simulate unrelated captions as a training metric (#66)

* add unrelated embedding metric

* change to torch.roll

Co-authored-by: nousr <z@localhost.com>
Co-authored-by: nousr <>

README.md

@@ -902,7 +902,7 @@ Please note that the script internally passes text_embed and image_embed to the
 ### Usage 
 
 ```bash
-$ pyhon train_diffusion_prior.py
+$ python train_diffusion_prior.py
 ```
 
 The most significant parameters for the script are as follows:
@@ -967,7 +967,7 @@ Once built, images will be saved to the same directory the command is invoked
 - [x] make sure DDPMs can be run with traditional resnet blocks (but leave convnext as an option for experimentation)
 - [x] make sure for the latter unets in the cascade, one can train on crops for learning super resolution (constrain the unet to be only convolutions in that case, or allow conv-like attention with rel pos bias)
 - [ ] become an expert with unets, cleanup unet code, make it fully configurable, port all learnings over to https://github.com/lucidrains/x-unet (test out unet² in ddpm repo) - consider https://github.com/lucidrains/uformer-pytorch attention-based unet
-- [ ] 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
+- [ ] make sure the cascading ddpm in the repository can be trained unconditionally, offer a one-line CLI tool for training on a folder of images
 - [ ] transcribe code to Jax, which lowers the activation energy for distributed training, given access to TPUs
 - [ ] pull logic for training diffusion prior into a class DiffusionPriorTrainer, for eventual script based + CLI based training
 - [ ] train on a toy task, offer in colab
@@ -980,6 +980,7 @@ Once built, images will be saved to the same directory the command is invoked
 - [ ] interface out the vqgan-vae so a pretrained one can be pulled off the shelf to validate latent diffusion + DALL-E2
 - [ ] make sure FILIP works with DALL-E2 from x-clip https://arxiv.org/abs/2111.07783
 - [ ] make sure resnet hyperparameters can be configurable across unet depth (groups and expansion factor)
+- [ ] offer save / load methods on the trainer classes to automatically take care of state dicts for scalers / optimizers / saving versions and checking for breaking changes
 
 ## Citations
 
@@ -1047,4 +1048,14 @@ Once built, images will be saved to the same directory the command is invoked
 }
 ```
 
+```bibtex
+@article{Yu2022CoCaCC,
+    title   = {CoCa: Contrastive Captioners are Image-Text Foundation Models},
+    author  = {Jiahui Yu and Zirui Wang and Vijay Vasudevan and Legg Yeung and Mojtaba Seyedhosseini and Yonghui Wu},
+    journal = {ArXiv},
+    year    = {2022},
+    volume  = {abs/2205.01917}
+}
+```
+
 *Creating noise from data is easy; creating data from noise is generative modeling.* - <a href="https://arxiv.org/abs/2011.13456">Yang Song's paper</a>

dalle2_pytorch/dalle2_pytorch.py

@@ -23,9 +23,14 @@ from dalle2_pytorch.vqgan_vae import NullVQGanVAE, VQGanVAE
 
 from resize_right import resize
 
+# rotary embeddings
+
+from rotary_embedding_torch import RotaryEmbedding
+
 # use x-clip
 
 from x_clip import CLIP
+from coca_pytorch import CoCa
 
 # helper functions
 
@@ -113,9 +118,10 @@ EmbeddedText = namedtuple('EmbedTextReturn', ['text_embed', 'text_encodings', 't
 EmbeddedImage = namedtuple('EmbedImageReturn', ['image_embed', 'image_encodings'])
 
 class BaseClipAdapter(nn.Module):
-    def __init__(self, clip):
+    def __init__(self, clip, **kwargs):
         super().__init__()
         self.clip = clip
+        self.overrides = kwargs
 
     @property
     def dim_latent(self):
@@ -173,6 +179,39 @@ class XClipAdapter(BaseClipAdapter):
         image_embed = self.clip.to_visual_latent(image_cls)
         return EmbeddedImage(l2norm(image_embed), image_encodings)
 
+class CoCaAdapter(BaseClipAdapter):
+    @property
+    def dim_latent(self):
+        return self.clip.dim
+
+    @property
+    def image_size(self):
