Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Apr 29, 2019

• Also an option to prune projection matrix in LSTM.

  Show description for 816c21d
  

  lingvo-bot

  authored and

  copybara-github

  committed
  816c21d

  Copy full SHA for 816c21d

• n/a

  Show description for 6a52356
  

  lingvo-bot

  authored and

  copybara-github

  committed
  6a52356

  Copy full SHA for 6a52356

• Replace deprecated FastGFile call with GFile.

  Show description for 1ad5436
  

  lingvo-bot

  authored and

  copybara-github

  committed
  1ad5436

  Copy full SHA for 1ad5436

Commits on Apr 26, 2019

• Enable pruning for SRUCell.

  Show description for c926b4e
  

  lingvo-bot

  authored and

  copybara-github

  committed
  c926b4e

  Copy full SHA for c926b4e

• Add

  Show description for 3b98146
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  3b98146

  Copy full SHA for 3b98146

• Set shapes to weights (theta) that were passed to recurrent Fwd/Bak functions.

  Show description for 405dd47
  

  bignamehyp

  authored and

  copybara-github

  committed
  405dd47

  Copy full SHA for 405dd47

Commits on Apr 25, 2019

• Include metric name in logging message for easier debugging.

  Show description for 7a55484
  

  lingvo-bot

  authored and

  copybara-github

  committed
  7a55484

  Copy full SHA for 7a55484

• Adds 'parent' to BaseLayer.

  Show description for b9e2023
  

  Ruoming Pang

  authored and

  copybara-github

  committed
  b9e2023

  Copy full SHA for b9e2023

• Remove remaining references to step_state, step_seed, or global_step.

  Show description for ecb7771
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  ecb7771

  Copy full SHA for ecb7771

Commits on Apr 24, 2019

• Fix accumulator incorrectly being initialized from disabled state in recurrent (and add to the test to ensure that initial state carries).

  Show description for 1237db0
  

  Stella Laurenzo

  authored and

  copybara-github

  committed
  1237db0

  Copy full SHA for 1237db0

• Need to apply mask in FProp so that the inference graph can be generated correctly with masked weights when pruning is used in training.

  Show description for bf766f4
  

  lingvo-bot

  authored and

  copybara-github

  committed
  bf766f4

  Copy full SHA for bf766f4

• Check to see that a value is present in the summary before accessing.

  Show description for ae5c04f
  

  Vijay Vasudevan

  authored and

  copybara-github

  committed
  ae5c04f

  Copy full SHA for ae5c04f

Commits on Apr 23, 2019

• Allow input_device to be specified for non-trainer jobs.

  Show description for dcb6d03
  

  Vijay Vasudevan

  authored and

  copybara-github

  committed
  dcb6d03

  Copy full SHA for dcb6d03

• Update required bazel_version to 0.24.1 for the python_version attribute of python rules.

  Show description for e2ec7e3
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  e2ec7e3

  Copy full SHA for e2ec7e3

• n/a

  Show description for 4d49de6
  

  lingvo-bot

  authored and

  copybara-github

  committed
  4d49de6

  Copy full SHA for 4d49de6

Commits on Apr 22, 2019

• In AUCMetric also draws the curve as a tf image summary.

  Show description for ed29100
  

  lingvo-bot

  authored and

  copybara-github

  committed
  ed29100

  Copy full SHA for ed29100

• Make GenerateStepSeedPair more explicit by requiring a passed in global_step.

  Show description for e9785ed
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  e9785ed

  Copy full SHA for e9785ed

Commits on Apr 21, 2019

• Remove global_step arg from Attention functions and use theta.global_step

  Show description for 9f604b0
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  9f604b0

  Copy full SHA for 9f604b0

• Use CreateChild to create child layers. AddChild should only add an existing

  Show description for 82bba07
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  82bba07

  Copy full SHA for 82bba07

• Remove step_counter in Attention layers and use GenerateStepSeedPair.

  Show description for a0f7590
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  a0f7590

  Copy full SHA for a0f7590

Commits on Apr 20, 2019

• [GPipe] Handle cases when num_layers is not divisible by num_splits

  Show description for f8cb772
  

  lingvo-bot

  authored and

  copybara-github

  committed
  f8cb772

  Copy full SHA for f8cb772

Commits on Apr 19, 2019

• check for undefined tensor when exporting multi-headed attention summary

  Show description for b84918c
  

  lingvo-bot

  authored and

  copybara-github

  committed
  b84918c

  Copy full SHA for b84918c

• Add global_step to thetas for all layers.

  Show description for bf3e3fa
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  bf3e3fa

  Copy full SHA for bf3e3fa

• Minor fixes to BUILD rules.

  Show description for 9416f3d
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  9416f3d

  Copy full SHA for 9416f3d

Commits on Apr 18, 2019

• Expose more hparams of TransformerAttentionLayer in base_config.py

  Show description for 53a9d09
  

  orhanf

  authored and

  copybara-github

  committed
  53a9d09

  Copy full SHA for 53a9d09

• Only cast to fprop_dtype tensors that match the original dtype.

  Show description for dc951f6
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  dc951f6

  Copy full SHA for dc951f6

• Add a utility function that forces to rematerialize the computation as

  Show description for 65f9a85
  

  Yonghui Wu

  authored and

  copybara-github

  committed
  65f9a85

  Copy full SHA for 65f9a85

Commits on Apr 17, 2019

• Global step summary name should be global_step/sec.

  Show description for f51e278
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  f51e278

  Copy full SHA for f51e278

• Add a GlobalStepContext under which the same global step tensor (not variable)

  Show description for b9445a3
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  b9445a3

  Copy full SHA for b9445a3

• Fix typo in _BeamSearchStep docstring.

  Show description for f4493ce
  

  lingvo-bot

  authored and

  copybara-github

  committed
  f4493ce

  Copy full SHA for f4493ce

• Update run_locally docstring.

  Show description for 207aae0
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  207aae0

  Copy full SHA for 207aae0

• Split the implementation of GetOrCreateGlobalStep into GetGlobalStep and

  Show description for 74fca55
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  74fca55

  Copy full SHA for 74fca55

Commits on Apr 16, 2019

• Add a test_utils.TestCase subclass that ensures global_step variable is created

  Show description for 0f1edb2
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  0f1edb2

  Copy full SHA for 0f1edb2

• Certain non-tensorflow flags should not be reset during API doc generation to

  Show description for 3e4582a
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  3e4582a

  Copy full SHA for 3e4582a

• Rename GetOrCreateGlobalStep to GetGlobalStep.

  Show description for 38a8a0c
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  38a8a0c

  Copy full SHA for 38a8a0c

Pagination

Previous

Next