update sample, and set default gradient clipping value for decoder training

README.md

@@ -18,7 +18,11 @@ There was enough interest for a <a href="https://github.com/lucidrains/dalle2-ja
 
 - A research group has used the code in this repository to train a functional diffusion prior for their CLIP generations. Will share their work once they release their preprint. This, and <a href="https://github.com/crowsonkb">Katherine's</a> own experiments, validate OpenAI's finding that the extra prior increases variety of generations.
 
-- Decoder is now verified working for unconditional generation on my experimental setup for Oxford flowers
+- Decoder is now verified working for unconditional generation on my experimental setup for Oxford flowers. 2 researchers have also confirmed Decoder is working for them.
+
+<img src="./samples/oxford.png" width="600px" />
+
+*ongoing at 21k steps*
 
 ## Install
 
@@ -820,8 +824,8 @@ clip = CLIP(
 
 # mock data
 
-text = torch.randint(0, 49408, (32, 256)).cuda()
-images = torch.randn(32, 3, 256, 256).cuda()
+text = torch.randint(0, 49408, (512, 256)).cuda()
+images = torch.randn(512, 3, 256, 256).cuda()
 
 # prior networks (with transformer)
 
@@ -854,7 +858,7 @@ diffusion_prior_trainer.update()  # this will update the optimizer as well as th
 # after much of the above three lines in a loop
 # you can sample from the exponential moving average of the diffusion prior identically to how you do so for DiffusionPrior
 
-image_embeds = diffusion_prior_trainer.sample(text) # (4, 512) - exponential moving averaged image embeddings
+image_embeds = diffusion_prior_trainer.sample(text, max_batch_size = 4) # (512, 512) - exponential moving averaged image embeddings
 ```
 
 ## Bonus
@@ -867,7 +871,7 @@ ex.
 
 ```python
 import torch
-from dalle2_pytorch import Unet, Decoder
+from dalle2_pytorch import Unet, Decoder, DecoderTrainer
 
 # unet for the cascading ddpm
 
@@ -890,20 +894,24 @@ decoder = Decoder(
     unconditional = True
 ).cuda()
 
-# mock images (get a lot of this)
+# decoder trainer
+
+decoder_trainer = DecoderTrainer(decoder)
+
+# images (get a lot of this)
 
 images = torch.randn(1, 3, 512, 512).cuda()
 
 # feed images into decoder
 
 for i in (1, 2):
-    loss = decoder(images, unet_number = i)
-    loss.backward()
+    loss = decoder_trainer(images, unet_number = i)
+    decoder_trainer.update(unet_number = i)
 
-# do the above for many many many many steps
+# do the above for many many many many images
 # then it will learn to generate images
 
-images = decoder.sample(batch_size = 2) # (2, 3, 512, 512)
+images = decoder_trainer.sample(batch_size = 36, max_batch_size = 4) # (36, 3, 512, 512)
 ```
 
 ## Dataloaders

dalle2_pytorch/optimizer.py

@@ -7,7 +7,7 @@ def separate_weight_decayable_params(params):
 
 def get_optimizer(
     params,
-    lr = 2e-5,
+    lr = 1e-4,
     wd = 1e-2,
     betas = (0.9, 0.999),
     eps = 1e-8,

dalle2_pytorch/trainer.py

@@ -235,6 +235,16 @@ class EMA(nn.Module):
 
 # diffusion prior trainer
 
+def prior_sample_in_chunks(fn):
+    @wraps(fn)
+    def inner(self, *args, max_batch_size = None, **kwargs):
+        if not exists(max_batch_size):
+            return fn(self, *args, **kwargs)
+
+        outputs = [fn(self, *chunked_args, **chunked_kwargs) for _, (chunked_args, chunked_kwargs) in split_args_and_kwargs(*args, split_size = max_batch_size, **kwargs)]
+        return torch.cat(outputs, dim = 0)
+    return inner
+
 class DiffusionPriorTrainer(nn.Module):
     def __init__(
         self,
@@ -295,11 +305,13 @@ class DiffusionPriorTrainer(nn.Module):
 
     @torch.no_grad()
     @cast_torch_tensor
+    @prior_sample_in_chunks
     def p_sample_loop(self, *args, **kwargs):
         return self.ema_diffusion_prior.ema_model.p_sample_loop(*args, **kwargs)
 
     @torch.no_grad()
     @cast_torch_tensor
+    @prior_sample_in_chunks
     def sample(self, *args, **kwargs):
         return self.ema_diffusion_prior.ema_model.sample(*args, **kwargs)
 
@@ -351,10 +363,10 @@ class DecoderTrainer(nn.Module):
         self,
         decoder,
         use_ema = True,
-        lr = 2e-5,
+        lr = 1e-4,
         wd = 1e-2,
         eps = 1e-8,
-        max_grad_norm = None,
+        max_grad_norm = 0.5,
         amp = False,
         **kwargs
     ):

setup.py

@@ -10,7 +10,7 @@ setup(
       'dream = dalle2_pytorch.cli:dream'
     ],
   },
-  version = '0.2.46',
+  version = '0.3.2',
   license='MIT',
   description = 'DALL-E 2',
   author = 'Phil Wang',