Commits

Commits on Jul 6, 2022

remove caption mask from rainbow_dalle example, as it is no longer needed in most recent version
lucidrains
committedJul 6, 2022

Commits on May 30, 2022

add packaging package
lucidrains
committedMay 30, 2022
fix version problem if researcher did not install from pip
lucidrains
committedMay 30, 2022

Commits on May 16, 2022

add an assert on torch version in order to use OpenAIDiscreteVAE
lucidrains
committedMay 16, 2022

Commits on May 10, 2022

fix remaining issues with transparencies when generating and training DALL-E
lucidrains
committedMay 10, 2022
fix all issues with training on transparent pngs
lucidrains
committedMay 10, 2022

Commits on May 9, 2022

support for transparent images
lucidrains
committedMay 9, 2022

Commits on Apr 13, 2022

cleaner way to organize code around RiverHasWings application of classifier free guidance to transformers
lucidrains
committedApr 13, 2022

Commits on Apr 8, 2022

link to dalle-2
lucidrains
committedApr 8, 2022

Commits on Mar 24, 2022

cleanup conv-like attention
lucidrains
committedMar 24, 2022
switch to causal convolution for conv-like attention using unfold
lucidrains
committedMar 24, 2022

Commits on Mar 16, 2022

link to NUWA
lucidrains
committedMar 16, 2022

Commits on Jan 18, 2022

1.4.2
lucidrains
committedJan 18, 2022
fix tokenizer decoding (lucidrains#410 )
tillfalko
committedJan 18, 2022

Commits on Jan 12, 2022

remove g-mlps
lucidrains
committedJan 12, 2022

Commits on Jan 11, 2022

1.4.0 for improved inference speeds, thanks to @borzunov
lucidrains
committedJan 11, 2022
Implement cached inference (lucidrains#409 )
borzunov
committedJan 11, 2022

Commits on Jan 7, 2022

make sure generate_text works when shift token feature is turned on
lucidrains
committedJan 7, 2022

Commits on Jan 4, 2022

add super-conditioning feature, idea originated from @crowsonkb
lucidrains
committedJan 4, 2022
remove some dead code, text mask is not needed when training DALL-E
lucidrains
committedJan 4, 2022

Commits on Jan 3, 2022

remove softmax and torch multinomial when sampling
lucidrains
committedJan 3, 2022
stable softmax when generating images
lucidrains
committedJan 3, 2022

Commits on Dec 25, 2021

Commits on Dec 2, 2021

fix distributed webdataset : divide batch size by world size in webdataset batching (lucidrains#397 )
rom1504
committedDec 2, 2021

Commits on Dec 1, 2021

switch to stable softmax when sampling
lucidrains
committedDec 1, 2021

Commits on Nov 26, 2021

fix: device designation && sum() function error to run on colab environment (lucidrains#391 )
snoop2head
committedNov 26, 2021

Commits on Nov 12, 2021

1.1.4
lucidrains
committedNov 12, 2021

Commits on Nov 9, 2021

1.1.2
lucidrains
committedNov 9, 2021
Fix rotary embeddings (lucidrains#383 )
borzunov
committedNov 9, 2021

Commits on Nov 6, 2021

add additional stabilization technique, looking at ruDALL-E results
lucidrains
committedNov 6, 2021

Commits on Nov 4, 2021

date
lucidrains
committedNov 4, 2021
ruDALL-E link
lucidrains
committedNov 4, 2021

Commits on Oct 19, 2021

add sandwich norm, from the coqview paper, for stabilizing training even more, hidden behind feature flag
lucidrains
committedOct 19, 2021

Commits on Sep 29, 2021

no rotary embedding for transformers within CLIP model
lucidrains
committedSep 29, 2021