just take care of the logic for AdamW and transformers

README.md

@@ -348,7 +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):
@@ -376,14 +377,224 @@ 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.
 
+## 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
+```
+
+## OpenAI CLIP
+
+Although there is the possibility they are using an unreleased, more powerful CLIP, you can use one of the released ones, if you do not wish to train your own CLIP from scratch. This will also allow the community to more quickly validate the conclusions of the paper.
+
+First you'll need to install <a href="https://github.com/openai/CLIP#usage">the prerequisites</a>
+
+Then to use a pretrained OpenAI CLIP, simply import `OpenAIClipAdapter` and pass it into the `DiffusionPrior` or `Decoder` like so
+
+```python
+import torch
+from dalle2_pytorch import DALLE2, DiffusionPriorNetwork, DiffusionPrior, Unet, Decoder, OpenAIClipAdapter
+
+# openai pretrained clip - defaults to ViT/B-32
+
+clip = OpenAIClipAdapter()
+
+# mock data
+
+text = torch.randint(0, 49408, (4, 256)).cuda()
+images = torch.randn(4, 3, 256, 256).cuda()
+
+# prior networks (with transformer)
+
+prior_network = DiffusionPriorNetwork(
+    dim = 512,
+    depth = 6,
+    dim_head = 64,
+    heads = 8
+).cuda()
+
+diffusion_prior = DiffusionPrior(
+    net = prior_network,
+    clip = clip,
+    timesteps = 100,
+    cond_drop_prob = 0.2
+).cuda()
+
+loss = diffusion_prior(text, images)
+loss.backward()
+
+# do above for many steps ...
+
+# decoder (with unet)
+
+unet1 = Unet(
+    dim = 128,
+    image_embed_dim = 512,
+    cond_dim = 128,
+    channels = 3,
+    dim_mults=(1, 2, 4, 8)
+).cuda()
+
+unet2 = Unet(
+    dim = 16,
+    image_embed_dim = 512,
+    cond_dim = 128,
+    channels = 3,
+    dim_mults = (1, 2, 4, 8, 16)
+).cuda()
+
+decoder = Decoder(
+    unet = (unet1, unet2),
+    image_sizes = (128, 256),
+    clip = clip,
+    timesteps = 100,
+    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):
+    loss = decoder(images, unet_number = unet_number) # this can optionally be decoder(images, text) if you wish to condition on the text encodings as well, though it was hinted in the paper it didn't do much
+    loss.backward()
+
+# do above for many steps
+
+dalle2 = DALLE2(
+    prior = diffusion_prior,
+    decoder = decoder
+)
+
+images = dalle2(
+    ['a butterfly trying to escape a tornado'],
+    cond_scale = 2. # classifier free guidance strength (> 1 would strengthen the condition)
+)
+
+# save your image (in this example, of size 256x256)
+```
+
+Now you'll just have to worry about training the Prior and the Decoder!
+
 ## 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.
+This repository decides to take the next step and offer DALL-E v2 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.
 
+The repository also comes equipped with all the necessary settings to recreate `ViT-VQGan` from the <a href="https://arxiv.org/abs/2110.04627">Improved VQGans</a> paper. Furthermore, the <a href="https://github.com/lucidrains/vector-quantize-pytorch">vector quantization</a> library also comes equipped to do <a href="https://arxiv.org/abs/2203.01941">residual or multi-headed quantization</a>, which I believe will give an even further boost in performance to the autoencoder.
+
 ```python
 import torch
 from dalle2_pytorch import Unet, Decoder, CLIP, VQGanVAE
@@ -524,14 +735,16 @@ Once built, images will be saved to the same directory the command is invoked
 - [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
-- [ ] spend one day cleaning up tech debt in decoder
+- [x] use inheritance just this once for sharing logic between decoder and prior network ddpms
+- [x] bring in vit-vqgan https://arxiv.org/abs/2110.04627 for the latent diffusion
+- [x] 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
+- [ ] think about how best to design a declarative training config that handles preencoding for prior and training of multiple networks in decoder
 - [ ] 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
 
@@ -577,4 +790,14 @@ Once built, images will be saved to the same directory the command is invoked
 }
 ```
 
