Commits

Branch selector

 master

User selector

All users

All time

Commit History

Commits on Dec 15, 2017

• Spelling fix: 'fo' => 'to' (#471)

  

  cbockman

  authored and

  rsepassi

  committed
  9a934cc

  Copy full SHA for 9a934cc

Commits on Dec 4, 2017

• Merge pull request #456 from lukaszkaiser/push

  Show description for d9f807c
  lukaszkaiser

  authored
  d9f807c

  Copy full SHA for d9f807c

• Release 1.3.2 with colab improvements.

  Show description for f2fb96b
  

  Lukasz Kaiser

  committed
  f2fb96b

  Copy full SHA for f2fb96b

• add translate_enfr_wmt32k_packed problem.

  Show description for b8aa4ea
  

  nshazeer

  authored and

  lukaszkaiser

  committed
  b8aa4ea

  Copy full SHA for b8aa4ea

• Store attention-weight tensors as part of Transformer model class for easier access when vizualizing.

  Show description for 9a8f203
  

  Lukasz Kaiser

  committed
  9a8f203

  Copy full SHA for 9a8f203

• Update colab notebook to use v1.3.1

  Show description for d517a62
  

  Ryan Sepassi

  authored and

  lukaszkaiser

  committed
  d517a62

  Copy full SHA for d517a62

Commits on Dec 1, 2017

• Merge pull request #451 from rsepassi/push

  Show description for 970dac9
  lukaszkaiser

  authored
  970dac9

  Copy full SHA for 970dac9

• TF Eager improvements for T2TModel

  Show description for 889fc84
  

  Ryan Sepassi

  committed
  889fc84

  Copy full SHA for 889fc84

• Fix decoding and training issues in external colab.

  Show description for 654f74e
  

  Niki Parmar

  authored and

  Ryan Sepassi

  committed
  654f74e

  Copy full SHA for 654f74e

• Enable Transformer fast decoding in eager mode

  Show description for e133a1a
  

  Niki Parmar

  authored and

  Ryan Sepassi

  committed
  e133a1a

  Copy full SHA for e133a1a

• New BLEU cleanup and small correction to VAE.

  Show description for c93a188
  

  Lukasz Kaiser

  authored and

  Ryan Sepassi

  committed
  c93a188

  Copy full SHA for c93a188

• v1.3.1

  Show description for b1abcf4
  

  Ryan Sepassi

  committed
  b1abcf4

  Copy full SHA for b1abcf4

• T2T depends on TF 1.4+, daisy_chain_getter bug fix, some Eager-mode improvements/fixes

  Show description for aa2c0b7
  

  Ryan Sepassi

  committed
  aa2c0b7

  Copy full SHA for aa2c0b7

• Clean up transformer_vae and add refining.

  Show description for 24c1fd7
  

  Lukasz Kaiser

  authored and

  Ryan Sepassi

  committed
  24c1fd7

  Copy full SHA for 24c1fd7

• Remove TranslateEndeWmtBpe32kPacked. We have TranslateEndeWmt32kPacked.

  Show description for 01030eb
  

  nshazeer

  authored and

  Ryan Sepassi

  committed
  01030eb

  Copy full SHA for 01030eb

• Make Parallelism object use reuse=True by default. Solves tpu checkpoint compatibility bug.

  Show description for 7f3ef1e
  

  nshazeer

  authored and

  Ryan Sepassi

  committed
  7f3ef1e

  Copy full SHA for 7f3ef1e

• Packed datasets - combine examples to constant length for efficient TPU training. Modify transformer to keep the packed-together examples from attending to one another.

  Show description for c9144df
  

  nshazeer

  authored and

  Ryan Sepassi

  committed
  c9144df

  Copy full SHA for c9144df

• Discrete autoencoder with VQ-VAE as in https://arxiv.org/abs/1711.00937.

  Show description for 20c7e41
  

  T2T Team

  authored and

  Ryan Sepassi

  committed
  20c7e41

  Copy full SHA for 20c7e41

• Merge pull request #436 from martinpopel/bleu

  Show description for bb1173a
  lukaszkaiser

  authored
  bb1173a

  Copy full SHA for bb1173a

Commits on Nov 30, 2017

• if no reference or translation is provided fail

  Show description for 7ba78a2
  martinpopel

  committed
  7ba78a2

  Copy full SHA for 7ba78a2

• add t2t-bleu to setup.py

  martinpopel

  committed
  004888a

  Copy full SHA for 004888a

• first try a new checkpoint before checking the time elapsed

  Show description for 7fc50b9
  martinpopel

  committed
  7fc50b9

  Copy full SHA for 7fc50b9

• more options for t2t-bleu

  Show description for a43eff3
  martinpopel

  committed
  a43eff3

  Copy full SHA for a43eff3

• fix tests for Python2

  martinpopel

  committed
  1583032

  Copy full SHA for 1583032

• fix tests because of the added smoothing in BLEU

  martinpopel

  committed
  b912213

  Copy full SHA for b912213

• add t2t-bleu script for proper BLEU evaluation

  martinpopel

  committed
  aade7ec

  Copy full SHA for aade7ec

• allow t2t-decoder to load any checkpoint

  Show description for 685c9fc
  martinpopel

  committed
  685c9fc

  Copy full SHA for 685c9fc

• add BLEU-compatible tokenization

  Show description for 4160979
  martinpopel

  committed
  4160979

  Copy full SHA for 4160979

• add BLEU smoothing

  Show description for 86ecec0
  martinpopel

  committed
  86ecec0

  Copy full SHA for 86ecec0

Commits on Nov 29, 2017

• Merge pull request #449 from rsepassi/push

  Show description for e3cd447
  lukaszkaiser

  authored
  e3cd447

  Copy full SHA for e3cd447

• v1.3

  Show description for 69701e4
  

  Ryan Sepassi

  committed
  69701e4

  Copy full SHA for 69701e4

• Disable image summary in eager mode (tf.Eager doesn't like it for now).

  Show description for c65a646
  

  Lukasz Kaiser

  authored and

  Ryan Sepassi

  committed
  c65a646

  Copy full SHA for c65a646

• fix bug with bpe32k - prepend <pad> and <EOS> to vocab.

  Show description for 7909c69
  

  nshazeer

  authored and

  Ryan Sepassi

  committed
  7909c69

  Copy full SHA for 7909c69

• Make default Parallelism in T2TModel reusing (removes "parallel_0" from variable names in colab).

  Show description for 2ca3232
  

  Lukasz Kaiser

  authored and

  Ryan Sepassi

  committed
  2ca3232

  Copy full SHA for 2ca3232

• Correction for eager-mode decoding scopes (to use with pre-trained checkpoints).

  Show description for 8f05bab
  

  Lukasz Kaiser

  authored and

  Ryan Sepassi

  committed
  8f05bab

  Copy full SHA for 8f05bab