Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Jan 28, 2022

• Remove non-existing source_ids key.

  Show description for 7f0dd76
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  7f0dd76

  Copy full SHA for 7f0dd76

• GLam with bp16.

  Show description for 9b1de8d
  

  bignamehyp

  authored and

  copybara-github

  committed
  9b1de8d

  Copy full SHA for 9b1de8d

• Update punctuator codelab to use tf.data DataSource based input generator.

  Show description for 55caa2a
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  55caa2a

  Copy full SHA for 55caa2a

• Instead of getting users to run a tree_map over gdas, etc and run asyncio.run, absorb those APIs into the gda serde library.

  Show description for 8cb6b6e
  

  lingvo-bot

  authored and

  copybara-github

  committed
  8cb6b6e

  Copy full SHA for 8cb6b6e

• Internal Change

  Show description for 0142fa3
  

  lingvo-bot

  authored and

  copybara-github

  committed
  0142fa3

  Copy full SHA for 0142fa3

• Add a de-duplicating AverageMetric.

  Show description for 273c318
  

  Vijay Vasudevan

  authored and

  copybara-github

  committed
  273c318

  Copy full SHA for 273c318

• Move the sampling output postprocessing logic to a helper function in BaseBeamSearchDecoder to allow reusing between different decoders.

  Show description for 614a703
  

  lingvo-bot

  authored and

  copybara-github

  committed
  614a703

  Copy full SHA for 614a703

Commits on Jan 27, 2022

• By default, only use train program to create executor checkpointer, only use eval/decode programs when train program not specified, e.g. train_executions_per_eval=0.

  Show description for 64d481e
  

  lingvo-bot

  authored and

  copybara-github

  committed
  64d481e

  Copy full SHA for 64d481e

• Internal change.

  Show description for 058a2ab
  

  lingvo-bot

  authored and

  copybara-github

  committed
  058a2ab

  Copy full SHA for 058a2ab

• Skip a test entirely in learner (instead of partly in optimizer).

  Show description for 1d7e9fb
  

  ukoxyz

  authored and

  copybara-github

  committed
  1d7e9fb

  Copy full SHA for 1d7e9fb

Commits on Jan 26, 2022

• Support callable learning rate for more optimizers.

  Show description for 137bb62
  

  lingvo-bot

  authored and

  copybara-github

  committed
  137bb62

  Copy full SHA for 137bb62

• Add Copy() when accessing program_schedule_params in GetExecutorParams(), in case users may share params instance in configuration.

  Show description for a22da88
  

  lingvo-bot

  authored and

  copybara-github

  committed
  a22da88

  Copy full SHA for a22da88

• Some internal changes

  Show description for e02c941
  

  bignamehyp

  authored and

  copybara-github

  committed
  e02c941

  Copy full SHA for e02c941

• Fix checkpoints.py when using GDA

  Show description for d340d3d
  

  laurentes

  authored and

  copybara-github

  committed
  d340d3d

  Copy full SHA for d340d3d

• Allow dataset specific task override for multitask model.

  Show description for c62c6c9
  

  lingvo-bot

  authored and

  copybara-github

  committed
  c62c6c9

  Copy full SHA for c62c6c9

Commits on Jan 25, 2022

• Make sure that spmd decode supports GDA.

  Show description for 4cc679b
  

  lingvo-bot

  authored and

  copybara-github

  committed
  4cc679b

  Copy full SHA for 4cc679b

• Fix TransformerLmSpmdAdafactor param, otherwise it hits assertion in TransformerLm.

  Show description for c52b0f5
  

  lingvo-bot

  authored and

  copybara-github

  committed
  c52b0f5

  Copy full SHA for c52b0f5

• add local attention support for conformer

  Show description for 93b23b0
  

  lingvo-bot

  authored and

  copybara-github

  committed
  93b23b0

  Copy full SHA for 93b23b0

• Cleanup addressing post-submit comments.