+```bibtex
+@article{Yu2021VectorquantizedIM,
+    title   = {Vector-quantized Image Modeling with Improved VQGAN},
+    author  = {Jiahui Yu and Xin Li and Jing Yu Koh and Han Zhang and Ruoming Pang and James Qin and Alexander Ku and Yuanzhong Xu and Jason Baldridge and Yonghui Wu},
+    journal = {ArXiv},
+    year    = {2021},
+    volume  = {abs/2110.04627}
+}
+```
+
 *Creating noise from data is easy; creating data from noise is generative modeling.* - Yang Song's <a href="https://arxiv.org/abs/2011.13456">paper</a>

dalle2_pytorch/__init__.py

@@ -1,4 +1,5 @@
 from dalle2_pytorch.dalle2_pytorch import DALLE2, DiffusionPriorNetwork, DiffusionPrior, Unet, Decoder
+from dalle2_pytorch.dalle2_pytorch import OpenAIClipAdapter
 
 from dalle2_pytorch.vqgan_vae import VQGanVAE
 from x_clip import CLIP

dalle2_pytorch/attention.py

@@ -1,125 +0,0 @@
-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/dalle2_pytorch.py

@@ -7,6 +7,7 @@ from contextlib import contextmanager
 import torch
 import torch.nn.functional as F
 from torch import nn, einsum
+import torchvision.transforms as T
 
 from einops import rearrange, repeat
 from einops.layers.torch import Rearrange
@@ -17,7 +18,6 @@ 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
 
@@ -36,6 +36,10 @@ def default(val, d):
 def cast_tuple(val, length = 1):
     return val if isinstance(val, tuple) else ((val,) * length)
 
+@contextmanager
+def null_context(*args, **kwargs):
+    yield
+
 def eval_decorator(fn):
     def inner(model, *args, **kwargs):
         was_training = model.training
@@ -84,7 +88,146 @@ 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)
+
+# image normalization functions
+# ddpms expect images to be in the range of -1 to 1
+# but CLIP may otherwise
+
+def normalize_img(img):
+    return img * 2 - 1
+
+def unnormalize_img(normed_img):
+    return (normed_img + 1) * 0.5
+
+# clip related adapters
+
+class BaseClipAdapter(nn.Module):
+    def __init__(self, clip):
+        super().__init__()
+        self.clip = clip
+
+    @property
+    def dim_latent(self):
+        raise NotImplementedError
+
+    @property
+    def image_size(self):
+        raise NotImplementedError
+
+    @property
+    def image_channels(self):
+        raise NotImplementedError
+
+    @property
+    def max_text_len(self):
+        raise NotImplementedError
+
+    def embed_text(self, text):
+        raise NotImplementedError
+
+    def embed_image(self, image):
+        raise NotImplementedError
+
+class XClipAdapter(BaseClipAdapter):
+    @property
+    def dim_latent(self):
+        return self.clip.dim_latent
+
+    @property
+    def image_size(self):
+        return self.clip.image_size
+
+    @property
+    def image_channels(self):
+        return self.clip.image_channels
+
+    @property
+    def max_text_len(self):
+        return self.clip.text_seq_len
+
+    @torch.no_grad()
+    def embed_text(self, text):
+        text = text[..., :self.max_text_len]
+        text_mask = text != 0
+        encoder_output = self.clip.text_transformer(text)
+        text_cls, text_encodings = encoder_output[:, 0], encoder_output[:, 1:]
+        text_embed = self.clip.to_text_latent(text_cls)
+        return l2norm(text_embed), text_encodings, text_mask
+
+    @torch.no_grad()
+    def embed_image(self, image):
+        image = resize_image_to(image, self.image_size)
+        encoder_output = self.clip.visual_transformer(image)
+        image_cls, image_encodings = encoder_output[:, 0], encoder_output[:, 1:]
+        image_embed = self.clip.to_visual_latent(image_cls)
+        return l2norm(image_embed), image_encodings
+
+class OpenAIClipAdapter(BaseClipAdapter):
+    def __init__(
+        self,
+        name = 'ViT-B/32'
+    ):
+        try:
+            import clip
+        except ImportError:
+            print('you must install openai clip in order to use this adapter - `pip install git+https://github.com/openai/CLIP.git` - more instructions at https://github.com/openai/CLIP#usage')
+
+        openai_clip, _ = clip.load(name)
+        super().__init__(openai_clip)
+
+        text_attention_final = self.find_layer('ln_final')
+        self.handle = text_attention_final.register_forward_hook(self._hook)
+        self.clip_normalize = T.Normalize((0.48145466, 0.4578275, 0.40821073), (0.26862954, 0.26130258, 0.27577711))
+        self.cleared = False
+
+    def find_layer(self,  layer):
+        modules = dict([*self.clip.named_modules()])
+        return modules.get(layer, None)
+
+    def clear(self):
+        if self.cleared:
+            return
+
+        self.handle()
+
+    def _hook(self, _, inputs, outputs):
+        self.text_encodings = outputs
+
+    @property
+    def dim_latent(self):
+        return 512
+
+    @property
+    def image_size(self):
+        return self.clip.visual.input_resolution
+
+    @property
+    def image_channels(self):
+        return 3
+
+    @property
+    def max_text_len(self):
+        return self.clip.context_length
+
+    @torch.no_grad()
+    def embed_text(self, text):
+        text = text[..., :self.max_text_len]
+        text_mask = text != 0
+        assert not self.cleared
+
+        text_embed = self.clip.encode_text(text)
+        text_encodings = self.text_encodings
+        del self.text_encodings
+        return text_embed.float(), text_encodings.float(), text_mask
+
+    @torch.no_grad()
+    def embed_image(self, image):
+        assert not self.cleared
+        image = resize_image_to(image, self.image_size)
+        image = self.clip_normalize(unnormalize_img(image))
+        image_embed = self.clip.encode_image(image)
+        return image_embed.float(), None
 
