Commits

Branch selector

 main

User selector

All users

Datepicker

All time

Commit History

Commits on Aug 12, 2022

• Free some resources after each step to avoid OOM

  andrewmoise

  committed
  b25168e

  Copy full SHA for b25168e

Commits on Aug 10, 2022

• add self conditioning for elucidated ddpm

  lucidrains

  committed
  beb2f2d

  Copy full SHA for beb2f2d

• fix sampling ddpm tqdm

  lucidrains

  committed
  f0d59ac

  Copy full SHA for f0d59ac

• add the new self conditioning technique from hintons group from bit diffusion paper

  Show description for 689593a
  lucidrains

  committed
  689593a

  Copy full SHA for 689593a

Commits on Jul 29, 2022

• higher epsilon for fp16 in layernorm

  lucidrains

  committed
  eba4449

  Copy full SHA for eba4449

Commits on Jul 27, 2022

• rescale values to prevent linear attention from overflowing in fp16 setting

  lucidrains

  committed
  12f95b3

  Copy full SHA for 12f95b3

Commits on Jul 25, 2022

• fix accelerator prepare bug for dataloader

  lucidrains

  committed
  6b504c4

  Copy full SHA for 6b504c4

Commits on Jul 19, 2022

• fix a bug with ddim and predict x0 objective

  lucidrains

  committed
  37334ae

  Copy full SHA for 37334ae

Commits on Jul 18, 2022

• refresh images

  lucidrains

  committed
  6eba6cd

  Copy full SHA for 6eba6cd

• take a gamble on cosine sim attention

  lucidrains

  committed
  555566c

  Copy full SHA for 555566c

Commits on Jul 12, 2022

• move accelerator backward outside of autocast context, also calculate total loss correctly across gradient accumulated steps

  lucidrains

  committed
  2b742dd

  Copy full SHA for 2b742dd

Commits on Jul 10, 2022

• do not noise at the last timestep for ddim

  lucidrains

  committed
  1345a8a

  Copy full SHA for 1345a8a

Commits on Jul 9, 2022

• bring in ddim sampling

  lucidrains

  committed
  931a5af

  Copy full SHA for 931a5af

• optimizer should be saved and loaded

  lucidrains

  committed
  a0c3443

  Copy full SHA for a0c3443

Commits on Jul 8, 2022

• add convert_image_to keyword argument, for forcing images being loaded to be converted to some format, greyscale, rgb, rgba, whatever

  lucidrains

  committed
  6621728

  Copy full SHA for 6621728

Commits on Jul 6, 2022

• also make sure grad scaler actually exists in the saved pt file

  lucidrains

  committed
  0248b5e

  Copy full SHA for 0248b5e

• support multi-gpu training using huggingface accelerate, addressing lucidrains#54

  lucidrains

  committed
  6b56af0

  Copy full SHA for 6b56af0

Commits on Jun 30, 2022

• tqdm auto instead

  lucidrains

  committed
  d442024

  Copy full SHA for d442024

Commits on Jun 29, 2022

• reorganize elucidating code

  lucidrains

  committed
  a536e5b

  Copy full SHA for a536e5b

• add clamping option to elucidated diffusion

  lucidrains

  committed
  1b85379

  Copy full SHA for 1b85379

• patch

  lucidrains

  committed
  8859864

  Copy full SHA for 8859864

• Merge pull request lucidrains#52 from AryaAftab/patch-1

  Show description for 32657f0
  lucidrains

  authored
  32657f0

  Copy full SHA for 32657f0

• Solve issue lucidrains#26

  AryaAftab

  authored
  d97bc02

  Copy full SHA for d97bc02

• fix bug in elucidating sampling

  lucidrains

  committed
  8408775

  Copy full SHA for 8408775

• release elucidating diffusion

  lucidrains

  committed
  86fcb67

  Copy full SHA for 86fcb67

• Merge pull request lucidrains#51 from lucidrains/pw/elucidating-ddpm

  Show description for c535d31
  lucidrains

  authored
  c535d31

  Copy full SHA for c535d31

• refactor sigmas and gamma generation

  lucidrains

  committed
  5db64fe

  Copy full SHA for 5db64fe

Commits on Jun 28, 2022

• fix off by one

  lucidrains

  committed
  b87ea27

  Copy full SHA for b87ea27

• no clamping when training from sigmas drawn from log normal distribution, clamp final images being sampled

  lucidrains

  committed
  a8403b8

  Copy full SHA for a8403b8

• complete a first pass of elucidated ddpm

  lucidrains

  committed
  f4b1d7a

  Copy full SHA for f4b1d7a

• clamp the sigma coming out of the log normal distribution

  lucidrains

  committed
  6184937

  Copy full SHA for 6184937

• take care of equation 7 in the paper

  lucidrains

  committed
  76b79aa

  Copy full SHA for 76b79aa

• cleanup again

  lucidrains

  committed
  be2bd8d

  Copy full SHA for be2bd8d

• cleanup

  lucidrains

  committed
  c3d1607

  Copy full SHA for c3d1607

• get training working

  lucidrains

  committed
  06b2e52

  Copy full SHA for 06b2e52

Pagination

Previous

Next