Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Dec 9, 2022

• tf 2.9.2 works

  yejingxin

  committed
  df8e417

  Copy full SHA for df8e417

Commits on Dec 5, 2022

• Refactor so that every extractor has a RightContext() API (just like the Stride() API), and GenericEncoder should call this RightContext() API

  Show description for c35badc
  

  lingvo-bot

  authored and

  copybara-github

  committed
  c35badc

  Copy full SHA for c35badc

• Fix MultitaskAdapterEinsumLayer crash.

  Show description for 0968499
  

  ds-hwang

  authored and

  copybara-github

  committed
  0968499

  Copy full SHA for 0968499

Commits on Dec 3, 2022

• MoE lifelong pretraining: LM sparse model

  Show description for d2d88f1
  

  lingvo-bot

  authored and

  copybara-github

  committed
  d2d88f1

  Copy full SHA for d2d88f1

Commits on Dec 2, 2022

• Disable VN in eval mode.

  Show description for 0e212c2
  

  ds-hwang

  authored and

  copybara-github

  committed
  0e212c2

  Copy full SHA for 0e212c2

• Remove references to deprecated 'sklearn'; published as lingvo 0.12.6.

  Show description for 81fd983
  

  lingvo-bot

  authored and

  copybara-github

  committed
  81fd983

  Copy full SHA for 81fd983

• Clean up ConformerLayer.adapter_tpl.

  Show description for 33bd750
  

  ds-hwang

  authored and

  copybara-github

  committed
  33bd750

  Copy full SHA for 33bd750

Commits on Dec 1, 2022

• Cache GetAllDatasetParams.

  Show description for 969916d
  

  ds-hwang

  authored and

  copybara-github

  committed
  969916d

  Copy full SHA for 969916d

Commits on Nov 30, 2022

• gshard builder and layers for MoE lifelong pretraining

  Show description for 794110e
  

  lingvo-bot

  authored and

  copybara-github

  committed
  794110e

  Copy full SHA for 794110e

• Loosen test tolerances due to BF16 GPU being enabled

  Show description for 9445632
  

  freeblee

  authored and

  copybara-github

  committed
  9445632

  Copy full SHA for 9445632

• Encode dicts of params into text.

  Show description for 14b035d
  

  lingvo-bot

  authored and

  copybara-github

  committed
  14b035d

  Copy full SHA for 14b035d

Commits on Nov 29, 2022

• Add various options for adapters.

  Show description for be05a35
  

  Qiujia Li

  authored and

  copybara-github

  committed
  be05a35

  Copy full SHA for be05a35

• Fix ConformerLayer's adapter_tpl for StreamStep

  Show description for 786e981
  

  ds-hwang

  authored and

  copybara-github

  committed
  786e981

  Copy full SHA for 786e981

• Refactor the way large initial numbers are used so that they don't cause overflow.

  Show description for 974f60f
  

  lingvo-bot

  authored and

  copybara-github

  committed
  974f60f

  Copy full SHA for 974f60f

• Add tests for the case the model whose datasets are defined by GetAllDatasetParams and public methods.

  Show description for 0f8caa2
  

  ds-hwang

  authored and

  copybara-github

  committed
  0f8caa2

  Copy full SHA for 0f8caa2

Commits on Nov 28, 2022

• Add a NestedMap compare using multi-line string compare for unittests.

  Show description for 0e43ced
  

  lingvo-bot

  authored and

  copybara-github

  committed
  0e43ced

  Copy full SHA for 0e43ced

Commits on Nov 24, 2022

• Add EmaDecaySchedule.