 # classifier free guidance functions
 
@@ -143,6 +286,103 @@ 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)
+
+        if loss_type == 'l1':
+            loss_fn = F.l1_loss
+        elif loss_type == 'l2':
+            loss_fn = F.mse_loss
+        elif loss_type == 'huber':
+            loss_fn = F.smooth_l1_loss
+        else:
+            raise NotImplementedError()
+
+        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)
+
+        # 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):
@@ -421,25 +661,41 @@ 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
 
@@ -455,16 +711,6 @@ 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)
@@ -475,115 +721,49 @@ 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_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):
+            if isinstance(clip, CLIP):
+                clip = XClipAdapter(clip)
+
+            assert isinstance(clip, BaseClipAdapter)
+            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.image_embed_dim = default(image_embed_dim, lambda: clip.dim_latent)
+        self.channels = default(image_channels, lambda: clip.image_channels)
+
         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):
-        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):
-        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)
-        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):
         pred = self.net(x, t, **text_cond)
 
@@ -620,37 +800,21 @@ 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):
+    def p_losses(self, image_embed, times, text_cond, noise = None):
         noise = default(noise, lambda: torch.randn_like(image_embed))
 
-        image_embed_noisy = self.q_sample(x_start = image_embed, t = t, noise = noise)
+        image_embed_noisy = self.q_sample(x_start = image_embed, t = times, noise = noise)
 
-        x_recon = self.net(
+        pred = self.net(
             image_embed_noisy,
-            t,
+            times,
             cond_drop_prob = self.cond_drop_prob,
             **text_cond
         )
 
-        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)
-        elif self.loss_type == 'l2':
-            loss = F.mse_loss(to_predict, x_recon)
-        elif self.loss_type == "huber":
-            loss = F.smooth_l1_loss(to_predict, x_recon)
-        else:
-            raise NotImplementedError()
+        target = noise if not self.predict_x_start else image_embed
 
+        loss = self.loss_fn(pred, target)
         return loss
 
     @torch.no_grad()
@@ -663,7 +827,12 @@ class DiffusionPrior(nn.Module):
         batch_size = text.shape[0]
         image_embed_dim = self.image_embed_dim
 
-        text_cond = self.get_text_cond(text)
+        text_embed, text_encodings, text_mask = self.clip.embed_text(text)
+
+        text_cond = dict(text_embed = text_embed)
+
+        if self.condition_on_text_encodings:
+            text_cond = {**text_cond, 'text_encodings': text_encodings, 'mask': text_mask}
 
         image_embeds = self.p_sample_loop((batch_size, image_embed_dim), text_cond = text_cond)
         text_embeds = text_cond['text_embed']
@@ -679,16 +848,43 @@ 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.clip.embed_image(image)
 
-        loss = self.p_losses(image_embed, times, text_cond = text_cond, *args, **kwargs)
-        return loss
+        # calculate text conditionings, based on what is passed in
+
+        if exists(text):
+            text_embed, text_encodings, text_mask = self.clip.embed_text(text)
+
+        text_cond = dict(text_embed = text_embed)
+
+        if self.condition_on_text_encodings:
+            assert exists(text_encodings), 'text encodings must be present for diffusion prior if specified'
+            text_cond = {**text_cond, '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
+
+        return self.p_losses(image_embed, times, text_cond = text_cond, *args, **kwargs)
 
 # decoder
 
@@ -840,6 +1036,7 @@ class Unet(nn.Module):
         dim,
         *,
         image_embed_dim,
+        text_embed_dim = None,
         cond_dim = None,
         num_image_tokens = 4,
         num_time_tokens = 2,
@@ -853,6 +1050,7 @@ 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__()
@@ -892,7 +1090,7 @@ 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.text_to_cond = nn.LazyLinear(cond_dim)
+        self.text_to_cond = nn.LazyLinear(cond_dim) if not exists(text_embed_dim) else nn.Linear(text_embed_dim, cond_dim)
 
