Commits

Breadcrumbs

History for

onmaster

User selector

All users

All time

Commit History

Commits on Dec 1, 2017

• 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

Commits on Nov 30, 2017

• allow t2t-decoder to load any checkpoint

  Show description for 685c9fc
  martinpopel

  committed
  685c9fc

  Copy full SHA for 685c9fc

Commits on Nov 29, 2017

• Make daisy_chain_variables and hparam instead of flag and unset to allow LSTM to train in distributed mode.

  Show description for 5adacd0
  

  Lukasz Kaiser

  authored and

  Ryan Sepassi

  committed
  5adacd0

  Copy full SHA for 5adacd0

• merge PRs.

  Show description for 6d9b5e1
  

  Lukasz Kaiser

  authored and

  Ryan Sepassi

  committed
  6d9b5e1

  Copy full SHA for 6d9b5e1

• More tiny sketch fixes

  Show description for 6cf47f9
  

  Ryan Sepassi

  committed
  6cf47f9

  Copy full SHA for 6cf47f9

Commits on Nov 23, 2017

• Merge pull request #422 from urvashik/master

  Show description for c25e43f
  lukaszkaiser

  authored
  c25e43f

  Copy full SHA for c25e43f

Commits on Nov 17, 2017

• fix the semantics of decode_to_file

  Show description for 985b637
  martinpopel

  committed
  985b637

  Copy full SHA for 985b637

Commits on Nov 14, 2017

• Raw text data generation for train, dev and test; script to compute rouge using the official pyrouge dist; a bash script to tokenize targets and predictions before computing rouge

  urvashik

  committed
  3f52fb9

  Copy full SHA for 3f52fb9

Commits on Nov 11, 2017

• Get rid of "last_position_only" by adding the corresponding property to Modality.

  Show description for fa46070
  

  Lukasz Kaiser

  authored and

  Ryan Sepassi

  committed
  fa46070

  Copy full SHA for fa46070

• Add CTC loss SymbolModality, correct get_bleu script, LSTM without inputs, small en-fr generator, play with adversarial transformer.

  Show description for e1bde97
  

  Lukasz Kaiser

  authored and

  Ryan Sepassi

  committed
  e1bde97

  Copy full SHA for e1bde97

• Minor changes to decoding to enable reporting of full beams and scores:

  Show description for b0f580c
  

  nshazeer

  authored and

  Ryan Sepassi

  committed
  b0f580c

  Copy full SHA for b0f580c

• Enable Xception and ImageNet on TPU

  Show description for ca22374
  

  Ryan Sepassi

  committed
  ca22374

  Copy full SHA for ca22374

Commits on Nov 4, 2017

• fix decode_from_file

  vince62s

  committed
  16396e0

  Copy full SHA for 16396e0

Commits on Oct 29, 2017

• Rouge pipeline complete

  urvashik

  committed
  f711de9

  Copy full SHA for f711de9

Commits on Oct 26, 2017

• 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

• Decoding corrections for problems with no inputs.

  Show description for e0100e8
  

  Lukasz Kaiser

  authored and

  Ryan Sepassi

  committed
  e0100e8

  Copy full SHA for e0100e8

• Allow parallel decoding for decode_from_dataset

  Show description for 1c51666
  

  Niki Parmar

  authored and

  Ryan Sepassi

  committed
  1c51666

  Copy full SHA for 1c51666

• 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

Commits on Oct 16, 2017

• Add support for custom record delimiter in decoding

  Show description for 545ec34
  

  Ryan Sepassi

  committed
  545ec34

  Copy full SHA for 545ec34

• Fix formatting issue when decoding the results

  Show description for d3faf90
  

  Etienne Pot

  authored and

  Ryan Sepassi

  committed
  d3faf90

  Copy full SHA for d3faf90

Commits on Sep 30, 2017

• 1d Dilated masked and unmasked self-attention. Added spaces between tokens for logging during inference.

  Show description for ed7862c
  

  Ashish Vaswani

  authored and

  Ryan Sepassi

  committed
  ed7862c

  Copy full SHA for ed7862c

Commits on Sep 29, 2017

• input_pipeline uses Problem.dataset

  Show description for c6710dd
  

  Ryan Sepassi

  committed
  c6710dd

  Copy full SHA for c6710dd

• Support using "-test" data for EVAL/PREDICT with --eval_use_test_set flag

  Show description for 41b7c70
  

  Ryan Sepassi

  committed
  41b7c70

  Copy full SHA for 41b7c70

Commits on Sep 22, 2017

• Correct README for decoding

  Show description for f191c78
  

  Ryan Sepassi

  committed
  f191c78

  Copy full SHA for f191c78

Commits on Sep 21, 2017

• Fix formatting in identity output

  Show description for bc191b5
  

  Niki Parmar

  authored and

  Ryan Sepassi

  committed
  bc191b5

  Copy full SHA for bc191b5

• Use decode_hparams.batch_size when decoding from dataset

  Show description for 3237538
  

  Etienne Pot

  authored and

  Ryan Sepassi

  committed
  3237538

  Copy full SHA for 3237538

• fix off-by-one num_samples bug in decode_from_dataset

  Show description for 2138599
  

  Etienne Pot

  authored and

  Ryan Sepassi

  committed
  2138599

  Copy full SHA for 2138599

Commits on Sep 10, 2017

• Refactor and move decode flags into --decode_hparams

  Show description for 42f089b
  

  Ryan Sepassi

  committed
  42f089b

  Copy full SHA for 42f089b

Commits on Sep 8, 2017

• Move to core Estimator and improve decoding

  Show description for 2ebead2
  

  Ryan Sepassi

  committed
  2ebead2

  Copy full SHA for 2ebead2

Commits on Aug 29, 2017

• Fix decode_from_dataset so that it decodes from multiple batches again

  Show description for ee3296f
  

  Ryan Sepassi

  committed
  ee3296f

  Copy full SHA for ee3296f

Commits on Aug 25, 2017

• change decode_from_dataset to write outside the loop and extra cond for decode_from_file

  Show description for a237064
  

  Niki Parmar

  authored and

  Ryan Sepassi

  committed
  a237064

  Copy full SHA for a237064

Commits on Aug 17, 2017

• Typo and VAE improvements.

  Show description for fd8a222
  

  Lukasz Kaiser

  authored and

  Ryan Sepassi

  committed
  fd8a222

  Copy full SHA for fd8a222

Commits on Aug 11, 2017

• Rename train_generator to just generator and port wiki_32k to Problem. Also cleaning and speeding up vocab generation, algorithmic problems, wmt_zhen and BPE download.

  Show description for 94eca0c
  

  Lukasz Kaiser

  authored and

  Ryan Sepassi

  committed
  94eca0c

  Copy full SHA for 94eca0c

• Add TransformerEncoder and TransformerDecoder models

  Show description for d30ec6b
  

  Ryan Sepassi

  committed
  d30ec6b

  Copy full SHA for d30ec6b