+        assert 'image_size' in self.overrides
+        return self.overrides['image_size']
+
+    @property
+    def image_channels(self):
+        assert 'image_channels' in self.overrides
+        return self.overrides['image_channels']
+
+    @property
+    def max_text_len(self):
+        assert 'max_text_len' in self.overrides
+        return self.overrides['max_text_len']
+
+    @torch.no_grad()
+    def embed_text(self, text):
+        text = text[..., :self.max_text_len]
+        text_mask = text != 0
+        text_embed, text_encodings = self.clip.embed_text(text)
+        return EmbeddedText(text_embed, text_encodings, text_mask)
+
+    @torch.no_grad()
+    def embed_image(self, image):
+        image = resize_image_to(image, self.image_size)
+        image_embed, image_encodings = self.clip.embed_image(image)
+        return EmbeddedImage(image_embed, image_encodings)
+
 class OpenAIClipAdapter(BaseClipAdapter):
     def __init__(
         self,
@@ -225,7 +264,7 @@ class OpenAIClipAdapter(BaseClipAdapter):
         text_embed = self.clip.encode_text(text)
         text_encodings = self.text_encodings
         del self.text_encodings
-        return EmbeddedText(text_embed.float(), text_encodings.float(), text_mask)
+        return EmbeddedText(l2norm(text_embed.float()), text_encodings.float(), text_mask)
 
     @torch.no_grad()
     def embed_image(self, image):
@@ -233,7 +272,7 @@ class OpenAIClipAdapter(BaseClipAdapter):
         image = resize_image_to(image, self.image_size)
         image = self.clip_normalize(unnormalize_img(image))
         image_embed = self.clip.encode_image(image)
-        return EmbeddedImage(image_embed.float(), None)
+        return EmbeddedImage(l2norm(image_embed.float()), None)
 
 # classifier free guidance functions
 
@@ -531,7 +570,8 @@ class Attention(nn.Module):
         heads = 8,
         dropout = 0.,
         causal = False,
-        post_norm = False
+        post_norm = False,
+        rotary_emb = None
     ):
         super().__init__()
         self.scale = dim_head ** -0.5
@@ -547,6 +587,8 @@ class Attention(nn.Module):
         self.to_q = nn.Linear(dim, inner_dim, bias = False)
         self.to_kv = nn.Linear(dim, dim_head * 2, bias = False)
 
+        self.rotary_emb = rotary_emb
+
         self.to_out = nn.Sequential(
             nn.Linear(inner_dim, dim, bias = False),
             LayerNorm(dim) if post_norm else nn.Identity()
@@ -559,6 +601,12 @@ class Attention(nn.Module):
         q, k, v = (self.to_q(x), *self.to_kv(x).chunk(2, dim = -1))
 
         q = rearrange(q, 'b n (h d) -> b h n d', h = self.heads)
+        q = q * self.scale
+
+        # rotary embeddings
+
+        if exists(self.rotary_emb):
+            q, k = map(self.rotary_emb.rotate_queries_or_keys, (q, k))
 
         # add null key / value for classifier free guidance in prior net
 
@@ -566,7 +614,7 @@ class Attention(nn.Module):
         k = torch.cat((nk, k), dim = -2)
         v = torch.cat((nv, v), dim = -2)
 
-        q = q * self.scale
+        # calculate query / key similarities
 
         sim = einsum('b h i d, b j d -> b h i j', q, k)
 
@@ -616,15 +664,18 @@ class CausalTransformer(nn.Module):
         attn_dropout = 0.,
         ff_dropout = 0.,
         final_proj = True,
-        normformer = False
+        normformer = False,
+        rotary_emb = True
     ):
         super().__init__()
         self.rel_pos_bias = RelPosBias(heads = heads)
 
+        rotary_emb = RotaryEmbedding(dim = min(32, dim_head)) if rotary_emb else None
+
         self.layers = nn.ModuleList([])
         for _ in range(depth):
             self.layers.append(nn.ModuleList([
-                Attention(dim = dim, causal = True, dim_head = dim_head, heads = heads, dropout = attn_dropout, post_norm = normformer),
+                Attention(dim = dim, causal = True, dim_head = dim_head, heads = heads, dropout = attn_dropout, post_norm = normformer, rotary_emb = rotary_emb),