         # finer control over whether to condition on image embeddings and text encodings
         # so one can have the latter unets in the cascading DDPMs only focus on super-resoluting
@@ -903,7 +1101,7 @@ 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
 
@@ -956,13 +1154,14 @@ class Unet(nn.Module):
         self,
         *,
         lowres_cond,
-        channels
+        channels,
+        cond_on_image_embeds
     ):
-        if lowres_cond == self.lowres_cond and channels == self.channels:
+        if lowres_cond == self.lowres_cond and channels == self.channels and cond_on_image_embeds == self.cond_on_image_embeds:
             return self
 
-        updated_kwargs = {**self._locals, 'lowres_cond': lowres_cond, 'channels': channels}
-        return self.__class__(**updated_kwargs)
+        updated_kwargs = {'lowres_cond': lowres_cond, 'channels': channels, 'cond_on_image_embeds': cond_on_image_embeds}
+        return self.__class__(**{**self._locals, **updated_kwargs})
 
     def forward_with_cond_scale(
         self,
@@ -1031,7 +1230,7 @@ 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)
@@ -1099,7 +1298,7 @@ class LowresConditioner(nn.Module):
         target_image_size = cast_tuple(target_image_size, 2)
 
         if self.training and self.downsample_first and exists(downsample_image_size):
-            cond_fmap = resize_image_to(cond_fmap, target_image_size, mode = self.cond_upsample_mode)
+            cond_fmap = resize_image_to(cond_fmap, downsample_image_size, mode = self.cond_upsample_mode)
 
         if self.training:
             # when training, blur the low resolution conditional image
@@ -1111,7 +1310,7 @@ class LowresConditioner(nn.Module):
 
         return cond_fmap
 
-class Decoder(nn.Module):
+class Decoder(BaseGaussianDiffusion):
     def __init__(
         self,
         unet,
@@ -1129,14 +1328,26 @@ class Decoder(nn.Module):
         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__()
-        assert isinstance(clip, CLIP)
+        super().__init__(
+            beta_schedule = beta_schedule,
+            timesteps = timesteps,
+            loss_type = loss_type
+        )
+
+        if isinstance(clip, CLIP):
+            clip = XClipAdapter(clip)
+
         freeze_model_and_make_eval_(clip)
+        assert isinstance(clip, BaseClipAdapter)
+
         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
 
@@ -1157,6 +1368,7 @@ class Decoder(nn.Module):
 
             one_unet = one_unet.cast_model_parameters(
                 lowres_cond = not is_first,
+                cond_on_image_embeds = is_first,
                 channels = unet_channels
             )
 
@@ -1192,55 +1404,6 @@ 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
@@ -1260,39 +1423,11 @@ 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)
-        image_cls = image_encoding[:, 0]
-        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 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
+        image_embed, _ = self.clip.embed_image(image)
+        return image_embed
 
     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)
@@ -1338,22 +1473,14 @@ class Decoder(nn.Module):
 
         return img
 
-    def q_sample(self, x_start, t, noise=None):
+    def p_losses(self, unet, x_start, times, *, image_embed, lowres_cond_img = None, text_encodings = None, predict_x_start = False, 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
-        )
+        x_noisy = self.q_sample(x_start = x_start, t = times, noise = noise)
 
-    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)
-
-        x_recon = unet(
+        pred = unet(
             x_noisy,
-            t,
+            times,
             image_embed = image_embed,
             text_encodings = text_encodings,
             lowres_cond_img = lowres_cond_img,
@@ -1362,15 +1489,7 @@ class Decoder(nn.Module):
 
         target = noise if not predict_x_start else x_start
 
-        if self.loss_type == 'l1':
-            loss = F.l1_loss(target, x_recon)
-        elif self.loss_type == 'l2':
-            loss = F.mse_loss(target, x_recon)
-        elif self.loss_type == "huber":
-            loss = F.smooth_l1_loss(target, x_recon)
-        else:
-            raise NotImplementedError()
-
+        loss = self.loss_fn(pred, target)
         return loss
 
     @torch.no_grad()
@@ -1378,12 +1497,20 @@ class Decoder(nn.Module):
     def sample(self, image_embed, text = None, cond_scale = 1.):
         batch_size = image_embed.shape[0]
 
