Commits

Branch selector

 main

User selector

All users

Datepicker

All time

Commit History

Commits on Jul 6, 2022

• remove caption mask from rainbow_dalle example, as it is no longer needed in most recent version

  lucidrains

  committed
  daf30d0

  Copy full SHA for daf30d0

Commits on May 30, 2022

• add packaging package

  lucidrains

  committed
  d7bd745

  Copy full SHA for d7bd745

• fix version problem if researcher did not install from pip

  lucidrains

  committed
  343f8b0

  Copy full SHA for 343f8b0

Commits on May 16, 2022

• add an assert on torch version in order to use OpenAIDiscreteVAE

  lucidrains

  committed
  d03df6a

  Copy full SHA for d03df6a

Commits on May 10, 2022

• fix remaining issues with transparencies when generating and training DALL-E

  lucidrains

  committed
  2612a51

  Copy full SHA for 2612a51

• fix all issues with training on transparent pngs

  lucidrains

  committed
  a6776c8

  Copy full SHA for a6776c8

Commits on May 9, 2022

• support for transparent images

  lucidrains

  committed
  bebc280

  Copy full SHA for bebc280

Commits on Apr 13, 2022

• cleaner way to organize code around RiverHasWings application of classifier free guidance to transformers

  lucidrains

  committed
  fcd35de

  Copy full SHA for fcd35de

Commits on Apr 8, 2022

• link to dalle-2

  lucidrains

  authored
  144ce50

  Copy full SHA for 144ce50

Commits on Mar 24, 2022

• cleanup conv-like attention

  lucidrains

  committed
  2969103

  Copy full SHA for 2969103

• switch to causal convolution for conv-like attention using unfold

  lucidrains

  committed
  3064403

  Copy full SHA for 3064403

Commits on Mar 16, 2022

• link to NUWA

  lucidrains

  authored
  0cf55f2

  Copy full SHA for 0cf55f2

Commits on Jan 18, 2022

• 1.4.2

  lucidrains

  authored
  289fddb

  Copy full SHA for 289fddb

• fix tokenizer decoding (lucidrains#410)

  tillfalko

  authored
  3f2cb7c

  Copy full SHA for 3f2cb7c

Commits on Jan 12, 2022

• remove g-mlps

  lucidrains

  committed
  9526cd6

  Copy full SHA for 9526cd6

Commits on Jan 11, 2022

• 1.4.0 for improved inference speeds, thanks to @borzunov

  lucidrains

  committed
  6aff102

  Copy full SHA for 6aff102

• Implement cached inference (lucidrains#409)

  Show description for 2094474
  borzunov

  authored
  2094474

  Copy full SHA for 2094474

Commits on Jan 7, 2022

• make sure generate_text works when shift token feature is turned on

  lucidrains

  committed
  01433fb

  Copy full SHA for 01433fb

Commits on Jan 4, 2022

• add super-conditioning feature, idea originated from @crowsonkb

  lucidrains

  committed
  459c46a

  Copy full SHA for 459c46a

• remove some dead code, text mask is not needed when training DALL-E

  lucidrains

  committed
  4511d2d

  Copy full SHA for 4511d2d

Commits on Jan 3, 2022

• remove softmax and torch multinomial when sampling

  lucidrains

  committed
  07f6f3f

  Copy full SHA for 07f6f3f

• stable softmax when generating images

  lucidrains

  committed
  2783801

  Copy full SHA for 2783801

Commits on Dec 25, 2021

• more cleanup

  lucidrains

  committed
  4a7958d

  Copy full SHA for 4a7958d

• start storing version of dalle-pytorch alongside model weights - also throw an error if trying to generate with a dalle-pytorch whose VAE is not the same as the type with which it was trained on

  lucidrains

  committed
  1c25f54

  Copy full SHA for 1c25f54

Commits on Dec 2, 2021

• fix distributed webdataset : divide batch size by world size in webdataset batching (lucidrains#397)

  Show description for e1d10b9
  rom1504

  authored
  e1d10b9

  Copy full SHA for e1d10b9

Commits on Dec 1, 2021

• switch to stable softmax when sampling

  lucidrains

  committed
  1ad3ab8

  Copy full SHA for 1ad3ab8

Commits on Nov 26, 2021

• fix: device designation && sum() function error to run on colab environment (lucidrains#391)

  Show description for 2df5ff0
  snoop2head

  authored
  2df5ff0

  Copy full SHA for 2df5ff0

Commits on Nov 12, 2021

• 1.1.4

  lucidrains

  committed
  ae4d88a

  Copy full SHA for ae4d88a

Commits on Nov 9, 2021

• 1.1.2

  lucidrains

  authored
  a34d5d9

  Copy full SHA for a34d5d9

• Fix rotary embeddings (lucidrains#383)

  borzunov

  authored
  e31bbfc

  Copy full SHA for e31bbfc

Commits on Nov 6, 2021

• add additional stabilization technique, looking at ruDALL-E results

  lucidrains

  committed
  04b6feb

  Copy full SHA for 04b6feb

Commits on Nov 4, 2021

• date

  lucidrains

  authored
  1dcf6c3

  Copy full SHA for 1dcf6c3

• ruDALL-E link

  lucidrains

  authored
  8c370c8

  Copy full SHA for 8c370c8

Commits on Oct 19, 2021

• add sandwich norm, from the coqview paper, for stabilizing training even more, hidden behind feature flag

  lucidrains

  committed
  e4e101f

  Copy full SHA for e4e101f

Commits on Sep 29, 2021

• no rotary embedding for transformers within CLIP model

  lucidrains

  committed
  15d2f35

  Copy full SHA for 15d2f35

Pagination

Previous

Next