Commits

Branch selector

 main

User selector

All users

All time

Commit History

Commits on Oct 24, 2022

• product management

  lucidrains

  authored
  8ec3321

  Copy full SHA for 8ec3321

• remove big blocker

  lucidrains

  committed
  b54013b

  Copy full SHA for b54013b

• bring in the pseudo 3d conv for the 3dunet from make-a-video, and add the new prediction objective from the progressive distillation paper, which allows for distillation, and noted in imagen video …

  Show description for 0a3afb1
  lucidrains

  committed
  0a3afb1

  Copy full SHA for 0a3afb1

Commits on Oct 9, 2022

• quick fix for elucidated ddpm

  lucidrains

  committed
  fa29d24

  Copy full SHA for fa29d24

Commits on Oct 5, 2022

• todo

  lucidrains

  authored
  76aa736

  Copy full SHA for 76aa736

• todo

  lucidrains

  authored
  76d812d

  Copy full SHA for 76d812d

• 1.11.14

  lucidrains

  authored
  2cbf8e1

  Copy full SHA for 2cbf8e1

• Bugfix in image size assertion (lucidrains#230)

  ezhang7423

  authored
  c3afdf7

  Copy full SHA for c3afdf7

Commits on Sep 29, 2022

• product management

  lucidrains

  authored
  35b72c9

  Copy full SHA for 35b72c9

Commits on Sep 19, 2022

• remove a source of confusion

  lucidrains

  committed
  f879ec9

  Copy full SHA for f879ec9

Commits on Sep 9, 2022

• make sure raw texts never include empty strings

  lucidrains

  committed
  e10e92e

  Copy full SHA for e10e92e

• make sure batch size is set correctly, if only inpainting images are passed in, for unconditional image synthesis. also allow for forcing unconditional image synthesis on network trained conditiona…

  Show description for 038f300
  lucidrains

  committed
  038f300

  Copy full SHA for 038f300

• add an extra assert to make sure researcher / user cannot sample without text on a network with cross attention blocks

  lucidrains

  committed
  77cc6a3

  Copy full SHA for 77cc6a3

Commits on Sep 7, 2022

• 1.11.8

  lucidrains

  committed
  b6563fd

  Copy full SHA for b6563fd

• pass max length to tokenizer (lucidrains#213)

  haukened

  authored
  8305f58

  Copy full SHA for 8305f58

• 1.11.7

  lucidrains

  authored
  ed6426f

  Copy full SHA for ed6426f

• fix error when passing use_tqdm to trainer.sample() (lucidrains#212)

  Show description for da6b6b2
  Nodja

  authored
  da6b6b2

  Copy full SHA for da6b6b2

Commits on Sep 3, 2022

• fix ddp and self conditioning in elucidated ddpm

  lucidrains

  committed
  f973004

  Copy full SHA for f973004

Commits on Sep 1, 2022

• make sure self conditining works in distributed training, cleanup later, bump

  lucidrains

  committed
  a5d69b9

  Copy full SHA for a5d69b9

• Fixing DDP Unet Access when Training in a MultiGPU Setting (lucidrains#206)

  Show description for 349675a
  vfragoso

  authored
  349675a

  Copy full SHA for 349675a

Commits on Aug 31, 2022

• fix a bug with inpainting for non-elucidated ddpm

  lucidrains

  committed
  52d5a73

  Copy full SHA for 52d5a73

Commits on Aug 30, 2022

• allow for one to pass in a tuple to use_linear_attn kwargs in unet, to finely specify which layers get sparse linear attention, and to turn off attention altogether

  lucidrains

  committed
  d4c45ab

  Copy full SHA for d4c45ab

Commits on Aug 27, 2022

• product management

  lucidrains

  committed
  59b5e55

  Copy full SHA for 59b5e55

• an extra assert to make sure images are not integer type

  lucidrains

  committed
  ceb23d6

  Copy full SHA for ceb23d6

Commits on Aug 25, 2022

• update with the best type of downsample

  lucidrains

  committed
  48361f6

  Copy full SHA for 48361f6

Commits on Aug 23, 2022

• exclude masks from being casted to float16

  lucidrains

  committed
  967b1be

  Copy full SHA for 967b1be

Commits on Aug 22, 2022

• dry custom layernorm

  lucidrains

  committed
  852a39a

  Copy full SHA for 852a39a

• make sure fp16 works

  lucidrains

  committed
  818ab5d

  Copy full SHA for 818ab5d

Commits on Aug 21, 2022

• 1.9.6

  lucidrains

  authored
  38cde97

  Copy full SHA for 38cde97

• load checkpoint model on cpu first using map_location (lucidrains#190)

  deepglugs

  authored
  b22daa1

  Copy full SHA for b22daa1

Commits on Aug 20, 2022

• all attention softmax in float32 and cast back

  lucidrains

  committed
  afed17a

  Copy full SHA for afed17a

Commits on Aug 19, 2022

• 1.9.4

  lucidrains

  committed
  97020dd

  Copy full SHA for 97020dd

• Return only the model when model_only==True (lucidrains#188)

  Show description for 10962d3
  deepglugs

  authored
  10962d3

  Copy full SHA for 10962d3

Commits on Aug 18, 2022

• release @deepglugs change to flexible state dict loading

  lucidrains

  committed
  8e9ca76

  Copy full SHA for 8e9ca76

• try to restore as much as possible if state_dict fails (lucidrains#184)

  deepglugs

  authored
  b9b45ad

  Copy full SHA for b9b45ad

Pagination

Previous

Next