-        text_encodings = self.get_text_encodings(text) if exists(text) else None
+        text_encodings = None
+        if exists(text):
+            _, text_encodings, _ = self.clip.embed_text(text)
+
+        assert not (self.condition_on_text_encodings and not exists(text_encodings)), 'text or text encodings must be passed into decoder if specified'
+        assert not (not self.condition_on_text_encodings and exists(text_encodings)), 'decoder specified not to be conditioned on text, yet it is presented'
 
         img = None
 
         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):
+
+            context = self.one_unet_in_gpu(unet = unet) if image_embed.is_cuda else null_context()
+
+            with context:
                 lowres_cond_img = None
                 shape = (batch_size, channel, image_size, image_size)
 
@@ -1436,9 +1563,14 @@ class Decoder(nn.Module):
         times = torch.randint(0, self.num_timesteps, (b,), device = device, dtype = torch.long)
 
         if not exists(image_embed):
-            image_embed = self.get_image_embed(image)
+            image_embed, _ = self.clip.embed_image(image)
 
-        text_encodings = self.get_text_encodings(text) if exists(text) and not exists(text_encodings) else None
+        text_encodings = None
+        if exists(text) and not exists(text_encodings):
+            _, text_encodings, _ = self.clip.embed_text(text)
+
+        assert not (self.condition_on_text_encodings and not exists(text_encodings)), 'text or text encodings must be passed into decoder if specified'
+        assert not (not self.condition_on_text_encodings and exists(text_encodings)), 'decoder specified not to be conditioned on text, yet it is presented'
 
         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
         image = resize_image_to(image, target_image_size)
@@ -1467,14 +1599,19 @@ 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
+
+        self.to_pil = T.ToPILImage()
 
     @torch.no_grad()
     @eval_decorator
     def forward(
         self,
         text,
-        cond_scale = 1.
+        cond_scale = 1.,
+        return_pil_images = False
     ):
         device = next(self.parameters()).device
         one_text = isinstance(text, str) or (not is_list_str(text) and text.shape[0] == 1)
@@ -1484,9 +1621,15 @@ class DALLE2(nn.Module):
             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 return_pil_images:
+            images = list(map(self.to_pil, images.unbind(dim = 0)))
 
         if one_text:
             return images[0]
 
         return images
+

dalle2_pytorch/openai_clip.py

@@ -0,0 +1,84 @@
+import torch
+from PIL import Image
+
+from dalle2_pytorch.dalle2_pytorch import BaseClipAdapter
+import torchvision.transforms as T
+
+def find_layer(model, layer):
+    modules = dict([*model.named_modules()])
+    return modules.get(layer, None)
+
+def hook(_, input, output):
+    print(output.shape)
+
+import clip
+# image = preprocess(Image.open("CLIP.png")).unsqueeze(0).to(device)
+text = clip.tokenize(["a diagram", "a dog", "a cat"]).cuda()
+image = torch.randn(1, 3, 224, 224).cuda()
+
+
+class OpenAIClipAdapter(BaseClipAdapter):
+    def __init__(self, name = 'ViT-B/32'):
+        try:
+            import clip
+        except ImportError:
+            print('you must install openai clip in order to use this adapter - `pip install git+https://github.com/openai/CLIP.git` - more instructions at https://github.com/openai/CLIP#usage')
+
+        openai_clip, _ = clip.load(name)
+        super().__init__(openai_clip)
+
+        text_attention_final = self.find_layer(self.clip, 'ln_final')
+        self.handle = text_attention_final.register_forward_hook(self._hook)
+        self.clip_normalize = T.Normalize((0.48145466, 0.4578275, 0.40821073), (0.26862954, 0.26130258, 0.27577711))
+        self.cleared = False
+
+    def find_layer(self,  layer):
+        modules = dict([*self.clip.named_modules()])
+        return modules.get(layer, None)
+
+    def clear(self):
+        if self.cleared:
+            return
+
+        self.handle()
+
+    def _hook(self, _, inputs, outputs):
+        self.text_encodings = outputs
+
+    @property
+    def dim_latent(self):
+        return 512
+
+    @property
+    def image_size(self):
+        return self.clip.visual.input_resolution
+
+    @property
+    def image_channels(self):
+        return 3
+
+    @torch.no_grad()
+    def embed_text(self, text):
+        assert not self.cleared
+
+        text_embed = self.clip.encode_text(text)
+        text_encodings = self.text_encodings
+        del self.text_encodings
+        return text_embed, text_encodings
+
+    @torch.no_grad()
+    def embed_image(self, image):
+        assert not self.cleared
+
+        image = self.clip_normalize(image)
+        image_embed = self.clip.encode_image(image)
+        return image_embed, None
+
+clip_adapter = OpenAIClipAdapter().cuda()
+
+# print(model)
+with torch.no_grad():
+    image_features, _ = clip_adapter.embed_image(image)
+    text_features, text_encodings = clip_adapter.embed_text(text)
+    print(text_features.shape, image_features.shape)
+    print(text_encodings.shape)

