Commits

Branch selector

 v1.2.6

User selector

All users

All time

Commit History

Commits on Oct 27, 2017

• Merge pull request #383 from rsepassi/push

  Show description for 9bdc801
  lukaszkaiser

  authored
  9bdc801

  Copy full SHA for 9bdc801

• v1.2.6

  Show description for ba47b61
  

  Ryan Sepassi

  committed
  ba47b61

  Copy full SHA for ba47b61

• Add back commented-out enfr datasets

  Show description for 9d86cf7
  

  Ryan Sepassi

  committed
  9d86cf7

  Copy full SHA for 9d86cf7

• Internal merge PR#370

  Show description for 90aa796
  

  Ryan Sepassi

  committed
  90aa796

  Copy full SHA for 90aa796

Commits on Oct 26, 2017

• Fast beam search decoding.

  Show description for 86703a2
  

  T2T Team

  authored and

  Ryan Sepassi

  committed
  86703a2

  Copy full SHA for 86703a2

• Remove debug lines from decoding.py that got in by mistake in cl/173421571.

  Show description for 2245033
  

  T2T Team

  authored and

  Ryan Sepassi

  committed
  2245033

  Copy full SHA for 2245033

• Decoding corrections for no input, when calling decode_from_file.

  Show description for b840c0c
  

  T2T Team

  authored and

  Ryan Sepassi

  committed
  b840c0c

  Copy full SHA for b840c0c

• Extra check in SubwordTextEncoder decoding for empty tokens.

  Show description for ba75966
  

  nshazeer

  authored and

  Ryan Sepassi

  committed
  ba75966

  Copy full SHA for ba75966

• Decoding corrections for problems with no inputs.

  Show description for e0100e8
  

  Lukasz Kaiser

  authored and

  Ryan Sepassi

  committed
  e0100e8

  Copy full SHA for e0100e8

• Stop inference after predicting EOS when batch_size=1

  Show description for 480ee06
  

  Etienne Pot

  authored and

  Ryan Sepassi

  committed
  480ee06

  Copy full SHA for 480ee06

• Modify tpu_trainer_lib to make text autoregressive models work on TPUs.

  Show description for e9d61f5
  

  T2T Team

  authored and

  Ryan Sepassi

  committed
  e9d61f5

  Copy full SHA for e9d61f5

• Add transformer_prepend_v1 hparams for backwards compatibility

  Show description for ec820d3
  

  T2T Team

  authored and

  Ryan Sepassi

  committed
  ec820d3

  Copy full SHA for ec820d3

• Ensure shard 0 is read correctly in parallel decoding

  Show description for 4b28166
  

  T2T Team

  authored and

  Ryan Sepassi

  committed
  4b28166

  Copy full SHA for 4b28166

• Ensure that training with "continuous_train_and_eval" schedule uses local devices.

  Show description for 3b11bbf
  

  T2T Team

  authored and

  Ryan Sepassi

  committed
  3b11bbf

  Copy full SHA for 3b11bbf

• Fix visualizations.

  Show description for 8797199
  

  T2T Team

  authored and

  Ryan Sepassi

  committed
  8797199

  Copy full SHA for 8797199

• Fix to SymbolModality to allow weight-sharing between target

  Show description for e4aa5f2
  

  nshazeer

  authored and

  Ryan Sepassi

  committed
  e4aa5f2

  Copy full SHA for e4aa5f2

• Update @recompute_grad to respect tf.contrib.framework.arg_scope

  Show description for 516a369
  

  Ryan Sepassi

  committed
  516a369

  Copy full SHA for 516a369

• Typo, minor hparams changes for the attention_lm_moe

  Show description for 30df3e9
  

  Etienne Pot

  authored and

  Ryan Sepassi

  committed
  30df3e9

  Copy full SHA for 30df3e9

• Allow parallel decoding for decode_from_dataset

  Show description for 1c51666
  

  Niki Parmar

  authored and

  Ryan Sepassi

  committed
  1c51666

  Copy full SHA for 1c51666

• Fix TPU input ranks

  Show description for 7d6348e
  

  Ryan Sepassi

  committed
  7d6348e

  Copy full SHA for 7d6348e

• Ensure shapes of inputs are fully defined in tpu trainer

  Show description for 4d81643
  

  Ryan Sepassi

  committed
  4d81643

  Copy full SHA for 4d81643

• Add ImageEncoder that allows to operate on images (only saving for now).

  Show description for 62aba9d
  

  Lukasz Kaiser

  authored and

  Ryan Sepassi

  committed
  62aba9d

  Copy full SHA for 62aba9d

• A MultiNLI classification problem for Tensor2tensor.

  Show description for 68023da
  

  T2T Team

  authored and

  Ryan Sepassi

  committed
  68023da

  Copy full SHA for 68023da

• Added a cache to improve performance of SubwordTextEncoder.

  Show description for 3f1a3f0
  

  nshazeer

  authored and

  Ryan Sepassi

  committed
  3f1a3f0

  Copy full SHA for 3f1a3f0

• Code style improvements in CNN/DailyMail generator.

  Show description for 41e0bfb
  

  Lukasz Kaiser

  authored and

  Ryan Sepassi

  committed
  41e0bfb

  Copy full SHA for 41e0bfb

Commits on Oct 25, 2017

• Merge pull request #370 from vince62s/translate

  Show description for a836d66
  lukaszkaiser

  authored
  a836d66

  Copy full SHA for a836d66

Commits on Oct 24, 2017

• fix

  vince62s

  committed
  b43f833

  Copy full SHA for b43f833

Commits on Oct 22, 2017

• adjust vocab with random lines

  vince62s

  committed
  9f59a50

  Copy full SHA for 9f59a50

Commits on Oct 19, 2017

• another warning fix

  vince62s

  committed
  fc35144

  Copy full SHA for fc35144

• fix warning

  vince62s

  committed
  dd08f9d

  Copy full SHA for dd08f9d

• Rename wmt.py to translate.py

  Show description for eacde9d
  vince62s

  committed
  eacde9d

  Copy full SHA for eacde9d

Commits on Oct 17, 2017

• Merge pull request #363 from urvashik/master

  Show description for 0ecbef8
  lukaszkaiser

  authored
  0ecbef8

  Copy full SHA for 0ecbef8

Commits on Oct 16, 2017

• Merge branch 'master' into master

  urvashik

  authored
  f084b5b

  Copy full SHA for f084b5b

• fixing encoding issues on cnn/dailymail (#1)

  

  pltrdy

  authored and

  urvashik

  committed
  a18541a

  Copy full SHA for a18541a

• Merge pull request #361 from rsepassi/push

  Show description for 3c5823f
  lukaszkaiser

  authored
  3c5823f

  Copy full SHA for 3c5823f