                 FeedForward(dim = dim, mult = ff_mult, dropout = ff_dropout, post_activation_norm = normformer)
             ]))
 
@@ -714,7 +765,7 @@ class DiffusionPriorNetwork(nn.Module):
         # but let's just do it right
 
         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
+            mask = F.pad(mask, (0, 3), value = True) # extend mask for text embedding, noised image embedding, time step embedding, and learned query
 
         time_embed = self.time_embeddings(diffusion_timesteps)
         time_embed = rearrange(time_embed, 'b d -> b 1 d')
@@ -725,6 +776,7 @@ class DiffusionPriorNetwork(nn.Module):
             text_encodings,
             text_embed,
             time_embed,
+            image_embed,
             learned_queries
         ), dim = -2)
 
@@ -754,7 +806,9 @@ class DiffusionPrior(BaseGaussianDiffusion):
         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
         sampling_clamp_l2norm = False,
+        training_clamp_l2norm = False,
         image_embed_scale = None,           # this is for scaling the l2-normed image embedding, so it is more suitable for gaussian diffusion, as outlined by Katherine (@crowsonkb) https://github.com/lucidrains/DALLE2-pytorch/issues/60#issue-1226116132
+        clip_adapter_overrides = dict()
     ):
         super().__init__(
             beta_schedule = beta_schedule,
@@ -764,7 +818,9 @@ class DiffusionPrior(BaseGaussianDiffusion):
 
         if exists(clip):
             if isinstance(clip, CLIP):
-                clip = XClipAdapter(clip)
+                clip = XClipAdapter(clip, **clip_adapter_overrides)
+            elif isinstance(clip, CoCa):
+                clip = CoCaAdapter(clip, **clip_adapter_overrides)
 
             assert isinstance(clip, BaseClipAdapter)
             freeze_model_and_make_eval_(clip)
@@ -788,6 +844,7 @@ class DiffusionPrior(BaseGaussianDiffusion):
 
         # whether to force an l2norm, similar to clipping denoised, when sampling
         self.sampling_clamp_l2norm = sampling_clamp_l2norm
+        self.training_clamp_l2norm = training_clamp_l2norm
 
     def p_mean_variance(self, x, t, text_cond, clip_denoised: bool):
         pred = self.net(x, t, **text_cond)
@@ -840,6 +897,9 @@ class DiffusionPrior(BaseGaussianDiffusion):
             **text_cond
         )
 
+        if self.predict_x_start and self.training_clamp_l2norm:
+            pred = l2norm(pred) * self.image_embed_scale
+
         target = noise if not self.predict_x_start else image_embed
 
         loss = self.loss_fn(pred, target)
@@ -1487,7 +1547,8 @@ class Decoder(BaseGaussianDiffusion):
         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
         clip_denoised = True,
-        clip_x_start = True
+        clip_x_start = True,
+        clip_adapter_overrides = dict()
     ):
         super().__init__(
             beta_schedule = beta_schedule,
@@ -1500,7 +1561,9 @@ class Decoder(BaseGaussianDiffusion):
         self.clip = None
         if exists(clip):
             if isinstance(clip, CLIP):
-                clip = XClipAdapter(clip)
+                clip = XClipAdapter(clip, **clip_adapter_overrides)
+            elif isinstance(clip, CoCa):
+                clip = CoCaAdapter(clip, **clip_adapter_overrides)
 
             freeze_model_and_make_eval_(clip)
             assert isinstance(clip, BaseClipAdapter)

dalle2_pytorch/train.py

@@ -111,11 +111,6 @@ class DiffusionPriorTrainer(nn.Module):
         # exponential moving average
 
         self.use_ema = use_ema
-
-        if use_ema:
-            has_lazy_linear = any([type(module) == nn.LazyLinear for module in diffusion_prior.modules()])
-            assert not has_lazy_linear, 'you must set the text_embed_dim on your u-nets if you plan on doing automatic exponential moving average'
-
         if self.use_ema:
             self.ema_diffusion_prior = EMA(diffusion_prior, **ema_kwargs)
 