dalle2_pytorch/optimizer.py

@@ -0,0 +1,29 @@
+from torch.optim import AdamW, Adam
+
+def separate_weight_decayable_params(params):
+    no_wd_params = set([param for param in params if param.ndim < 2])
+    wd_params = set(params) - no_wd_params
+    return wd_params, no_wd_params
+
+def get_optimizer(
+    params,
+    lr = 3e-4,
+    wd = 1e-2,
+    betas = (0.9, 0.999),
+    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)
+
+    params = set(params)
+    wd_params, no_wd_params = separate_weight_decayable_params(params)
+
+    param_groups = [
+        {'params': list(wd_params)},
+        {'params': list(no_wd_params), 'weight_decay': 0},
+    ]
+
+    return AdamW(param_groups, lr = lr, weight_decay = wd, betas = betas)

dalle2_pytorch/train.py

@@ -35,7 +35,7 @@ class EMA(nn.Module):
 
         self.update_moving_average(self.ema_model, self.online_model)
 
-    def update_moving_average(ma_model, current_model):
+    def update_moving_average(self, ma_model, current_model):
         def calculate_ema(beta, old, new):
             if not exists(old):
                 return new

dalle2_pytorch/vqgan_vae.py

@@ -12,8 +12,8 @@ from torch.autograd import grad as torch_grad
 import torchvision
 
 from einops import rearrange, reduce, repeat
-
-from dalle2_pytorch.attention import QueryAttnUpsample
+from einops_exts import rearrange_many
+from einops.layers.torch import Rearrange
 
 # constants
 
@@ -146,6 +146,8 @@ class LayerNormChan(nn.Module):
         mean = torch.mean(x, dim = 1, keepdim = True)
         return (x - mean) / (var + self.eps).sqrt() * self.gamma
 
+# discriminator
+
 class Discriminator(nn.Module):
     def __init__(
         self,
@@ -179,6 +181,8 @@ class Discriminator(nn.Module):
 
         return self.to_logits(x)
 
+# positional encoding
+
 class ContinuousPositionBias(nn.Module):
     """ from https://arxiv.org/abs/2111.09883 """
 
@@ -213,6 +217,84 @@ class ContinuousPositionBias(nn.Module):
         bias = rearrange(rel_pos, 'i j h -> h i j')
         return x + bias
 
+# resnet encoder / decoder
+
+class ResnetEncDec(nn.Module):
+    def __init__(
+        self,
+        dim,
+        *,
+        channels = 3,
+        layers = 4,
+        layer_mults = None,
+        num_resnet_blocks = 1,
+        resnet_groups = 16,
+        first_conv_kernel_size = 5,
+        use_attn = True,
+        attn_dim_head = 64,
+        attn_heads = 8,
+        attn_dropout = 0.,
+    ):
+        super().__init__()
+        assert dim % resnet_groups == 0, f'dimension {dim} must be divisible by {resnet_groups} (groups for the groupnorm)'
+
+        self.layers = layers
+
+        self.encoders = MList([])
+        self.decoders = MList([])
+
+        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)
+
+        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,) * (layers - 1)), num_resnet_blocks)
+
+        if not isinstance(use_attn, tuple):
+            use_attn = (*((False,) * (layers - 1)), use_attn)
+
+        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(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.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))
+
+            for _ in range(layer_num_resnet_blocks):
+                append(self.encoders, ResBlock(dim_out, groups = resnet_groups))
+                prepend(self.decoders, GLUResBlock(dim_out, groups = resnet_groups))
+
+            if layer_use_attn:
+                append(self.encoders, VQGanAttention(dim = dim_out, heads = attn_heads, dim_head = attn_dim_head, dropout = attn_dropout))
+
+        prepend(self.encoders, nn.Conv2d(channels, dim, first_conv_kernel_size, padding = first_conv_kernel_size // 2))
+        append(self.decoders, nn.Conv2d(dim, channels, 1))
+
+    def get_encoded_fmap_size(self, image_size):
+        return image_size // (2 ** self.layers)
+
+    def encode(self, x):
+        for enc in self.encoders:
+            x = enc(x)
+        return x
+
+    def decode(self, x):
+        for dec in self.decoders:
+            x = dec(x)
+        return x
+
 class GLUResBlock(nn.Module):
     def __init__(self, chan, groups = 16):
         super().__init__()
@@ -246,6 +328,7 @@ class ResBlock(nn.Module):
         return self.net(x) + x
 
