update readme so users understand actual dalle is much bigger

README.md

@@ -29,9 +29,10 @@ from dalle_pytorch import DiscreteVAE
 vae = DiscreteVAE(
     image_size = 256,
     num_layers = 3,         # number of downsamples - ex. 256 / (2 ** 3) = (32 x 32 feature map)
-    num_tokens = 1024,      # number of visual tokens. iGPT had 512, so probably should have more
+    num_tokens = 8192,      # number of visual tokens. in the paper, they used 8192, but could be smaller for downsized projects
     codebook_dim = 512,     # codebook dimension
     hidden_dim = 64,        # hidden dimension
+    num_resnet_blocks = 1,  # number of resnet blocks
     temperature = 0.9       # gumbel softmax temperature, the lower this is, the more hard the discretization
 )
 
@@ -52,19 +53,20 @@ from dalle_pytorch import DiscreteVAE, DALLE
 vae = DiscreteVAE(
     image_size = 256,
     num_layers = 3,
-    num_tokens = 1024,
-    codebook_dim = 512,
+    num_tokens = 8192,
+    codebook_dim = 1024,
     hidden_dim = 64,
+    num_resnet_blocks = 1,
     temperature = 0.9
 )
 
 dalle = DALLE(
-    dim = 512,
+    dim = 1024,
     vae = vae,                  # automatically infer (1) image sequence length and (2) number of image tokens
     num_text_tokens = 10000,    # vocab size for text
     text_seq_len = 256,         # text sequence length
-    depth = 6,                  # should be 64
-    heads = 8,                  # attention heads
+    depth = 12,                 # should aim to be 64
+    heads = 16,                 # attention heads
     dim_head = 64,              # attention head dimension
     attn_dropout = 0.1,         # attention dropout
     ff_dropout = 0.1            # feedforward dropout
@@ -135,12 +137,12 @@ Simply set the `reversible` keyword to `True` for the `DALLE` class
 
 ```python
 dalle = DALLE(
-    dim = 512,
+    dim = 1024,
     vae = vae,
     num_text_tokens = 10000,
     text_seq_len = 256,
     depth = 64,
-    heads = 8,
+    heads = 16,
     reversible = True  # <-- reversible networks https://arxiv.org/abs/2001.04451
 )
 ```