dalle2_pytorch/train_vqgan_vae.py

@@ -3,14 +3,15 @@ import copy
 from random import choice
 from pathlib import Path
 from shutil import rmtree
+from PIL import Image
 
 import torch
 from torch import nn
-
-from PIL import Image
-from torchvision.datasets import ImageFolder
-import torchvision.transforms as T
+from torch.cuda.amp import autocast, GradScaler
 from torch.utils.data import Dataset, DataLoader, random_split
+
+import torchvision.transforms as T
+from torchvision.datasets import ImageFolder
 from torchvision.utils import make_grid, save_image
 
 from einops import rearrange
@@ -99,6 +100,7 @@ class VQGanVAETrainer(nn.Module):
         ema_update_after_step = 2000,
         ema_update_every = 10,
         apply_grad_penalty_every = 4,
+        amp = False
     ):
         super().__init__()
         assert isinstance(vae, VQGanVAE), 'vae must be instance of VQGanVAE'
@@ -120,6 +122,10 @@ class VQGanVAETrainer(nn.Module):
         self.optim = get_optimizer(vae_parameters, lr = lr, wd = wd)
         self.discr_optim = get_optimizer(discr_parameters, lr = lr, wd = wd)
 
+        self.amp = amp
+        self.scaler = GradScaler(enabled = amp)
+        self.discr_scaler = GradScaler(enabled = amp)
+
         # create dataset
 
         self.ds = ImageDataset(folder, image_size = image_size)
@@ -178,20 +184,22 @@ class VQGanVAETrainer(nn.Module):
             img = next(self.dl)
             img = img.to(device)
 
-            loss = self.vae(
-                img,
-                return_loss = True,
-                apply_grad_penalty = apply_grad_penalty
-            )
+            with autocast(enabled = self.amp):
+                loss = self.vae(
+                    img,
+                    return_loss = True,
+                    apply_grad_penalty = apply_grad_penalty
+                )
+
+
+                self.scaler.scale(loss / self.grad_accum_every).backward()
 
             accum_log(logs, {'loss': loss.item() / self.grad_accum_every})
 
-            (loss / self.grad_accum_every).backward()
-
-        self.optim.step()
+        self.scaler.step(self.optim)
+        self.scaler.update()
         self.optim.zero_grad()
 
-
         # update discriminator
 
         if exists(self.vae.discr):
@@ -200,12 +208,15 @@ class VQGanVAETrainer(nn.Module):
                 img = next(self.dl)
                 img = img.to(device)
 
-                loss = self.vae(img, return_discr_loss = True)
+                with autocast(enabled = self.amp):
+                    loss = self.vae(img, return_discr_loss = True)
+
+                    self.discr_scaler.scale(loss / self.grad_accum_every).backward()
+
                 accum_log(logs, {'discr_loss': loss.item() / self.grad_accum_every})
 
-                (loss / self.grad_accum_every).backward()
-
-            self.discr_optim.step()
+            self.discr_scaler.step(self.discr_optim)
+            self.discr_scaler.update()
             self.discr_optim.zero_grad()
 
             # log

setup.py

@@ -10,7 +10,7 @@ setup(
       'dream = dalle2_pytorch.cli:dream'
     ],
   },
-  version = '0.0.108',
+  version = '0.1.6',
   license='MIT',
   description = 'DALL-E 2',
   author = 'Phil Wang',
@@ -24,12 +24,14 @@ setup(
   install_requires=[
     'click',
     'clip-anytorch',
+    'coca-pytorch>=0.0.5',
     'einops>=0.4',
     'einops-exts>=0.0.3',
     'embedding-reader',
     'kornia>=0.5.4',
     'pillow',
     'resize-right>=0.0.2',
+    'rotary-embedding-torch',
     'torch>=1.10',
     'torchvision',
     'tqdm',

train_diffusion_prior.py

@@ -46,28 +46,60 @@ def save_model(save_path, state_dict):
     print("====================================== Saving checkpoint ======================================")
     torch.save(state_dict, save_path+'/'+str(time.time())+'_saved_model.pth')
 