 # vqgan attention layer
+
 class VQGanAttention(nn.Module):
     def __init__(
         self,
@@ -290,6 +373,145 @@ class VQGanAttention(nn.Module):
 
         return out + residual
 
+# ViT encoder / decoder
+
+class RearrangeImage(nn.Module):
+    def forward(self, x):
+        n = x.shape[1]
+        w = h = int(sqrt(n))
+        return rearrange(x, 'b (h w) ... -> b h w ...', h = h, w = w)
+
+class Attention(nn.Module):
+    def __init__(
+        self,
+        dim,
+        *,
+        heads = 8,
+        dim_head = 32
+    ):
+        super().__init__()
+        self.norm = nn.LayerNorm(dim)
+        self.heads = heads
+        self.scale = dim_head ** -0.5
+        inner_dim = dim_head * heads
+
+        self.to_qkv = nn.Linear(dim, inner_dim * 3, bias = False)
+        self.to_out = nn.Linear(inner_dim, dim)
+
+    def forward(self, x):
+        h = self.heads
+
+        x = self.norm(x)
+
+        q, k, v = self.to_qkv(x).chunk(3, dim = -1)
+        q, k, v = rearrange_many((q, k, v), 'b n (h d) -> b h n d', h = h)
+
+        q = q * self.scale
+        sim = einsum('b h i d, b h j d -> b h i j', q, k)
+
+        sim = sim - sim.amax(dim = -1, keepdim = True).detach()
+        attn = sim.softmax(dim = -1)
+
+        out = einsum('b h i j, b h j d -> b h i d', attn, v)
+
+        out = rearrange(out, 'b h n d -> b n (h d)')
+        return self.to_out(out)
+
+def FeedForward(dim, mult = 4):
+    return nn.Sequential(
+        nn.LayerNorm(dim),
+        nn.Linear(dim, dim * mult, bias = False),
+        nn.GELU(),
+        nn.Linear(dim * mult, dim, bias = False)
+    )
+
+class Transformer(nn.Module):
+    def __init__(
+        self,
+        dim,
+        *,
+        layers,
+        dim_head = 32,
+        heads = 8,
+        ff_mult = 4
+    ):
+        super().__init__()
+        self.layers = nn.ModuleList([])
+        for _ in range(layers):
+            self.layers.append(nn.ModuleList([
+                Attention(dim = dim, dim_head = dim_head, heads = heads),
+                FeedForward(dim = dim, mult = ff_mult)
+            ]))
+
+        self.norm = nn.LayerNorm(dim)
+
+    def forward(self, x):
+        for attn, ff in self.layers:
+            x = attn(x) + x
+            x = ff(x) + x
+
+        return self.norm(x)
+
+class ViTEncDec(nn.Module):
+    def __init__(
+        self,
+        dim,
+        channels = 3,
+        layers = 4,
+        patch_size = 8,
+        dim_head = 32,
+        heads = 8,
+        ff_mult = 4
+    ):
+        super().__init__()
+        self.encoded_dim = dim
+        self.patch_size = patch_size
+
+        input_dim = channels * (patch_size ** 2)
+
+        self.encoder = nn.Sequential(
+            Rearrange('b c (h p1) (w p2) -> b (h w) (p1 p2 c)', p1 = patch_size, p2 = patch_size),
+            nn.Linear(input_dim, dim),
+            Transformer(
+                dim = dim,
+                dim_head = dim_head,
+                heads = heads,
+                ff_mult = ff_mult,
+                layers = layers
+            ),
+            RearrangeImage(),
+            Rearrange('b h w c -> b c h w')
+        )
+
+        self.decoder = nn.Sequential(
+            Rearrange('b c h w -> b (h w) c'),
+            Transformer(
+                dim = dim,
+                dim_head = dim_head,
+                heads = heads,
+                ff_mult = ff_mult,
+                layers = layers
+            ),
+            nn.Sequential(
+                nn.Linear(dim, dim * 4, bias = False),
+                nn.Tanh(),
+                nn.Linear(dim * 4, input_dim, bias = False),
+            ),
+            RearrangeImage(),
+            Rearrange('b h w (p1 p2 c) -> b c (h p1) (w p2)', p1 = patch_size, p2 = patch_size)
+        )
+
+    def get_encoded_fmap_size(self, image_size):
+        return image_size // self.patch_size
+
+    def encode(self, x):
+        return self.encoder(x)
+
+    def decode(self, x):
+        return self.decoder(x)
+
+# main vqgan-vae classes
+
 class NullVQGanVAE(nn.Module):
     def __init__(
         self,
@@ -320,81 +542,45 @@ class VQGanVAE(nn.Module):
         image_size,
         channels = 3,
         layers = 4,
-        layer_mults = None,
         l2_recon_loss = False,
         use_hinge_loss = True,
-        num_resnet_blocks = 1,
         vgg = None,
+        vq_codebook_dim = 256,
         vq_codebook_size = 512,
         vq_decay = 0.8,
         vq_commitment_weight = 1.,
         vq_kmeans_init = True,
         vq_use_cosine_sim = True,
-        use_attn = True,
-        attn_dim_head = 64,
-        attn_heads = 8,
-        resnet_groups = 16,
-        attn_dropout = 0.,
-        first_conv_kernel_size = 5,
         use_vgg_and_gan = True,
+        vae_type = 'resnet',
+        discr_layers = 4,
         **kwargs
     ):
         super().__init__()
-        assert dim % resnet_groups == 0, f'dimension {dim} must be divisible by {resnet_groups} (groups for the groupnorm)'
-
         vq_kwargs, kwargs = groupby_prefix_and_trim('vq_', kwargs)
+        encdec_kwargs, kwargs = groupby_prefix_and_trim('encdec_', kwargs)
 
