Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Jun 21, 2023

• want to update batch stats during eval

  theAdamColton

  committed
  1fcd6b2

  Copy full SHA for 1fcd6b2

• fix sd loading, don't update affine during eval

  theAdamColton

  committed
  6d415ef

  Copy full SHA for 6d415ef

• loosen restriction around needing an separate optimizer for learnable codebook

  lucidrains

  committed
  60a494a

  Copy full SHA for 60a494a

• address straight through lucidrains#59

  lucidrains

  committed
  f867b33

  Copy full SHA for f867b33

Commits on Jun 19, 2023

• patch

  lucidrains

  committed
  1a59a89

  Copy full SHA for 1a59a89

• Merge pull request lucidrains#57 from theAdamColton/fix_mem_leak

  Show description for 8a9bb74
  lucidrains

  authored
  8a9bb74

  Copy full SHA for 8a9bb74

• detach buffer before registering

  theAdamColton

  committed
  16c56b5

  Copy full SHA for 16c56b5

• release inplace codebook optimizer, thanks to @theAdamColton

  lucidrains

  committed
  1f328ad

  Copy full SHA for 1f328ad

• Merge pull request lucidrains#54 from theAdamColton/inplace_update

  Show description for a82e6f3
  lucidrains

  authored
  a82e6f3

  Copy full SHA for a82e6f3

Commits on Jun 16, 2023

• fix mps detection

  lucidrains

  committed
  878234d

  Copy full SHA for 878234d

Commits on Jun 15, 2023

• fix memory leak issue, thanks to @fauzanardh

  lucidrains

  committed
  dea3bfa

  Copy full SHA for dea3bfa

Commits on Jun 14, 2023

• do not used detached quantized tensor for commit loss, to support learnable codebooks correctly

  lucidrains

  committed
  8a5a159

  Copy full SHA for 8a5a159

• make sure library can work on mac m1/m2 chip - lucidrains#55

  lucidrains

  committed
  4a44819

  Copy full SHA for 4a44819

Commits on Jun 3, 2023

• need to detach x

  theAdamColton

  committed
  3c86917

  Copy full SHA for 3c86917

Commits on Jun 1, 2023

• inplace codebook optimization

  theAdamColton

  committed
  ba6f263

  Copy full SHA for ba6f263

Commits on May 30, 2023

• reintroduce orthogonal regularization, due to bug that @kashif found

  lucidrains

  committed
  ae5db01

  Copy full SHA for ae5db01

• fix learnable codebook logic, thanks to @kashif

  lucidrains

  committed
  d23d1fd

  Copy full SHA for d23d1fd

• Merge pull request lucidrains#53 from kashif/patch-1

  Show description for 0b2fe93
  lucidrains

  authored
  0b2fe93

  Copy full SHA for 0b2fe93

• Update vector_quantize_pytorch.py

  Show description for f6ab8df
  kashif

  authored
  f6ab8df

  Copy full SHA for f6ab8df

Commits on May 29, 2023

• address cosine sim distributed lucidrains#51

  lucidrains

  committed
  6fff51a

  Copy full SHA for 6fff51a

Commits on May 28, 2023

• add some quick validations for now

  lucidrains

  committed
  f162fc3

  Copy full SHA for f162fc3

• add the extra hyperparameter v for controlling optimistic vs pessimistic update for the synchronous update rule in https://minyoungg.github.io/vqtorch/assets/draft_050523.pdf

  lucidrains

  committed
  5607471

  Copy full SHA for 5607471

• affine param now works distributed

  lucidrains

  committed
  8e4ef77

  Copy full SHA for 8e4ef77

• need to account for multi-headedness (multiple codebooks). cite einops for immeasurable time saved

  lucidrains

  committed
  9d6a34f

  Copy full SHA for 9d6a34f

Commits on May 27, 2023

• Merge pull request lucidrains#50 from kashif/example

  Show description for 4c090f1
  lucidrains

  authored
  4c090f1

  Copy full SHA for 4c090f1

• remove what reportedly does not work well

  lucidrains

  committed
  936d9be

  Copy full SHA for 936d9be

• buy the conclusions of the new MIT paper with their solution for the internal covariate shift

  lucidrains

  committed
  1c0e73e

  Copy full SHA for 1c0e73e

• fix import

  kashif

  committed
  615747c

  Copy full SHA for 615747c

• fix file name

  kashif

  committed
  793681e

  Copy full SHA for 793681e

• initial autoencoder example

  kashif

  committed
  52ae0d4

  Copy full SHA for 52ae0d4

Commits on May 26, 2023

• fix cosine sim and multi-headed codebooks

  lucidrains

  committed
  e513881

  Copy full SHA for e513881

Commits on May 25, 2023

• if not training, no need for straight through calculation nor stochastic sampling

  lucidrains

  committed
  9001056

  Copy full SHA for 9001056

• temperature should also be applied to logits for sampling, unless if temperature is 0

  lucidrains

  committed
  aac9487

  Copy full SHA for aac9487

• allow for temperature to be annealed

  lucidrains

  committed
  1da641b

  Copy full SHA for 1da641b

• fix a bug with straight-through, thanks to @goldbird5

  lucidrains

  committed
  7c9f1dc

  Copy full SHA for 7c9f1dc

Pagination

Previous

Next