Commits

Branch selector

 main

User selector

lucidrains

Datepicker

All time

Commit History

Commits on Apr 20, 2023

• force einops 0.6.1 or greater and call allow_ops_in_compiled_graph

  lucidrains

  committed
  00e07b7

  Copy full SHA for 00e07b7

• remove einops exts for better pytorch 2.0 compile compatibility

  lucidrains

  committed
  0069857

  Copy full SHA for 0069857

Commits on Mar 7, 2023

• use .to(device) to avoid copy, within one_unet_in_gpu context

  lucidrains

  committed
  580274b

  Copy full SHA for 580274b

Commits on Mar 5, 2023

• always rederive the predicted noise from the clipped x0 for ddim + predict noise objective

  lucidrains

  committed
  848e8a4

  Copy full SHA for 848e8a4

Commits on Mar 4, 2023

• bump to newer package of clip-anytorch that allows for text encodings < maximum context length

  lucidrains

  committed
  cc58f75

  Copy full SHA for cc58f75

Commits on Feb 12, 2023

• fix for self conditioning in diffusion prior network lucidrains#273

  lucidrains

  committed
  3b2cf7b

  Copy full SHA for 3b2cf7b

Commits on Dec 23, 2022

• default ddim sampling eta to 0

  lucidrains

  committed
  984d62a

  Copy full SHA for 984d62a

Commits on Dec 15, 2022

• extra insurance in case eos id is not there

  lucidrains

  committed
  683dd98

  Copy full SHA for 683dd98

Commits on Nov 23, 2022

• address lucidrains#266

  lucidrains

  committed
  067ac32

  Copy full SHA for 067ac32

Commits on Oct 29, 2022

• bring in prediction of v objective, combining the findings from progressive distillation paper and imagen-video to the eventual extension of dalle2 to make-a-video

  lucidrains

  committed
  fbba0f9

  Copy full SHA for fbba0f9

Commits on Oct 23, 2022

• fix openclipadapter to be able to use latest open sourced sota model

  lucidrains

  committed
  c3df46e

  Copy full SHA for c3df46e

Commits on Oct 19, 2022

• fix a dtype conversion issue for the diffusion timesteps in the diffusion prior, thanks to @JiaHeng-DLUT

  lucidrains

  committed
  41fabf2

  Copy full SHA for 41fabf2

Commits on Sep 29, 2022

• fix for use with larger openai clip models by extracting dimension of last layernorm in clip

  lucidrains

  committed
  c18c080

  Copy full SHA for c18c080

Commits on Sep 20, 2022

• fix readme dataloader example

  lucidrains

  authored
  b39653c

  Copy full SHA for b39653c

Commits on Sep 19, 2022

• show example of using SOTA open sourced open clip

  lucidrains

  committed
  39f8b6c

  Copy full SHA for 39f8b6c

• handle open clip adapter image size being a tuple

  lucidrains

  committed
  d0c11b3

  Copy full SHA for d0c11b3

Commits on Sep 1, 2022

• in ddim, noise should be predicted after x0 is maybe clipped, thanks to @lukovnikov for pointing this out in another repository

  lucidrains

  committed
  0d82dff

  Copy full SHA for 0d82dff

• fix bug with misnamed variable in diffusion prior network

  lucidrains

  committed
  8bbc956

  Copy full SHA for 8bbc956

Commits on Aug 31, 2022

• todo

  lucidrains

  authored
  22019fd

  Copy full SHA for 22019fd

• fix ddim to use alpha_cumprod

  lucidrains

  committed
  6fb7e91

  Copy full SHA for 6fb7e91

Commits on Aug 28, 2022

• add two asserts to diffusion prior to ensure matching image embedding dimensions for clip, diffusion prior network, and what was set on diffusion prior

  lucidrains

  committed
  ba58ae0

  Copy full SHA for ba58ae0

Commits on Aug 25, 2022

• upgrade to best downsample

  lucidrains

  committed
  1cc5d0a

  Copy full SHA for 1cc5d0a

Commits on Aug 23, 2022

• fix classifier free guidance for diffusion prior, thanks to @jaykim9870 for spotting the issue

  lucidrains

  committed
  59fa101

  Copy full SHA for 59fa101

Commits on Aug 20, 2022

• cast attention matrix back to original dtype pre-softmax in attention

  lucidrains

  committed
  083508f

  Copy full SHA for 083508f

Commits on Aug 19, 2022

• make it work for @ethancohen123

  lucidrains

  committed
  7762edd

  Copy full SHA for 7762edd

Commits on Aug 17, 2022

• cite einops

  lucidrains

  authored
  de5e628

  Copy full SHA for de5e628

• gratitude

  lucidrains

  authored
  1b4046b

  Copy full SHA for 1b4046b

Commits on Aug 15, 2022

• make sure diffusion prior trainer can operate with no warmup

  lucidrains

  committed
  27f19ba

  Copy full SHA for 27f19ba

• give diffusion prior trainer cosine annealing lr too

  lucidrains

  committed
  8f38339

  Copy full SHA for 8f38339

• add cosine annealing lr schedule

  lucidrains

  committed
  6b9b4b9

  Copy full SHA for 6b9b4b9

Commits on Aug 14, 2022

• add weight standardization behind feature flag, which may potentially work well with group norm

  lucidrains

  committed
  44e09d5

  Copy full SHA for 44e09d5

Commits on Aug 13, 2022

• make it so diffusion prior p_sample_loop returns unnormalized image embeddings

  lucidrains

  committed
  3480666

  Copy full SHA for 3480666

Commits on Aug 12, 2022

• dry up some code around handling unet outputs with learned variance

  lucidrains

  committed
  dc816b1

  Copy full SHA for dc816b1

• fix self conditioning shape in diffusion prior

  lucidrains

  committed
  05192ff

  Copy full SHA for 05192ff

• make self conditioning technique work with diffusion prior

  lucidrains

  committed
  9440411

  Copy full SHA for 9440411

Pagination

Previous

Next