Releases: lucidrains/DALLE-pytorch
Releases · lucidrains/DALLE-pytorch
0.0.61
fix mask for full attention so that last text token can actually pred…
0.0.60
remove some dead code
0.0.59
remove ability to do bidirectional attention on text, to keep reposit…
0.0.58
handle other types being passed to `attn_types` keyword arg
0.0.56
always use mask if sparse axial or conv attention needs to be padded
0.0.55
complete sparse attention integration with transformer, user can now …
0.0.54
do not tie vae codebook to dall-e image embedding by default, and if …
0.0.53
fix causal masking in sparse conv attention
0.0.52
allow for dilation in sparse convolutional causal attention
0.0.51
complete sparse axial causal attention, where each row or column has …