Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on May 22, 2020

• Update Lingvo TF dep to 2.2.

  Show description for ca6f15d
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  ca6f15d

  Copy full SHA for ca6f15d

• Makes inference_graph_export log params line-by-line to avoid truncation.

  Show description for d04a0b1
  

  Ruoming Pang

  authored and

  copybara-github

  committed
  d04a0b1

  Copy full SHA for d04a0b1

Commits on May 21, 2020

• Trainer job may not have a summary_writer.

  Show description for b99c3ab
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  b99c3ab

  Copy full SHA for b99c3ab

• Fix two bugs in reported maximum recall metric.

  Show description for d6356c0
  

  lingvo-bot

  authored and

  copybara-github

  committed
  d6356c0

  Copy full SHA for d6356c0

• Move batch major transformer layers into Lingvo

  Show description for d67cc9b
  

  freeblee

  authored and

  copybara-github

  committed
  d67cc9b

  Copy full SHA for d67cc9b

• Fix text input generator and add a test using wpm tokenizer.

  Show description for b842405
  

  lingvo-bot

  authored and

  copybara-github

  committed
  b842405

  Copy full SHA for b842405

• Remove outside_all_rewrites from TPU input generator flows.

  Show description for a17bb85
  

  freeblee

  authored and

  copybara-github

  committed
  a17bb85

  Copy full SHA for a17bb85

• Use a merger layer to combine source attention. Also move MergerLayer to attention.py to avoid circular dependency.

  Show description for 08ebefb
  

  mrhuke

  authored and

  copybara-github

  committed
  08ebefb

  Copy full SHA for 08ebefb

Commits on May 20, 2020

• Move conv_layers_with_time_padding builder into Lingvo

  Show description for 9e19c1c
  

  freeblee

  authored and

  copybara-github

  committed
  9e19c1c

  Copy full SHA for 9e19c1c

• Allow TransformerAttentionLayer to customize its residual connection.

  Show description for 6bd44b2
  

  lingvo-bot

  authored and

  copybara-github

  committed
  6bd44b2

  Copy full SHA for 6bd44b2

• Clean-up rnn_cell test module using parameterized tests.

  Show description for 5a23365
  

  laurentes

  authored and

  copybara-github

  committed
  5a23365

  Copy full SHA for 5a23365

• Check in CCT Attention layer. This version is slightly different from the original in the paper.

  Show description for ae4883a
  

  Ankur Bapna

  authored and

  copybara-github

  committed
  ae4883a

  Copy full SHA for ae4883a

• Makes CreateTpuFeeds() work for non-2D inputs for use_partitioned_infeed_queue=True.

  Show description for 5ebb1d6
  

  Ruoming Pang

  authored and

  copybara-github

  committed
  5ebb1d6

  Copy full SHA for 5ebb1d6

• Add a GatingLayer.

  Show description for 34f86a7
  

  lingvo-bot

  authored and

  copybara-github

  committed
  34f86a7

  Copy full SHA for 34f86a7

Commits on May 19, 2020

• Add a composite optimizer abstract class to support multiple optimizers applied to different parts of the model.

  Show description for 6c5fbe0
  

  lingvo-bot

  authored and

  copybara-github

  committed
  6c5fbe0

  Copy full SHA for 6c5fbe0

• Move Builder pattern into Lingvo

  Show description for 0dede8e
  

  freeblee

  authored and

  copybara-github

  committed
  0dede8e

  Copy full SHA for 0dede8e

• Speed up executor decoder by placing tpu decode step in a tf while_loop.

  Show description for bf50110
  

  bignamehyp

  authored and

  copybara-github

  committed
  bf50110

  Copy full SHA for bf50110

• Adds name_scope to TransformerAttentionLayer.FProp() to simplify debugging.

  Show description for 4351f46
  

  Ruoming Pang

  authored and

  copybara-github

  committed
  4351f46

  Copy full SHA for 4351f46

• Fix typo: Actually test the dot product attention.

  Show description for a38ef38
  

  lingvo-bot

  authored and

  copybara-github

  committed
  a38ef38

  Copy full SHA for a38ef38

• Fix a proto dependency for tf.io.decode_proto(). Add a simple unit test for TextPackedInput reading a tfrecord.

  Show description for 229a56f
  

  lingvo-bot

  authored and

  copybara-github

  committed
  229a56f

  Copy full SHA for 229a56f

Commits on May 18, 2020

• Allow access to v2 while_loop.

  Show description for ec94891
  

  lingvo-bot

  authored and

  copybara-github

  committed
  ec94891

  Copy full SHA for ec94891

Commits on May 16, 2020

• Move text packed input generator to lingvo.

  Show description for a0b7f4d
  

  lingvo-bot

  authored and

  copybara-github

  committed
  a0b7f4d

  Copy full SHA for a0b7f4d

Commits on May 15, 2020

• Define a transformer layer class where it modifies layers_with_attention.TransformerLayer by adding a third attention layer to be used to attend to a transformer stack encoding context.

  Show description for ff907c8
  

  lingvo-bot

  authored and

  copybara-github

  committed
  ff907c8

  Copy full SHA for ff907c8

• Add publications bibtex source, and move update script to github root.

  Show description for cfc8f21
  

  ronw

  authored and

  copybara-github

  committed
  cfc8f21

  Copy full SHA for cfc8f21

• Add more flexibility to Recall at Precision calculations.

  Show description for 6d5e010
  

  lingvo-bot

  authored and

  copybara-github

  committed
  6d5e010

  Copy full SHA for 6d5e010

• Fix bib entry for Irie, et al paper from Interspeech 2019.

  Show description for 867c60f
  

  ronw

  authored and

  copybara-github

  committed
  867c60f

  Copy full SHA for 867c60f

• Move mass_op to lingvo.

  Show description for 73d0713
  

  lingvo-bot

  authored and

  copybara-github

  committed
  73d0713

  Copy full SHA for 73d0713

Commits on May 14, 2020

• Clean up the breakdown metrics.

  Show description for e6f2961
  

  lingvo-bot

  authored and

  copybara-github

  committed
  e6f2961

  Copy full SHA for e6f2961

• Checkpointer: restore state of previous checkpoint on restart.

  Show description for 75d2423
  

  Vijay Vasudevan

  authored and

  copybara-github

  committed
  75d2423

  Copy full SHA for 75d2423

Commits on May 13, 2020

• Fix disappeared xprof step time.

  Show description for 3e75b11
  

  lingvo-bot

  authored and

  copybara-github

  committed
  3e75b11

  Copy full SHA for 3e75b11

• Add CycleSchedule.

  Show description for 1cba53d
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  1cba53d

  Copy full SHA for 1cba53d

• Add primary source index to specify which attention probabilities will be returned.

  Show description for b3ebe46
  

  mrhuke

  authored and

  copybara-github

  committed
  b3ebe46

  Copy full SHA for b3ebe46

• Move pack_ops to lingvo.

  Show description for a733c97
  

  lingvo-bot

  authored and

  copybara-github

  committed
  a733c97

  Copy full SHA for a733c97

Commits on May 12, 2020

• Increase the number of histogram bins for distance breakdown metric.

  Show description for b369b30
  

  lingvo-bot

  authored and

  copybara-github

  committed
  b369b30

  Copy full SHA for b369b30

• Internal change.

  Show description for cfafec0
  

  lingvo-bot

  authored and

  copybara-github

  committed
  cfafec0

  Copy full SHA for cfafec0

Pagination

Previous

Next