
-            num_resnet_blocks = (*((0,) * (num_layers - 1)), num_resnet_blocks)
+            num_resnet_blocks = (*((0,) * (layers - 1)), num_resnet_blocks)
 
         if not isinstance(use_attn, tuple):
-            use_attn = (*((False,) * (num_layers - 1)), use_attn)
+            use_attn = (*((False,) * (layers - 1)), use_attn)
 
-        assert len(num_resnet_blocks) == num_layers, 'number of resnet blocks config must be equal to number of layers'
-        assert len(use_attn) == num_layers
+        assert len(num_resnet_blocks) == layers, 'number of resnet blocks config must be equal to number of layers'
+        assert len(use_attn) == layers
 
-        for layer_index, (dim_in, dim_out), layer_num_resnet_blocks, layer_use_attn in zip(range(num_layers), dim_pairs, num_resnet_blocks, use_attn):
+        for layer_index, (dim_in, dim_out), layer_num_resnet_blocks, layer_use_attn in zip(range(layers), dim_pairs, num_resnet_blocks, use_attn):
             append(self.encoders, nn.Sequential(nn.Conv2d(dim_in, dim_out, 4, stride = 2, padding = 1), leaky_relu()))
-            prepend(self.decoders, nn.Sequential(nn.Upsample(scale_factor = 2, mode = 'bilinear', align_corners = False), nn.Conv2d(dim_out, dim_in, 3, padding = 1), leaky_relu()))
+            prepend(self.decoders, nn.Sequential(nn.ConvTranspose2d(dim_out, dim_in, 4, 2, 1), leaky_relu()))
 
             if layer_use_attn:
                 prepend(self.decoders, VQGanAttention(dim = dim_out, heads = attn_heads, dim_head = attn_dim_head, dropout = attn_dropout))
@@ -405,6 +432,9 @@ class VQGanVAE(nn.Module):
         self.discr_loss = hinge_discr_loss if use_hinge_loss else bce_discr_loss
         self.gen_loss = hinge_gen_loss if use_hinge_loss else bce_gen_loss
 
+    def get_encoded_fmap_size(self, image_size):
+        return image_size // (2 ** self.layers)
+
     def copy_for_eval(self):
         device = next(self.parameters()).device
         vae_copy = copy.deepcopy(self.cpu())
@@ -434,12 +464,15 @@ class VQGanVAE(nn.Module):
 
         return fmap
 
-    def decode(self, fmap):
+    def decode(self, fmap, return_indices_and_loss = False):
         fmap, indices, commit_loss = self.vq(fmap)
 
         for dec in self.decoders:
             fmap = dec(fmap)
 
+        if not return_indices_and_loss:
+            return fmap
+
         return fmap, indices, commit_loss
 
     def forward(
@@ -447,7 +480,8 @@ class VQGanVAE(nn.Module):
         img,
         return_loss = False,
         return_discr_loss = False,
-        return_recons = False
+        return_recons = False,
+        add_gradient_penalty = True
     ):
         batch, channels, height, width, device = *img.shape, img.device
         assert height == self.image_size and width == self.image_size, 'height and width of input image must be equal to {self.image_size}'
@@ -455,7 +489,7 @@ class VQGanVAE(nn.Module):
 
         fmap = self.encode(img)
 
-        fmap, indices, commit_loss = self.decode(fmap)
+        fmap, indices, commit_loss = self.decode(fmap, return_indices_and_loss = True)
 
         if not return_loss and not return_discr_loss:
             return fmap
@@ -472,11 +506,11 @@ class VQGanVAE(nn.Module):
 
             fmap_discr_logits, img_discr_logits = map(self.discr, (fmap, img))
 
-            gp = gradient_penalty(img, img_discr_logits)
-
             discr_loss = self.discr_loss(fmap_discr_logits, img_discr_logits)
 
-            loss = discr_loss + gp
+            if add_gradient_penalty:
+                gp = gradient_penalty(img, img_discr_logits)
+                loss = discr_loss + gp
 
             if return_recons:
                 return loss, fmap

setup.py

@@ -10,7 +10,7 @@ setup(
       'dream = dalle2_pytorch.cli:dream'
     ],
   },
-  version = '0.0.36',
+  version = '0.0.51',
   license='MIT',
   description = 'DALL-E 2',
   author = 'Phil Wang',