Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Mar 24, 2019

• Refactors LayerNormalizedLSTMCellLean to allow subclasses to override layer normalization.

  Show description for 2cdfab4
  

  Ruoming Pang

  authored and

  copybara-github

  committed
  2cdfab4

  Copy full SHA for 2cdfab4

Commits on Mar 23, 2019

• Avoid manually updating step_state.time_step and delegate to py_utils.GenerateStepSeedPair instead.

  Show description for 6ef4032
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  6ef4032

  Copy full SHA for 6ef4032

Commits on Mar 22, 2019

• Evolved Transformer decoder layer and Swish activation.

  Show description for adc146b
  

  lingvo-bot

  authored and

  copybara-github

  committed
  adc146b

  Copy full SHA for adc146b

• Adding an option to specify what function to use to obtain LSTM initial states (zero_state). Defaults to tf.zeros, so the current behavior will remain unchanged.

  Show description for 31f2ce8
  

  lingvo-bot

  authored and

  copybara-github

  committed
  31f2ce8

  Copy full SHA for 31f2ce8

• Optimize LocationSensitiveAttention for TPU by storing attention state as [batch, 2, seq_len] instead of [batch, seq_len, 2].

  Show description for 2370816
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  2370816

  Copy full SHA for 2370816

• Split py_binary into py_binary and py_library to avoid having py_binary in deps.

  Show description for 7afc7c0
  

  lingvo-bot

  authored and

  copybara-github

  committed
  7afc7c0

  Copy full SHA for 7afc7c0

• Factor out common code into _DecorateCellFn and _DecorateCellGrad.

  Show description for e7e5491
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  e7e5491

  Copy full SHA for e7e5491

Commits on Mar 21, 2019

• In CreateVariableStub, actually create variables (rather than dummy tensors) for the total samples counters so that we can access the `value` method during testing.

  Show description for 567055f
  

  Anjuli Kannan

  authored and

  copybara-github

  committed
  567055f

  Copy full SHA for 567055f

• Add option to set a max on float source inputs (rather than default of 1.0) when generating test data.

  Show description for ebafd13
  

  Anjuli Kannan

  authored and

  copybara-github

  committed
  ebafd13

  Copy full SHA for ebafd13

• Fix bug in PadOrTrimTo with dynamic shapes.

  Show description for a10b557
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  a10b557

  Copy full SHA for a10b557

• Merge pull request #50 from zh794390558:patch-1

  Show description for 551c756
  copybara-github

  committed
  551c756

  Copy full SHA for 551c756

• Qualify calls to some functions from <cmath>.

  Show description for 3b99f08
  

  lingvo-bot

  authored and

  copybara-github

  committed
  3b99f08

  Copy full SHA for 3b99f08

Commits on Mar 20, 2019

• Changes MT v1 decoder's ComputePredictions to return a NestedMap, making attention probs available downstream.

  Show description for aba9fcb
  

  Colin Cherry

  authored and

  copybara-github

  committed
  aba9fcb

  Copy full SHA for aba9fcb

• Changes to select decoder to use to calculate loss based on data source selected.

  Show description for 638107c
  

  lingvo-bot

  authored and

  copybara-github

  committed
  638107c

  Copy full SHA for 638107c

• Update base_layer.py

  zh794390558

  authored
  1d39b44

  Copy full SHA for 1d39b44

• Only set static shape in PadSequenceDim when it's an integer instead of a tensor.

  Show description for aef1748
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  aef1748

  Copy full SHA for aef1748

• Fix a typo.

  Show description for 57143d7
  

  toponado-zz

  authored and

  copybara-github

  committed
  57143d7

  Copy full SHA for 57143d7

Commits on Mar 19, 2019

• unify tensorboard var/grad histograms names with var/grad norm names

  Show description for 1e232db
  

  lingvo-bot

  authored and

  copybara-github

  committed
  1e232db

  Copy full SHA for 1e232db

• Moves record iterator creation out of the mutex.

  Show description for 52ddf54
  

  Zhifeng Chen

  authored and

  copybara-github

  committed
  52ddf54

  Copy full SHA for 52ddf54

• Add more debugging info when the ranks don't match.

  Show description for 5fc4cdd
  

  Vijay Vasudevan

  authored and

  copybara-github

  committed
  5fc4cdd

  Copy full SHA for 5fc4cdd

• Automated rollback of commit ec47ce52972c9eab289fae1f99a35995a87112ba

  Show description for 390beed
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  390beed

  Copy full SHA for 390beed

• Use py_utils.GetShape instead of tf.shape when computing conv output paddings. This will prevent conv layers from returning the output shape in a tensor when it is possible to return in a list of i…

  Show description for a664124
  

  lingvo-bot

  authored and

  copybara-github

  committed
  a664124

  Copy full SHA for a664124

Commits on Mar 18, 2019

• Adds bucket_adjust_every_n.

  Show description for 6c2f02c
  

  lingvo-bot

  authored and

  copybara-github

  committed
  6c2f02c

  Copy full SHA for 6c2f02c

Commits on Mar 15, 2019

• Fixed TPU training loop annotation.

  Show description for 7a7bcb3
  

  lingvo-bot

  authored and

  copybara-github

  committed
  7a7bcb3

  Copy full SHA for 7a7bcb3

• Support None fetches in ConvertSubgraphDictToProto.

  Show description for a289383
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  a289383

  Copy full SHA for a289383

• Support deterministic dropout in MultiSourceFRNNWithAttention.

  Show description for 29348cc
  

  lingvo-bot

  authored and

  copybara-github

  committed
  29348cc

  Copy full SHA for 29348cc

• Add a verbose option when compute WERs for multiple transcripts.

  Show description for 86775b2
  

  lingvo-bot

  authored and

  copybara-github

  committed
  86775b2

  Copy full SHA for 86775b2

Commits on Mar 14, 2019

• Rename GetOpSeedPair->GenerateStepSeedPair.

  Show description for 93f9296
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  93f9296

  Copy full SHA for 93f9296

• Set add_summary=False in inference_graph_exporter.

  Show description for c407ed4
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  c407ed4

  Copy full SHA for c407ed4

• Add a cosines lr schedule with linear warmup.

  Show description for c8ff2e1
  

  bignamehyp

  authored and

  copybara-github

  committed
  c8ff2e1

  Copy full SHA for c8ff2e1

• Set xla_device = 'tpu' when exporting inference graphs for tpu.

  Show description for 14a52b8
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  14a52b8

  Copy full SHA for 14a52b8

• Merge pull request #30 from zh794390558:patch-3

  Show description for 0732063
  copybara-github

  committed
  0732063

  Copy full SHA for 0732063

Commits on Mar 13, 2019

• Remove trailing spaces.

  jonathanasdf

  authored
  e6ed781

  Copy full SHA for e6ed781

• Do not write disabled_ summaries for better TPU compatibility.

  Show description for 11e11ab
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  11e11ab

  Copy full SHA for 11e11ab

• Qualify calls to certain functions from the cmath library.

  Show description for 4f8a8c9
  

  lingvo-bot

  authored and

  copybara-github

  committed
  4f8a8c9

  Copy full SHA for 4f8a8c9

Pagination

Previous

Next