         self.image_size = image_size
         self.channels = channels
-        self.layers = layers
-        self.fmap_size = image_size // (layers ** 2)
         self.codebook_size = vq_codebook_size
 
-        self.encoders = MList([])
-        self.decoders = MList([])
+        if vae_type == 'resnet':
+            enc_dec_klass = ResnetEncDec
+        elif vae_type == 'vit':
+            enc_dec_klass = ViTEncDec
+        else:
+            raise ValueError(f'{vae_type} not valid')
 
-        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,) * (layers - 1)), num_resnet_blocks)
-
-        if not isinstance(use_attn, tuple):
-            use_attn = (*((False,) * (layers - 1)), use_attn)
-
-        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(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.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))
-
-            for _ in range(layer_num_resnet_blocks):
-                append(self.encoders, ResBlock(dim_out, groups = resnet_groups))
-                prepend(self.decoders, GLUResBlock(dim_out, groups = resnet_groups))
-
-            if layer_use_attn:
-                append(self.encoders, VQGanAttention(dim = dim_out, heads = attn_heads, dim_head = attn_dim_head, dropout = attn_dropout))
-
-        prepend(self.encoders, nn.Conv2d(channels, dim, first_conv_kernel_size, padding = first_conv_kernel_size // 2))
-        append(self.decoders, nn.Conv2d(dim, channels, 1))
+        self.enc_dec = enc_dec_klass(
+            dim = dim,
+            channels = channels,
+            layers = layers,
+            **encdec_kwargs
+        )
 
         self.vq = VQ(
-            dim = codebook_dim,
+            dim = self.enc_dec.encoded_dim,
+            codebook_dim = vq_codebook_dim,
             codebook_size = vq_codebook_size,
             decay = vq_decay,
             commitment_weight = vq_commitment_weight,
@@ -427,13 +613,21 @@ class VQGanVAE(nn.Module):
 
         # gan related losses
 
+        layer_mults = list(map(lambda t: 2 ** t, range(discr_layers)))
+        layer_dims = [dim * mult for mult in layer_mults]
+        dims = (dim, *layer_dims)
+
         self.discr = Discriminator(dims = dims, channels = channels)
 
         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
 
+    @property
+    def encoded_dim(self):
+        return self.enc_dec.encoded_dim
+
     def get_encoded_fmap_size(self, image_size):
-        return image_size // (2 ** self.layers)
+        return self.enc_dec.get_encoded_fmap_size(image_size)
 
     def copy_for_eval(self):
         device = next(self.parameters()).device
@@ -459,16 +653,13 @@ class VQGanVAE(nn.Module):
         return self.vq.codebook
 
     def encode(self, fmap):
-        for enc in self.encoders:
-            fmap = enc(fmap)
-
+        fmap = self.enc_dec.encode(fmap)
         return 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)
+        fmap = self.enc_dec.decode(fmap)
 
         if not return_indices_and_loss:
             return fmap

setup.py

@@ -10,7 +10,7 @@ setup(
       'dream = dalle2_pytorch.cli:dream'
     ],
   },
-  version = '0.0.46',
+  version = '0.0.70',
   license='MIT',
   description = 'DALL-E 2',
   author = 'Phil Wang',
@@ -31,7 +31,7 @@ setup(
     'torchvision',
     'tqdm',
     'vector-quantize-pytorch',
-    'x-clip>=0.4.4',
+    'x-clip>=0.5.1',
     'youtokentome'
   ],
   classifiers=[