  Show description for 1705c24
  

  zhangqiaorjc

  authored and

  copybara-github

  committed
  1705c24

  Copy full SHA for 1705c24

• Add gradient tests against TF's CTC implementation.

  Show description for 8cd537d
  

  yotarok

  authored and

  copybara-github

  committed
  8cd537d

  Copy full SHA for 8cd537d

• Implement ChunkwiseAttention/ChunkwiseAttentionXL in batch_major_attention

  Show description for b3fe586
  

  yqwangustc

  authored and

  copybara-github

  committed
  b3fe586

  Copy full SHA for b3fe586

• add SpecAugment implementation

  Show description for 3543ff3
  

  lingvo-bot

  authored and

  copybara-github

  committed
  3543ff3

  Copy full SHA for 3543ff3

Commits on Jan 24, 2022

• Fix a bug where we wrongly assume dim_per_head must be model_dim // num_heads in extend_step().

  Show description for 5422803
  

  lingvo-bot

  authored and

  copybara-github

  committed
  5422803

  Copy full SHA for 5422803

Commits on Jan 23, 2022

• Add training metrics that allows optimizers to summarize important metrics.

  Show description for 34b21ee
  

  rohan-anil

  authored and

  copybara-github

  committed
  34b21ee

  Copy full SHA for 34b21ee

Commits on Jan 22, 2022

• Cleans up ResNet50 on ImageNet.

  Show description for 87414c0
  

  lingvo-bot

  authored and

  copybara-github

  committed
  87414c0

  Copy full SHA for 87414c0

• For decoding spmd models: Each jax process runs process_decode_out() on its own (manually created from a fully replicated output) shard.

  Show description for 7277dc1
  

  lingvo-bot

  authored and

  copybara-github

  committed
  7277dc1

  Copy full SHA for 7277dc1

• Add padding support for ConvBNAct

  Show description for c27fae4
  

  lingvo-bot

  authored and

  copybara-github

  committed
  c27fae4

  Copy full SHA for c27fae4

• add configure files for the XEndec approach and remove useless codes in its implementation.

  Show description for daaefe0
  

  lingvo-bot

  authored and

  copybara-github

  committed
  daaefe0

  Copy full SHA for daaefe0

• Taking the first step towards integrating flax.core to lingvo-jax by adding flax.core.scope to JaxContext and removing _forward_updated_vars from BaseLayer to handle side-effects in layer fprop cle…

  Show description for 9b21acd
  

  zhangqiaorjc

  authored and

  copybara-github

  committed
  9b21acd

  Copy full SHA for 9b21acd

• Automated rollback of commit a8ead2348f2583f37f7872c6d9518b439111fb6f

  Show description for 4c140b0
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  4c140b0

  Copy full SHA for 4c140b0

Commits on Jan 21, 2022

• Add TF2 compatibility of adafactor and ggt optimizers

  Show description for 86d7166
  

  lingvo-bot

  authored and

  copybara-github

  committed
  86d7166

  Copy full SHA for 86d7166

• Previously, each executor program has its own Checkpointer for restoring. The

  Show description for 8dd8cb6
  

  Jonathan Shen

  authored and

  copybara-github

  committed
  8dd8cb6

  Copy full SHA for 8dd8cb6

• Add collapse_and_remove_blanks, used to remove blanks and repeats in a typical CTC output stream.

  Show description for ad9a92a
  

  lingvo-bot

  authored and

  copybara-github

  committed
  ad9a92a

  Copy full SHA for ad9a92a

• Add log probs on the selected tokens (including prefix) to decode outputs.

  Show description for 52c861f
  

  lingvo-bot

  authored and

  copybara-github

  committed
  52c861f

  Copy full SHA for 52c861f

• Add StackedConformer layer which stacks N layers Conformers.

  Show description for 94a2074
  

  lingvo-bot

  authored and

  copybara-github

  committed
  94a2074

  Copy full SHA for 94a2074

Pagination

Previous

Next