-def report_cosine_sims(diffusion_prior,image_reader,text_reader,train_set_size,val_set_size,NUM_TEST_EMBEDDINGS,device):
+
+def report_cosine_sims(diffusion_prior, image_reader, text_reader, train_set_size, val_set_size, NUM_TEST_EMBEDDINGS, device):
     cos = nn.CosineSimilarity(dim=1, eps=1e-6)
 
     tstart = train_set_size+val_set_size
     tend = train_set_size+val_set_size+NUM_TEST_EMBEDDINGS
 
-    for embt, embi in zip(text_reader(batch_size = NUM_TEST_EMBEDDINGS, start=tstart, end = tend),image_reader(batch_size = NUM_TEST_EMBEDDINGS, start=tstart, end = tend)):
+    for embt, embi in zip(text_reader(batch_size=NUM_TEST_EMBEDDINGS, start=tstart, end=tend), image_reader(batch_size=NUM_TEST_EMBEDDINGS, start=tstart, end=tend)):
+        # make a copy of the text embeddings for shuffling
         text_embed = torch.tensor(embt[0]).to(device)
-        text_embed = text_embed /  text_embed.norm(dim=1, keepdim=True)
-        test_text_cond = dict(text_embed = text_embed)
+        text_embed_shuffled = text_embed.clone()
 
+        # roll the text embeddings to simulate "unrelated" captions
+        rolled_idx = torch.roll(torch.arange(NUM_TEST_EMBEDDINGS), 1)
+        text_embed_shuffled = text_embed_shuffled[rolled_idx]
+        text_embed_shuffled = text_embed_shuffled / \
+            text_embed_shuffled.norm(dim=1, keepdim=True)
+        test_text_shuffled_cond = dict(text_embed=text_embed_shuffled)
+
+        # prepare the text embedding
+        text_embed = text_embed / text_embed.norm(dim=1, keepdim=True)
+        test_text_cond = dict(text_embed=text_embed)
+
+        # prepare image embeddings
         test_image_embeddings = torch.tensor(embi[0]).to(device)
-        test_image_embeddings = test_image_embeddings /  test_image_embeddings.norm(dim=1, keepdim=True)
+        test_image_embeddings = test_image_embeddings / \
+            test_image_embeddings.norm(dim=1, keepdim=True)
 
-        predicted_image_embeddings = diffusion_prior.p_sample_loop((NUM_TEST_EMBEDDINGS, 768), text_cond = test_text_cond)
-        predicted_image_embeddings = predicted_image_embeddings / predicted_image_embeddings.norm(dim=1, keepdim=True)
+        # predict on the unshuffled text embeddings
+        predicted_image_embeddings = diffusion_prior.p_sample_loop(
+            (NUM_TEST_EMBEDDINGS, 768), text_cond=test_text_cond)
+        predicted_image_embeddings = predicted_image_embeddings / \
+            predicted_image_embeddings.norm(dim=1, keepdim=True)
 
-        original_similarity = cos(text_embed,test_image_embeddings).cpu().numpy()
-        predicted_similarity = cos(text_embed,predicted_image_embeddings).cpu().numpy()
+        # predict on the shuffled embeddings
+        predicted_unrelated_embeddings = diffusion_prior.p_sample_loop(
+            (NUM_TEST_EMBEDDINGS, 768), text_cond=test_text_shuffled_cond)
+        predicted_unrelated_embeddings = predicted_unrelated_embeddings / \
+            predicted_unrelated_embeddings.norm(dim=1, keepdim=True)
 
-        wandb.log({"CosineSimilarity(text_embed,image_embed)": np.mean(original_similarity)})
-        wandb.log({"CosineSimilarity(text_embed,predicted_image_embed)":np.mean(predicted_similarity)})
+        # calculate similarities
+        original_similarity = cos(
+            text_embed, test_image_embeddings).cpu().numpy()
+        predicted_similarity = cos(
+            text_embed, predicted_image_embeddings).cpu().numpy()
+        unrelated_similarity = cos(
+            text_embed, predicted_unrelated_embeddings).cpu().numpy()
+
+        wandb.log(
+            {"CosineSimilarity(text_embed,image_embed)": np.mean(original_similarity)})
+        wandb.log({"CosineSimilarity(text_embed,predicted_image_embed)": np.mean(
+            predicted_similarity)})
+        wandb.log({"CosineSimilarity(text_embed,predicted_unrelated_embed)": np.mean(
+            unrelated_similarity)})
 
     return np.mean(predicted_similarity - original_similarity)