Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Dec 28, 2017

• Add option to use SampleEmbeddingHelper with a temperature.

  oahziur

  committed
  365e738

• Add num_layers back to hparams for compatible reasons

  

  lmthang

  authored and

  oahziur

  committed
  e6ecab2

• Add initial value for info["avg_grad_norm"] so print_step_info in train.py

  

  lmthang

  authored and

  oahziur

  committed
  ecc35a5

• Fix printing of avg_step_time, speed, and summary of train_ppl. Add info dict and method before_train() to simplify code.

  

  lmthang

  authored and

  oahziur

  committed
  63352d7

• Add back decay_scheme luong5

  

  lmthang

  authored and

  oahziur

  committed
  b2a1647

• Smartly place embeddings on devices (cpu if vocab size >

  

  lmthang

  authored and

  oahziur

  committed
  8251bd2

• Add option to avg checkpoints for external evaluation. (Inspired by Tensor2Tensor)

  oahziur

  committed
  bf760a6

• Allow GNMT encoder to pair with non-GNMT attention (i.e., encoder_type=gnmt

  

  lmthang

  authored and

  oahziur

  committed
  1899199

• Add num_encoder_layers / num_decoder_layers flags. Update logics on num

  

  lmthang

  authored and

  oahziur

  committed
  2f16efd

• Add tests for nmt.py.

  oahziur

  committed
  e9218aa

• Fix typo

  

  Meng-Gen

  authored and

  oahziur

  committed
  34a4368

Commits on Dec 4, 2017

• add command line option to input the num_inter_threads and num_intra_threads to control threads of session

  

  cuixiaom

  authored and

  oahziur

  committed
  20dff79

• change embed_file variable name

  

  Germey

  authored and

  oahziur

  committed
  bd2c626

• Add 'num_keep_ckpts' to hparams. Defaults to 5.

  oahziur

  committed
  f83bfd6

• fix moses bleu SPM

  

  Joost Bastings

  authored and

  oahziur

  committed
  616bd66

• Only clean training data

  

  Joost Bastings

  authored and

  oahziur

  committed
  c196cd5

• Correct other spelling issue

  

  nave01314

  authored and

  oahziur

  committed
  9be6794

• Remove flags start_decay_step, decay_steps, decay_factor, source_reverse.

  

  lmthang

  authored and

  oahziur

  committed
  9ffc354

Commits on Nov 27, 2017

• Revert _sample_decode print_out fixes for issue tensorflow#175 and issue tensorflow#169, we have already done the utf-8 converting in the print_out method, so the error seems come from other sour…

  oahziur

  committed
  82692af

• Add options to load pre-trained embeddings in txt (glove) format.

  oahziur

  committed
  d65d5e1

• Make sure the final decoded ids shape is [beam_width, batch_size, max_time] when

  oahziur

  committed
  1f8f4d8

Commits on Nov 20, 2017

• Update _sample_decode print_out to resolve potential python3 unicode issue.

  oahziur

  committed
  58abf51

• Add output_attention flag to control standard attention wrapper's behavior.

  oahziur

  committed
  f6846ff

• Factor out code snippets that initialize, accumulate, and reset statistics

  

  lmthang

  authored and

  oahziur

  committed
  3cdfbfc

• Add luong10 decay scheme, same as luong but decay 10 times

  

  lmthang

  authored and

  oahziur

  committed
  5ed775b

• Update stable branch link to tensorflow 1.4.

  oahziur

  committed
  2b7ca76

• Add header=None field to misc_utils.print_hparams

  

  lmthang

  authored and

  oahziur

  committed
  087483b

Commits on Nov 2, 2017

• Add github link to README

  oahziur

  committed
  05f8cc1

• Cleanup bpe_delimiter as it is relpaced by subword_options.

  oahziur

  committed
  192b0ec

• Update README.md

  

  v1n337

  authored and

  oahziur

  committed
  72b19b4

• Adding word_accuracy evaluation metric.

  

  Anonymous

  authored and

  oahziur

  committed
  4023608

Commits on Oct 26, 2017

• Add Metadata about the tutorial.

  oahziur

  committed
  2b9fb37

• Merge pull request tensorflow#153: updated to tf 1.4 API

  

  kashif

  authored and

  oahziur

  committed
  1a7f1a6

Commits on Oct 25, 2017

• Add warmup_scheme to replace warmup_factor.

  

  lmthang

  authored and

  oahziur

  committed
  cff5291

• Add rnn.NASCell (through --unit_type=nas); set decay factor to 1.0 to fix a bug in learning rate decay when start_decay_step is not set.

  

  lmthang

  authored and

  oahziur

  committed
  d6d7c46

Pagination

PreviousNext