  Show description for 3e19cc5
  

  ds-hwang

  authored and

  copybara-github

  committed
  3e19cc5

  Copy full SHA for 3e19cc5

Commits on Nov 22, 2022

• Add a 'keep' parameter to the function CopyFieldsTo in hyperparams.py

  Show description for 523ff4a
  

  tonybruguier

  authored and

  copybara-github

  committed
  523ff4a

  Copy full SHA for 523ff4a

Commits on Nov 21, 2022

• Implement EMA schedule

  Show description for 8ca46ae
  

  ds-hwang

  authored and

  copybara-github

  committed
  8ca46ae

  Copy full SHA for 8ca46ae

Commits on Nov 17, 2022

• Breaks internal targets

  Show description for 4c0f0c2
  

  lingvo-bot

  authored and

  copybara-github

  committed
  4c0f0c2

  Copy full SHA for 4c0f0c2

• Fix seemingly broken padding semantics in praxis/convolutions.py and lingvo/conv_layers_with_time_padding.py

  Show description for 3c3df53
  

  lingvo-bot

  authored and

  copybara-github

  committed
  3c3df53

  Copy full SHA for 3c3df53

• Use tf.tensordot for projecting the last dimension

  Show description for 1f1735b
  

  lingvo-bot

  authored and

  copybara-github

  committed
  1f1735b

  Copy full SHA for 1f1735b

• `NamedSharding` is the replacement **name** for `MeshPspecSharding`.

  Show description for cb46259
  

  lingvo-bot

  authored and

  copybara-github

  committed
  cb46259

  Copy full SHA for cb46259

Commits on Nov 16, 2022

• Add input Reset in ExperimentalDecodeProgram at "end" of decoding infeed. This seems release some memory since currently after switching away from decode programs, some memories is still hold for t…

  Show description for e0fca38
  

  lingvo-bot

  authored and

  copybara-github

  committed
  e0fca38

  Copy full SHA for e0fca38

• Add cyclical_step flag to TransformerSchedule.

  Show description for 5e4b7fd
  

  ds-hwang

  authored and

  copybara-github

  committed
  5e4b7fd

  Copy full SHA for 5e4b7fd

Commits on Nov 15, 2022

• Making get_zero_batch in gshard_decode a @classmethod

  Show description for 2aef09d
  

  lingvo-bot

  authored and

  copybara-github

  committed
  2aef09d

  Copy full SHA for 2aef09d

• Improve DefineAndTrace type signature; lingvo 0.12.5 release

  Show description for f296b81
  

  lingvo-bot

  authored and

  copybara-github

  committed
  f296b81

  Copy full SHA for f296b81

• Save the current EMA varaibles to the checkpoint, instead of t-1.

  Show description for 9150f56
  

  ds-hwang

  authored and

  copybara-github

  committed
  9150f56

  Copy full SHA for 9150f56

Commits on Nov 11, 2022

• Prepare for update to python3.10

  Show description for 3bdda3f
  

  lingvo-bot

  authored and

  copybara-github

  committed
  3bdda3f

  Copy full SHA for 3bdda3f

• Reuse the mechanism of py_utils to have empty name for EMA variable_scope.

  Show description for 0ca2767
  

  ds-hwang

  authored and

  copybara-github

  committed
  0ca2767

  Copy full SHA for 0ca2767

• Now that we require Py3.8, use the walrus operator in a few places.

  Show description for 2c75e99
  

  lingvo-bot

  authored and

  copybara-github

  committed
  2c75e99

  Copy full SHA for 2c75e99

• Remove debugging lines

  Show description for 6569919
  

  lingvo-bot

  authored and

  copybara-github

  committed
  6569919

  Copy full SHA for 6569919

• In MultitaskAdapterEinsumLayer, cast tasks tensor to tf.int32 as it is required to be this type.

  Show description for 980c1b0
  

  lingvo-bot

  authored and

  copybara-github

  committed
  980c1b0

  Copy full SHA for 980c1b0

Commits on Nov 10, 2022

• An adhoc fix that adds variable `last_alpha_update_step` (from compression_op.py) to TF2 model checkpoints.

  Show description for 747e91a
  

  lingvo-bot

  authored and

  copybara-github

  committed
  747e91a

  Copy full SHA for 747e91a

Commits on Nov 9, 2022

• Fix testing target: layer.theta -> layer.vars.

  Show description for 94006a9
  

  aaroey

  authored and

  copybara-github

  committed
  94006a9

  Copy full SHA for 94006a9

Pagination

Previous

Next