Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Aug 4, 2020

• Add additional encoder inference graph support for beam search decoders.

  Show description for 3e42a1c
  

  mrhuke

  authored and

  copybara-github

  committed
  3e42a1c

  Copy full SHA for 3e42a1c

• Implements the routing attention based on k-means clustering.

  Show description for 11a216c
  

  lingvo-bot

  authored and

  copybara-github

  committed
  11a216c

  Copy full SHA for 11a216c

• Add generic_input op unit test for usage with tf.data

  Show description for 68dfe3a
  

  lingvo-bot

  authored and

  copybara-github

  committed
  68dfe3a

  Copy full SHA for 68dfe3a

Commits on Aug 3, 2020

• Closes #231: Disable eager execution in create_asr_features.

  Show description for 7d17151
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  7d17151

  Copy full SHA for 7d17151

• Move DistillationTask to its own file.

  Show description for 2393c80
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  2393c80

  Copy full SHA for 2393c80

• base_model.py refactor.

  Show description for 8ff6a9f
  

  lingvo-bot

  authored and

  copybara-github

  committed
  8ff6a9f

  Copy full SHA for 8ff6a9f

• Bug fix for ordering of PostTrainingStepUpdate.

  Show description for 069fbd5
  

  Vijay Vasudevan

  authored and

  copybara-github

  committed
  069fbd5

  Copy full SHA for 069fbd5

• Add residual_weight to layers_with_attention.TransformerFeedForwardLayer

  Show description for 1580ad3
  

  protoget

  authored and

  copybara-github

  committed
  1580ad3

  Copy full SHA for 1580ad3

Commits on Aug 1, 2020

• Increment total_records_skipped_ for CANCELLED errors

  Show description for 2b5a5d6
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  2b5a5d6

  Copy full SHA for 2b5a5d6

• Better error handling for imports.

  Show description for a79489d
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  a79489d

  Copy full SHA for a79489d

• in moe_layers.py allow state layer updates

  Show description for b4fd694
  

  lingvo-bot

  authored and

  copybara-github

  committed
  b4fd694

  Copy full SHA for b4fd694

• Don't copy the tensor content if the batch size is 1.

  Show description for b368092
  

  aaroey

  authored and

  copybara-github

  committed
  b368092

  Copy full SHA for b368092

• Add autoregressive decoding support for LocalSelfAttention.

  Show description for b4e6e49
  

  lingvo-bot

  authored and

  copybara-github

  committed
  b4e6e49

  Copy full SHA for b4e6e49

• Add per_core_infeed option to allow handling of per host infeed with data parallelism while num_partitions is > 1.

  Show description for 391780f
  

  lingvo-bot

  authored and

  copybara-github

  committed
  391780f

  Copy full SHA for 391780f

Commits on Jul 31, 2020

• Support streaming ground truth bounding boxes to reduce memory consumption for GroundTruthAugmentor preprocessor.

  Show description for e07e8d9
  

  lingvo-bot

  authored and

  copybara-github

  committed
  e07e8d9

  Copy full SHA for e07e8d9

• Add a test script to verify docstring formatting manually.

  Show description for a0ddb80
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  a0ddb80

  Copy full SHA for a0ddb80

• Add a flag for softmax scaling code motion and set it to false by default.

  Show description for e876b48
  

  lingvo-bot

  authored and

  copybara-github

  committed
  e876b48

  Copy full SHA for e876b48

• Set the output shapes in _DefineFunction.

  Show description for fb1f3f0
  

  aaroey

  authored and

  copybara-github

  committed
  fb1f3f0

  Copy full SHA for fb1f3f0

Commits on Jul 30, 2020

• Enable multiple learner with single loss.

  Show description for dd55b4d
  

  yzhang87

  authored and

  copybara-github

  committed
  dd55b4d

  Copy full SHA for dd55b4d

• Remove context managers such as tf.variable_scope inside of __init__.

  Show description for 43e4866
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  43e4866

  Copy full SHA for 43e4866

• Allow fprop_dtype propagation to ShardedVarLayer.

  Show description for 87f7963
  

  lingvo-bot

  authored and

  copybara-github

  committed
  87f7963

  Copy full SHA for 87f7963

• Fix string.

  Show description for d6a0318
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  d6a0318

  Copy full SHA for d6a0318

• Add v2_padding option to conv_layers_builder.py

  Show description for 0f13467
  

  lingvo-bot

  authored and

  copybara-github

  committed
  0f13467

  Copy full SHA for 0f13467

• Allow IdentityLayer inputs to be NestedMap.

  Show description for cf0cb4d
  

  lingvo-bot

  authored and

  copybara-github

  committed
  cf0cb4d

  Copy full SHA for cf0cb4d

• Fix Decode programs to be in cluster_factory.SetEval(True) context.

  Show description for c723825
  

  freeblee

  authored and

  copybara-github

  committed
  c723825

  Copy full SHA for c723825

• Add a context manager for rendezvous sharing in tf.function, and use it in

  Show description for 2a85330
  

  aaroey

  authored and

  copybara-github

  committed
  2a85330

  Copy full SHA for 2a85330

• Remove AddTrainerVerboseTensor.

  Show description for 5d255ea
  

  lingvo-bot

  authored and

  copybara-github

  committed
  5d255ea

  Copy full SHA for 5d255ea

• Rename relative_atten_util to attention_util

  Show description for 0dc242d
  

  lingvo-bot

  authored and

  copybara-github

  committed
  0dc242d

  Copy full SHA for 0dc242d

• Computes SmoothL1 loss on the coordinates of each corner.

  Show description for a56d76e
  

  lingvo-bot

  authored and

  copybara-github

  committed
  a56d76e

  Copy full SHA for a56d76e

• Stop using Defun for some simple test cases, by removing it or changing it to py_utils.CallDefun or tf.function.

  Show description for da04588
  

  aaroey

  authored and

  copybara-github

  committed
  da04588

  Copy full SHA for da04588

• Implement the k-means clustering of the routing transformer.

  Show description for 3e9cc1c
  

  lingvo-bot

  authored and

  copybara-github

  committed
  3e9cc1c

  Copy full SHA for 3e9cc1c

• Add helper methods to avoid redundant test code.

  Show description for 43f8099
  

  aaroey

  authored and

  copybara-github

  committed
  43f8099

  Copy full SHA for 43f8099

Commits on Jul 29, 2020

• Revert shape check op to internal op.

  Show description for 8c5373a
  

  jngiam

  authored and

  copybara-github

  committed
  8c5373a

  Copy full SHA for 8c5373a

• Add assertion failure details in specaug.

  Show description for 127243c
  

  anmolgulati

  authored and

  copybara-github

  committed
  127243c

  Copy full SHA for 127243c

• Enable tf.function in recurrent_test.

  Show description for 14f2fe5
  

  aaroey

  authored and

  copybara-github

  committed
  14f2fe5

  Copy full SHA for 14f2fe5

Pagination

Previous

Next